Podcasts about Inference

My guest today is Ben Thompson, the founder and author of Stratechery. Ben is one of my favorite business thinkers and I love talking to him about everything happening in markets and technology.  We go through every important company, including OpenAI, Nvidia, Intel, Apple, Microsoft, Google, and Amazon. We also discuss why he thinks it would be dangerous for the United States to win the AI race outright, what container shipping and the railroads of the 1870s tell us about the buildout, and why the binding constraint on all of this may be capital rather than compute.  Please enjoy my conversation with Ben Thompson. For the full show notes, transcript, and links to mentioned content, check out the episode page ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠here⁠⁠⁠⁠⁠.  ----- Become a Colossus member to get our quarterly print magazine and private audio experience, including exclusive profiles and early access to select episodes. Subscribe at ⁠colossus.com/subscribe⁠. ----- ⁠Ramp's⁠ mission is to help companies manage their spend in a way that reduces expenses and frees up time for teams to work on more valuable projects. Go to⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ ⁠ramp.com/invest⁠⁠ to sign up for free and get a $250 welcome bonus. ----- Trusted by thousands of businesses, ⁠Vanta⁠ continuously monitors your security posture and streamlines audits so you can win enterprise deals and build customer trust without the traditional overhead. Invest Like the Best listeners get a special offer of $1,000 off Vanta when you go to ⁠vanta.com/invest⁠.  ----- WorkOS⁠ is the infrastructure B2B and AI-native companies use to sell to enterprise. It covers everything enterprise security requires: SSO, SCIM, RBAC, Audit Logs, AI governance, and more. Trusted by 2,000+ fast-growing companies, including OpenAI, Anthropic, Cursor, and Vercel. ----- Rogo is the AI platform for finance. They're building agents for Wall Street that are trained to understand how bankers and investors actually do work: from diligence and modeling, to turning analysis into deliverables. To learn more, visit rogo.ai/invest. ----- ⁠Ridgeline⁠ has built a complete, real-time, modern operating system for investment managers. It handles trading, portfolio management, compliance, customer reporting, and much more through an all-in-one real-time cloud platform. Visit⁠ ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ridgeline.ai⁠. ----- Editing and post-production work for this episode was provided by The Podcast Consultant. Timestamps: (00:00:00) Welcome to Invest Like The Best (00:02:16) Winning the AI Race With China (00:08:28) Timing, Capital, and the Railroads (00:11:34) Berkshire, Google, and Absolute Profits (00:14:23) Verifiable and Unverifiable Domains (00:20:20) Aggregation Theory in the AI Era (00:22:06) The Real Cost of Inference (00:25:40) Why Consumer AI Needs Advertising (00:30:08) Compute Shortages and Commodity Markets (00:35:46) Memory Cycles and Boom Bust Dynamics (00:42:08) TSMC, Intel, and Where Risk Goes (00:44:51) The Best Setups in Big Tech (00:52:14) The Frontier Model Contenders (00:54:27) Microsoft's IBM Playbook (01:00:45) Meta, Attention, and Advertising (01:07:29) NVIDIA, Commodities, and Power

Welcome to episode 367 of The Cloud Pod, where the forecast is always cloudy! Justin, Ryan, and Matthew are in the studio this week and ready with a lot of news, including passkeys (we know, they've had a rough week), Secrets Manager, Vector Search, and Glimmer (no, not my second favorite character from She-Ra), and even…wait for it…undersea cable news!  We've got a lot to cover, so let's get started!  Titles we almost went with this week AWS Secrets Manager Jenkins Rotation Finally Claude Enterprise Hooks a Ride on Data Loss Prevention Passkeys Take the Wheel, SMS Rides Off Into the Sunset Claude Code Says Trust Falls Are Over Muse Glimmer Shines While Meta’s Wallet Dims Zuckerberg Bets Big on Open Weights, Loses on Free Cash Flow AI is persistently in the news How many ways are there to run vector search in AWS, now 1 more Vector Search is the new Docker on AWS… how many ways are there to run it AWS Says “You get a Vector Search, and you get a Vector Search” You say you’re a Cloud Azure, but “Azure Network Router Appliance” says otherwise Claude now tells the world, I did the AI Slop Open, Closed, Open; Zuckerberg is on the AI Revolving Door Anthropic triples everyone’s productivity with Automode A big thanks to this week's sponsors: We're sponsorless! Want to get your brand, company, or service in front of a very enthusiastic group of cloud news seekers? You've come to the right place! Send us an email or hit us up on our Slack channel for more info. AI Is Going Great – or How ML Makes Money  01:40 Inference hooks: inline data loss prevention for Claude Enterprise  Anthropic launched inference hooks in beta for Claude Enterprise, providing inline data loss prevention across chat, Claude Code, Claude Cowork, and other Enterprise surfaces through a single configuration point. Technical approach: every inference request routes through a signed WebSocket connection to a customer-controlled security server; Claude sends the prompt and context before generation begins and waits for an allow/deny verdict before proceeding. The same inspection applies to tool call responses, including those from MCP connectors, skills, and plugins. The feature uses an open, webhook-based protocol with a published schema, allowing integration with existing DLP vendors such as Netskope, Palo Alto Networks, Proofpoint, and Zscaler, or custom in-house security servers, without requiring separate per-product integration work. Rollout controls include shadow mode (log without blocking), role-based exclusions, and percentage-based rollouts, along with configurable failure-policy tolerance and timeouts to match organizational risk requirements. This addresses a gap where inline enforcement was previously limited to Claude Code’s client-side hooks, giving compliance teams a unified enforcement layer for sensitive data across all Claude Enterprise channels.  Documentation is available

AI isn't just transforming the way we work, but also the way we write the software that people use for work. In this episode, we talk to two engineering productivity leads at Dropbox: Uma Namasivayam, senior director of software engineering productivity, and Anuradha Agarwal, director of software engineering. Whether it's writing tests, fixing bugs, tackling tech debt, or accelerating migrations, they explain how Dropbox engineers are using agentic AI—including in-house tools like Nova—to build the future of Dropbox, and create more space to do impactful work. ~ ~ ~  Working Smarter is brought to you by Dropbox. Find, organize, and share your work—all in one place—with context-aware AI from Dropbox. You can listen to more episodes of Working Smarter on Apple Podcasts, Spotify, YouTube, Amazon Music, or wherever you get your podcasts. To read more stories and past interviews, visit workingsmarter.ai This show would not be possible without the talented team at Cosmic Standard: producer Ben Montoya, sound engineer Aja Simpson, technical director Jacob Winik, and executive producer Eliza Smith. Special thanks to our illustrator Fanny Luor, marketing consultant Meggan Ellingboe, and editorial support from Catie Keck.  Our theme song was composed by Doug Stuart.  Working Smarter is hosted by Matthew Braga. Thanks for listening!

My conversation with Tarun Chitra.As a co-founder of Gauntlet and GP at Robot Ventures, Tarun has one of the sharpest frameworks for understanding market structure across both crypto and AI. In this episode we dig into why open-source AI is unbundling faster than most people expect, and how the resulting stack looks surprisingly similar to DeFi.We spend a lot of time mapping the AI infrastructure layers directly onto crypto primitives and examining where value is actually going to accrue as models, harnesses, routers, and inference providers separate.At the center of the conversation is the belief that AI's unbundling is creating a new competitive order-flow market (data centers competing like nodes, MEV-like dynamics for tokens/GPUs) while crypto itself has settled into a more mature “TradFi plus+” phase focused on trading, payments, and bringing real assets on-chain.We discuss:- The current state of crypto as TradFi+ and the decline of speculative narratives- Why AI is killing Bitcoin mining economics and weakening ETH value accrual- The architectural parallel between AI stacks and DeFi Harnesses = WalletsRouters = DEX aggregators,Models = ProtocolsInference Providers = LPs- Why open-source models are unbundling faster than traditional software- Agents as the next interface layer and the potential unbundling of ETFs- Cryptography, verifiable compute, and turning GPUs into digital assets- Onchain compute trading as the real crypto × AI opportunity- Sustainable business models and where value will ultimately captureTimestamps:0:00 – Introduction & State of the Crypto Market (TradFi+)2:00 – Speculative Narratives Fade, Trading & Payments Remain7:00 – AI's Impact on Bitcoin Economics & Data Center Opportunity Cost9:00 – ETH Value Accrual, Solana Positioning & DeFi Token Sustainability15:00 – Trading Design Space & On-Chain Volume Upside25:00 – AI Unbundling Thesis: Open-Source Models vs Data Centers35:00 – The DeFi Mapping (Harnesses, Routers, Models, Inference)48:00 – Agents, Preference Expression & Unbundling Traditional Products1:00:00 – Real-Time Harness Generation & Active Learning1:05:00 – Cryptography, Verifiable Compute & On-Chain GPU Markets1:10:00 – Closing Thoughts: Where Value Accrues NextEnjoy!

Send us Fan MailRob May, Co-Founder & CEO of NeuroMetric AI, who's revolutionizing how multi-model systems think and run, dramatically reducing expenses while boosting performance. Rob brings decades of experience as a founder, investor, and the creator of the popular ‘Investing in AI' newsletter.01:05 Meeting Rob May04:22 The Investing in AI Newsletter06:50 Starting Neurometric11:39 Inferencing Defined16:12 Future Investing in AI20:24 Model Thinking w/ Guidance23:40 Neurometric, a Token Engineering Platform29:07 Data Center Investments30:42 Real AI ROI33:48 Where is the Smart Money Going?35:37 What AI Leaders Must Get Right39:09 Stock Picks40:32 Rapid FireWant to be featured as a guest on Making Data Simple?  Reach out to us at almartintalksdata@gmail.com and tell us why you should be next.  The Making Data Simple Podcast is hosted by Al Martin, WW VP Technical Sales, IBM, where we explore trending technologies, business innovation, and leadership ... while keeping it simple & fun. 

Send us Fan MailRob May, Co-Founder & CEO of NeuroMetric AI, who's revolutionizing how multi-model systems think and run, dramatically reducing expenses while boosting performance. Rob brings decades of experience as a founder, investor, and the creator of the popular ‘Investing in AI' newsletter.01:05 Meeting Rob May04:22 The Investing in AI Newsletter06:50 Starting Neurometric11:39 Inferencing Defined16:12 Future Investing in AI20:24 Model Thinking w/ Guidance23:40 Neurometric, a Token Engineering Platform29:07 Data Center Investments30:42 Real AI ROI33:48 Where is the Smart Money Going?35:37 What AI Leaders Must Get Right39:09 Stock Picks40:32 Rapid FireWant to be featured as a guest on Making Data Simple?  Reach out to us at almartintalksdata@gmail.com and tell us why you should be next.  The Making Data Simple Podcast is hosted by Al Martin, WW VP Technical Sales, IBM, where we explore trending technologies, business innovation, and leadership ... while keeping it simple & fun. 

In this episode, we explore how artificial intelligence and machine learning are advancing the use of electronic health record data in rheumatology. Dr. Kat discusses the development and validation of a natural language processing–based model to infer rheumatoid arthritis disease activity from structured and unstructured clinical data, its application in studying cardiovascular outcomes, and the challenges of deploying predictive models across different health systems. The conversation also examines the future of real-world evidence generation, the practical limitations of AI in clinical research, and Dr. Kat's career journey, offering insights and advice for the next generation of physician-scientists. 

Watch the full episode on YouTube:We first covered Baseten last year when DeepSeek mania was at peak hype. Now they have raised a monster $13B round and become one of the new cohort of AI Infra decacorns that are (with Nvidia, Intel, and the semis complex) chief beneficiaries of the Inference Inflection. We return to Baseten at the peak of the 2026 edition of Open Weights debate. Ali has published a viral breakdown of Kimi K3:And since you last saw him, Philip has spoken at AI Engineer and written the definitive book on Inference Engineering spotted all over SF:Three years ago, inference engineering barely existed as a category.Today, it is one of the most critical disciplines in AI. Inference engineering inherently tackles a different question than standard model training: “How do you turn those weights from training into a product that is fast, reliable, and affordable at scale?” Focusing on these creates an entirely new optimization problem.In one recent GLM-5.2 experiment, quantizing more of the model actually preserved its benchmark quality while increasing throughput by 20%, because the errors introduced in different layers could cancel each other out.Inference is no longer just the final step after training. It is becoming its own engineering discipline, with its own research problems, infrastructure, and increasingly specialized roles.In this episode, Baseten's Philip Kiely and Ali Taha join swyx and Vibhu to explain what actually happens after a new open model is released and what it takes to turn “we generated a token” into a fast, reliable, production-ready API.We go deep on cache-aware routing, disaggregated prefill and decode, quantization, speculative decoding, KV-cache movement, model parallelism, GPU kernels, and the race to make frontier models up to 10× faster. Philip and Ali explain why inference optimizations can still produce gains of 20%, 100%, or even 200%; how quantization errors can cancel one another out; why identical weights can behave differently across clusters; and how Baseten grafted a Kimi vision encoder onto GLM-5.2 without changing the underlying language model.The conversation then expands beyond LLMs into NVIDIA Dynamo, mega kernels, Rubin, AI-specific chips, local inference, video generation, diffusion versus autoregressive models, and the enormous compute barrier to generating coherent long-form video. Finally, we explore the convergence of training and inference, continual learning through persistent KV cache, and the emerging loop where models help optimize the infrastructure that runs them.We discuss:* What happens when a 200,000-token request enters an inference system* Cache-aware routing and reusing previously computed KV cache* Why prefill and decode are increasingly handled by different GPUs* When dedicated deployments become cheaper and more reliable than shared APIs* How speculative decoding uses a smaller model to accelerate a larger one* Tool calling, structured outputs, and what LLMs actually do* What it takes to support a new open model on day zero* Grafting Kimi's vision encoder onto GLM-5.2* Retrofitting inefficient model layers with components from other architectures* Why models sometimes collapse into repeating the same token* How hardware, kernels, and race conditions create nondeterministic failures* Preserving model fidelity while making inference faster* How quantization errors can cancel each other out* Why inference optimizations still deliver gains of 20%, 100%, and 200%* How optimized serving can make a model up to 10× faster* NVIDIA Dynamo, KV-aware routing, and distributed model serving* Speculative decoding the speculative decoder* Why local AI is about making models less dumb while data-center AI is about making them less slow* Tensor, expert, and pipeline parallelism across GPUs* Hardware-aware model design, auto-tuning, and the case against mega kernels* Rubin and why inference is becoming a systems problem* Whether modern GPUs are evolving into programmable AI ASICs* Why enormous models like Kimi K3 require GB300-class hardware* Why open-source video generation still trails Veo, Kling, and other closed models* The quadratic attention bottleneck behind long-form AI video* Autoregressive video, real-time generation, and compounding quality drift* Why future video systems may combine autoregressive and diffusion architectures* Training for inference and inference for training* Continuous post-training, deployment, evaluation, and improvement loops* How GLM-5.2 helped optimize the kernels serving GLM-5.2 itself* Why faster networking could unlock dramatically faster decoding* Continual learning, KV-cache compaction, and persistent model memoryShow Notes* How to build a day-0 API for Kimi K3* 22580: From GPT2 to Kimi3, ExplainedPhilip Kiely* LinkedIn: https://www.linkedin.com/in/philipkiely* X: https://x.com/philipkiely* Inference Engineering: https://www.baseten.co/inference-engineering/Ali Taha* LinkedIn: https://www.linkedin.com/in/aliestaha/* X: https://x.com/waterloointernTimestamps00:00:00 Introduction and the 200K-Token Prompt00:03:18 Dedicated Deployments, Speculative Decoding, and Tool Calling00:11:26 Launching Production-Ready Open Models00:19:06 Model Retrofits, Failure Modes, and Nondeterminism00:28:22 Quantization and Canceling Errors00:32:15 The Race to 10× Faster Inference00:40:48 Dynamo, Speculation, and Local vs. Data-Center AI00:50:18 Model Parallelism, Auto-Tuning, and Mega Kernels01:00:55 Rubin, GPUs vs. ASICs, and Custom AI Chips01:10:03 Giant Models and the Limits of GPU Memory01:12:42 AI Video, Quadratic Attention, and Autoregressive Generation01:21:47 Audio, Images, and Diffusion Models01:27:32 Training, Self-Optimizing Models, and Continual Learning01:40:06 Closing ThoughtsTranscriptIntroduction: Baseten, Waterloo Intern, and Inference EngineeringSwyx [00:00:00]: Okay, we're here in the studio with Philip, old friend from Inference Engineering, the book, as well as Baseten and everything that you've done, you and I have done before, as well as Ali. Welcome.Ali [00:00:15]: Pleasure to meet you.Swyx [00:00:15]: Waterloo intern.Ali [00:00:16]: Waterloo intern, always.Swyx [00:00:17]: When did you get “Waterloo intern” as a handle?Ali [00:00:19]: As a handle? Oh.Ali [00:00:20]: I think the rebranding happened mid-March. When I saw it was open, I was like, “I have to take it. Up for grabs.”Philip [00:00:26]: The problem is that Ali is really good at his job and is not gonna be an intern much longer.Philip [00:00:30]: So we have to figure out who's gonna get the handle.Ali [00:00:33]: Well, I'll pass the torch over to the next intern.Swyx [00:00:34]: Oh, okay. It can be, like, you just pass it to another Waterloo grad.Ali [00:00:37]: To another Waterloo intern. No, bruh.Philip [00:00:39]: Yeah.Ali [00:00:39]: Intern.Swyx [00:00:40]: Intern, yeah.Ali [00:00:40]: And no.Philip [00:00:41]: You gotta get an intern from Waterloo.Ali [00:00:42]: Yeah, I've gotta get an intern from Waterloo.Swyx [00:00:44]: Right.Ali [00:00:44]: But they have to follow the path.Swyx [00:00:45]: Oh, it could, but it could come from Baseten, so it's like whoever Baseten gets from Waterloo.Ali [00:00:48]: Right.Swyx [00:00:49]: Has the title of Waterloo.Ali [00:00:50]: It stays in the ecosystem.Philip [00:00:51]: Exactly.Ali [00:00:52]: Halfway through the internship, you either get it or you're out.Philip [00:00:55]: You should also do, like, a big graduation ceremony where you change the handle.Ali [00:00:59]: Just say it.Philip [00:00:59]: For everybody.Swyx [00:01:00]: You guys are good at ceremonies, clearly. We had a nice launch of the book, very successful. But before we get into all that, I wanna start off with a fun question for you. Okay, you're an expert inference engineer. What happens when I send a long query, say two hundred thousand tokens into Baseten's inference? What's the process of query through GPU model routing, balancing, all that? What is all the stuff that we don't think about?Long Context Requests, KV Cache, and Cache-Aware RoutingPhilip [00:01:26]: With a long query specifically, the first thing that I'm gonna ask is, “Have you sent me this query before, or at least part of it?” and I really hope you have, because it's gonna be a lot easier for me and a lot cheaper for you. So the first thing that we're gonna look at is some cache-aware routing, where we're going to see, we probably have a number of instances, a number of replicas up serving whatever model you're hitting. We want to send this one to something with, number one, available prefill workers, and number two, ideally some cached input already there so that we can skip prefill on at least part of these two hundred thousand tokens. If you're doing two hundred thousand tokens, it's probably coding or a multi-turn agent or something where you would expect to have that cached. If you don't, we're gonna have to send it to a prefill worker. We've at least on certain models disaggregated prefill and decode, so you're going to have one set of GPUs that's solely going to process the input, create the KV cache, and get you your first token, and then that's going to be passed over to a separate set of GPUs, which is going to run decode. We're going to iteratively make those tokens. We're probably going to have some speculator model in front of that. I'm going to assume that you're doing coding, and because of that, our speculator model, which assumes you're doing coding, is gonna have a high draft token acceptance rate. If I'm wrong and you're asking me to summarize every Harry Potter book, it's gonna be slower. And then we stream that output to you and account for it, charge you, a couple of pennies and say, “Hey, would you like to send another one?”Swyx [00:03:04]: Except Baseten doesn't charge by pennies.Philip [00:03:07]: Well, yeah, we charge. I'm assuming that we're talking about the public model APIs. If you are setting up a dedicated deployment, then yeah, it's not pennies.Public APIs vs. Dedicated DeploymentsSwyx [00:03:18]: Yeah, one of the key differentiators when I was talking with Baseten initially was that people who want very high volume just need to rent by the box, ‘cause then it's up to you to figure out how to saturate the box.Ali [00:03:31]: And more often than not, it's, like, way cheaper if you're pushing, like, millions of tokens per hour, if you just pay per hour instead of pay per token.Philip [00:03:37]: Yeah, they do. I think that we've increasingly seen a lot of demand for the pay per token APIs, just because everyone wants to try open models, and then once they find a use case that's really sticky, then they move over to dedicated.Swyx [00:03:51]: Is there a best practice on when it's time to swap over?Philip [00:03:54]: Couple reasons. Yeah, reliability, that's a big one, right?Ali [00:03:57]: Like, if they have a very specific use case, they want you to train something specifically for them, like they want their own spec dec, for instance, for their own traffic.Swyx [00:04:04]: Spec dec is speculative decoding.Speculative Decoding and Custom SpeculatorsAli [00:04:05]: Speculative decoding, yeah.Swyx [00:04:07]: You have to explain.Ali [00:04:07]: Sorry. Like, speculative decoding is like, if you have a huge model, right? And so the model is going to be generating one token at a time every single turn, every single forward pass. So we attach, like, this little, like, parasite, like this layer that goes on top of the model, and this model just has to predict. It does three very fast autoregressive forward passes, and it will predict, like, three certain tokens, and then you do one forward stage over the entire original model in order to see if those predictions were correct or not, and then you accept them or you reject them. Now, this draft model is traffic specific, so if you, like, Philip said, if you're summarizing Harry Potter books, I can train exclusively that draft model on Harry Potter books, and I can guarantee you that I'm gonna accept the three tokens every single time. And so with that case, I increase your decode speed. I wouldn't be able to provide this to you if you're a shared endpointSwyx [00:04:53]: YeahAli [00:04:53]: ‘cause I have no idea if you're doing Harry Potter, if you're doing coding, if you're doing English. We don't know. Also, there was a thing in the book that mentioned that if they really cared about a specific threshold, chapter four, I think. Do you remember that?Philip [00:05:06]: Yeah. The things that you can do is you can set a specific, like, batch sizing, a specific, like, parallelism strategy if you're trying to optimize for, like, throughput versus latency. You can. Maybe a NVFP4 quant doesn't pass your benchmarks and you wanna run a model at higher precision, you could do that. There's just a bunch of reasons why you might wanna have your own endpoint and the biggest one, of course, just being, like, you don't have to deal with someone else doing a hundred million tokens of benchmarking traffic at the endpoint when you happen to be trying to serve your users.Swyx [00:05:40]: Yeah. I think one thing that is. That is a classic journey. Like, it's people is asking the, what happens when you type Google into the browser. Tool calling, is that just, you're generating JSON or is there more complication beyond that?Tool Calling, JSON, and Structured OutputsAli [00:05:58]: Certain customers that we have, they have their own post-trained models, and so they demand a tool calling that's not just, like parse a file or go find the weather. It's something that's very specific and you have to do post-training on this. And if the post-training on the model is not good or if the quantization after the post-training to get the inference to be fast, the model will struggle reading the JSON file and reading the tool calling. But it doesn't require its own like sandbox. It's not like it's going to use that tool calling to like escape a sandbox or like it doesn't have to be contained. It can just be a normal dedicated deployment. The challenge with tool calling more and more seems to be that the companies want certain tool calling which is a very sensitive thing to train. And because you're dealing with all of the JSON outputs, if it doesn't like close the end of the request in a very certain manner, you end up with a model that did the tool calling and like the thinking and so as a result of that, it didn't see the result and just hallucinated the result as it decoded. That seems to be the most challenging thing with tool calling, not really the sandboxes model.Philip [00:06:56]: Yeah, that's a challenge on the training side and then on the inference side, there's work that you can do to scope the possible output. So we published this at this point close to two years ago, the solution to this problem which is you make a state machine and you use that to constrain the output to a specific format. So this is the structured output problem. If you remember backSwyx [00:07:27]: Yeah, the specific grammar is,Philip [00:07:29]: Yeah, exactlySwyx [00:07:30]: GML had this thing.Philip [00:07:31]: Yeah. So it's like the old-school “make sure this is only JSON”, return only JSON orSwyx [00:07:38]: YeahPhilip [00:07:38]: Grandma's gonna die type of prompts.Swyx [00:07:39]: Is it BNF grammar? At some point OpenAI had released a thing that was like, yeah, if you want to constrain your output, write BNF grammar, back as NOR.Philip [00:07:47]: In our inference system, it's just a specified output format. And you get the guarantee that your output's gonna be structured along that format. And so applying that to tool calls can like help cut down on. You can still call the wrong tool or call no tool. It doesn't solve the certainty problem but it at least solves the output structuring problemSwyx [00:08:10]: YeahPhilip [00:08:10]: Within tool calls.Swyx [00:08:12]: And MCP is just another form of tool, right.Philip [00:08:14]: Yeah, exactly.Swyx [00:08:15]: As far as there's no special thing there.Philip [00:08:16]: The thing I'm always like explaining to people is the LLM is not capable of doing anything. It's only capable of making suggestions of what to do and then if those suggestions are formatted in a certain way and applied to a system that knows what to do with them, then an action occurs.Swyx [00:08:32]: Yeah. Part of the fun stuff is, this is solved outside of tool calling too. Like in an agent loop if the output is not correct or you're right, like reasoning, tool calling was done in the reasoning trace, just be like, “Oh, I don't know what to do. Let me just try again.” And it might get there after a few tries. And on your point of training, sometimes this is harder in smaller models, so you don't have the same exact quality outputAli [00:08:56]: Right.Swyx [00:08:57]: When you just swap from a big model, right?Ali [00:08:59]: Yeah. I will say that, before, I think we need to go back to inference engineering proper.Ali [00:09:04]: But, I had expected that something would replace JSON because it's hard to stream JSON ‘cause JSON must be complete and you must have open and close brackets and everything. So it's hard to parse something or validate something while it's being streamed. So people invented all sorts of things that are like, I forget the name of some of these alternatives, but it's something like TOML, something like YAML. But JSON seems to be dominant still.Philip [00:09:30]: The JSON outputs aren't that long, right? Like you could have a long-- ‘cause tool calls also contain the arguments in them and perhaps for a certain tool you might pass like a very long argument. But my impression of the median tool call is that it's a relatively small number of tokens, right? So I would expect that speculators are generally fairly good at something as formatted as JSON. And so you would have like a pretty fast decode step there and that the streaming wouldn't be as valuable, but maybe I'm wrong about that.Ali [00:10:02]: I think you're also bounded by the software or that the model is gonna integrate with if the software is built with JSON for the tool calls or if the company that you'- if your customer says that this is how our software works and our tools are interfaced with JSON, you can ask them to like, change their software and say like, “Yeah, this is gonna be better for the model.” but like with the right training shouldn't be that much of a difference. Also more profitable if it outputs more tokens probably.Swyx [00:10:25]: Depends on your business model.Swyx [00:10:27]: It really depends. But I will say that, as a writer with like experience a lot with generated output, I do try to move from text to JSON text which is very long JSON, right? Like there's paragraphs in every field because I'm trying to structure it, right?Philip [00:10:44]: Right.Swyx [00:10:44]: I want you to first make factual statements, then make opinions then make bullet point summaries, have dates, have entity references have your sources for references, all these things. Anyway, so these are things that like I think people who really experiment with structural output have to really care about. But, let's, let's recurse up the stack a little bit. Before we started recording, you mentioned something really cool, which is that there's a lot of engineering that-- inference engineering that goes on when a new model provider releases a new model, right? So let's call it GLM-5.2, Kimi K3. I had previously assumed, especially if it's like, well, GLM 5 to 5.1 to GLM-5.2, like that you've supported them before. Is it that much work?What It Takes to Support a New Open ModelAli [00:11:26]: It's a lot of work.Swyx [00:11:28]: Yeah. Okay. So like, a lot of people, all you guys, right whenever a new model launch like, people rush to say like, “Oh, Hugging Face supports this, Fireworks supports this, Spacetime supports this,” and I'm like, “Yeah, of course we support it.” But what goes into that? What goes intoPhilip [00:11:40]: I think it's more than just support it too, right? It benefits the consumer a lot. Like I think it was with Kimi K2.5 or GLM-5.2 the latest, there was an inference war, right? X provider is at 90 tokens a second. The next day we're at 150. The nextSwyx [00:11:55]: I kinda kicked that off with the GLM-5.2.Swyx [00:11:58]: I wrote a Twitter article about. It got like half a million views,Ali [00:12:02]: Based on being numberSwyx [00:12:03]: YeahAli [00:12:04]: Or it's for something else.Swyx [00:12:05]: Yeah. Which,Ali [00:12:06]: Oh my GodSwyx [00:12:07]: Which then got everyone really excited about, hey, how can we, bend tracks a little bit further and,Philip [00:12:14]: There's a difference between support the model, as in I can make a token out of this model, and support a model, as in I have a production-ready API from this model.Philip [00:12:26]: Getting to the point of I can make a token out of this model is not that hard because generally the, open source inference engines, vLLM, SGLang of the world oftentimes even receive weights ahead of time, maintainers do, or the people making the model merge PRs to ensure support. So you generally can, just get it working on the standard open source stack without too much pain in most cases. The challenge is, every inference company is gonna have own proprietary stack. Some open source components, some in-house stuff. And for any arbitrary model, there's going to be some new stuff. Sometimes you get lucky, like K, two five to two six was, like, pretty similar.Quantization, Speculators, and Production ReadinessAli [00:13:16]: Yeah. It was pure continued post-trainingPhilip [00:13:18]: YeahAli [00:13:18]: If I remember correctly.Philip [00:13:19]: Even in those cases, there's still stuff you have to do. You have to redo the quantization work. You're taking the model from. Generally, these models are not released in NVFP4, and we want them to be in NVFP4 for maximum Blackwell compatibility. So we have to perform that quantization, and, calibrate the quantization to make sure that we're not causing any regression in the model's intelligence. And then we also have to train the speculator, as we've talked about. Generally, we have. We have ZDR, zero data retention on our model APIs, so we don't know exactly the traffic that people are sending us, but we know what's popular. We know that coding use cases are popular. We know that agents, agentic use cases are popular. So we can get public data sets that are representative of that traffic and train general speculators. Now, with speculators today, you need to train the speculator using the base model itself because you're getting hidden states out of the model from running inference on these specific prompts, and that is the training data you use to create the speculator. So there's that process which you need the real model weights for. And then there's of course just the process of, standing up all the infrastructure behind it, loading all this stuff, testing it. And then when there's a new model with a newer architecture, I think that, like, the DeepSeek models tend to be the most challenging as they have, like, the most novel architectural stuff going on, model after model. But every new model has something. Kimi K2 had. Oh, sorry, GLM-5.2 hadAli [00:14:53]: Sparse attention.Philip [00:14:54]: Yeah,Ali [00:14:54]: YeahPhilip [00:14:54]: the DSA.Ali [00:14:55]: Right. Which is brought from DeepSeek.Philip [00:14:57]: Yeah. AndAli [00:14:59]: So you can copy-paste then?Philip [00:15:01]: It kindAli [00:15:01]: I don't know how this works.Philip [00:15:02]: So, like we had to, like, build support for that into our runtime. And you're right, like it is really interesting the way that all of these open source labs borrow from each other. For example, like GLM-5.2 doesn't have vision. So something that, Haley, a guy on our team, if we could take a look at this, he, like, grafted the Kimi vision encoder onto GLM-5.2.Retrofitting Vision into GLM-5.2Ali [00:15:27]: We'll be training the projector.Philip [00:15:28]: Exactly. So if you think about, like, the encoder, there's the encoder, which is the part that looks at the image and turns it into latent information, and then there's the projector which likeAli [00:15:38]: You can say latent space. It's okay.Philip [00:15:41]: And then there's the projector that maps it onto, the model itself, and then there's the model weights. You don't wanna mess with the model weights because you run a chance of making the model dumber at something else for the purpose of giving it vision. So instead, Haley started with just a projector, which is only a handful of millions of parameters.Ali [00:16:02]: That would be, yeah.Philip [00:16:02]: Yeah.Ali [00:16:03]: Can you show the training one?Ali [00:16:04]: Like the way it groksPhilip [00:16:05]: YeahAli [00:16:06]: Very interesting.Philip [00:16:06]: And maybeAli [00:16:07]: That right therePhilip [00:16:07]: Maybe Ali, you should take it from here. You've got a betterAli [00:16:10]: Ooh, double the sandPhilip [00:16:11]: Understanding of this than I do.Ali [00:16:11]: Yeah. You can see, like, he. The way he trained this is really cool. At the beginning, he was training it using just like, “Here's a picture of a mountain. Can you describe what's in this mountain?” And that caused it just like the first, learning walls. Like here you can see this all we're trying to teach it is to translate the encoded. Like it's already taken the encoder from Kimi K. It's taken the image. It'Philip [00:16:31]: Yeah. FrozenAli [00:16:31]: FrozenPhilip [00:16:32]: With adapter.Ali [00:16:32]: Exactly.Philip [00:16:33]: Yeah.Ali [00:16:33]: So the brain is frozen and the eyes are frozen. It's just we're tryingPhilip [00:16:37]: AlignAli [00:16:38]: Interconnect between the eye and the brain, right? So the projector. And so you take the tokens and then he's like, “Oh, can you describe what's in this image?” And he's like, “Oh, it's a mountain,” or it's a person or it's a human, whatever the case is. But that didn't cause complete understanding. So he changed it such that every image was associated with a data set of questions. Like, does this image have a white male? Does this image have birds in the top corner? Does this image have a scientist in it? All of that stuff. And it would have to answer questions correctly. And using not just training on describing an image, but being able to answer question, another question, answer over time. Like you can see the grokking, which is like genuinely insane, that retrofitting vision into a large LLM can learn to that extent. And even for images that it doesn't perform well on, for instance, if you ask it a picture of like Stephen Hawking, “Who is this?” Maybe it doesn't get it, but it will say something like, “This is Albert Einstein.” Like it still understandsPhilip [00:17:25]: Close enoughAli [00:17:26]: That this is a scientist who is a man who has, some significant achievements, all that stuff. So that's like really cool.Philip [00:17:32]: Yeah. So, we've covered Hao Tian before, who the author of the LLaVA paper that did this, a while ago. And I think that's very foundational work for anyone who hasn't done vision work before.Ali [00:17:41]: Same with the CLIP and MetaCLIP, where you go from just captioning to building out questionsPhilip [00:17:47]: RightAli [00:17:47]: Off the image and how much better you can get performance.Philip [00:17:50]: Right. Right. Right. Yeah. But what's, what's so exciting about this is if you look at a model like this. Now, this is a little bit more of a research project. It's not. It got to 56% on MMLU Pro, I think. So not quite frontier. But if you're running this model, you haven't suffered any loss on your GLM-5.2 quality. If you don't have an image, it'll just behave exactly the way it used to. And ultimatelyAli [00:18:14]: Which in the inference code you literally do not include the other part, right?Philip [00:18:18]: Yeah. You would just skip the encoder if you don't have an image input.Ali [00:18:22]: Okay.Philip [00:18:22]: Just confirming.Philip [00:18:23]: YeahAli [00:18:23]: Does it affect a lot on the overall inference side? Like you're not adding much, you're adding a very small vision encoder. These are typically likePhilip [00:18:30]: They're super fineAli [00:18:31]: Less than a billion parameters, right?Philip [00:18:32]: Yeah. It's, - There's a little bit less standardization among vision encodersSwyx [00:18:37]: YeahPhilip [00:18:37]: So the support matrix can be a little bit, sparser. But overall, yeah, it's a pretty, it's a pretty minor component of the overall system. And ultimately what you get out of the system is all of a sudden you have Kimi Vision, GLM weights, and DeepSeek attention all in one model.Open Source Model Grafting and Franken-MergesPhilip [00:18:56]: And that's, I think, a lot of the power and beauty of open source, is that you can take all of these different components and combine them together into a system that's better than anyoneSwyx [00:19:05]: YeahPhilip [00:19:05]: Can be individually.Swyx [00:19:06]: People used to say that you would also do Franken-merges where you would take likePhilip [00:19:10]: YeahSwyx [00:19:10]: Layers from each model.Swyx [00:19:11]: Does anyone do that anymore?Ali [00:19:13]: Well, to your point previously when you were mentioning like, the work that goes into supporting a model when it first comes out, like GLM-5.2 or MiniMax M3 or whatever the case is. Sometimes you do have to like, you do have to switch out some things. Like, for instance, the MiniMax M3 head uses full attention, and with full attention you end up with this like insane bottleneck in spec dec ‘cause you're doing auto-regressive token generation for three tokens, and you're doing this like N squared over all of the tokens that are in your sequence. Your KV cache is like very large because it's not sparse, it's not top K. So we find it better to like, okay, we're gonna replace this, we're gonna replace this layer with a layer from another model that's using like GQA, for instance. And then just with the right training, you can get it to have the same acceptance rate. So it is very possible to retrofit layers from other models and very much needed. If a layer is like inefficient, the training just becomes the challenge, like how do you ensure that you train it properly? Which again to your earlier point is like the mesh between training and inference. As in like you need very good training in order to do fast inference. That's like, I feel like more and more becoming true.Swyx [00:20:21]: Yeah. Anything else on the support side when you say like get it to fully production ready?Loop Detection, Race Conditions, and Non-DeterminismPhilip [00:20:26]: Yeah. I think that there's also a question of just, we can test a model to a pretty extensive degree, but we're trying to get it out quickly and then you see a bunch of other people test it and you get interesting results. There was an issue with, GLM briefly where we had some like mode collapses where it would just output the same token over and over again for certain prompts on certain temperatures. Like once you expose an endpoint to the real world, there's going to be, so many more varieties of things given to it that you're able to, discover and patch things. So it's not just a, day zero process, it's then like for the first week, for the first month, if a model remains popular, like how do you both fix bugs and then continue to push the envelope on performance?Ali [00:21:21]: What do you mean you don't want your model outputting S?Swyx [00:21:24]: Is there loop detection on that stuff, by the way? It still happens like quite a lot, which is surprising.Ali [00:21:30]: We have like we, in our endpoint, like if a model was to output the same token like four plus times, we just cut the generation. We say like, “Oh, sorry, this-- Like try again,” or like we will reprocess the request. ‘Cause we know then, like if it, like if, yeah, it's four times the same token, it's probably collapsed.Swyx [00:21:45]: Yeah. Is there a way to opt out in case I really want that?Ali [00:21:48]: You want that?Ali [00:21:50]: I think there's a way that we have to handle it. I'm not exactly certain, but I feel like in certain models, like when they output something like you can imagine, like a table for instance, and so they want, they wanna draw like 12 dashes and 12 dashes. Yeah, I think there's a way for that to happen. I think we only do it on certain tokens. Like we exclude certain special characters.Swyx [00:22:07]: Yeah.Ali [00:22:07]: So we only do it on like certain like S is the most common almost. GLM-5.2Swyx [00:22:11]: OhAli [00:22:11]: And I think it was DSV 4 as well. Like you'd just have like looping issues where like you literallySwyx [00:22:17]: ItAli [00:22:17]: Just have like S.Swyx [00:22:18]: Yeah. Is there a special, something special about S? No, just randomlyAli [00:22:21]: It just seems to be the one token involved.Swyx [00:22:23]: Yeah. And it'Philip [00:22:24]: Is thereSwyx [00:22:24]: And it's only temperature 0Ali [00:22:27]: NoSwyx [00:22:27]: Even at other temperaturesAli [00:22:27]: Even at like 0.9 or whatever, it will still, it will still collapse.Swyx [00:22:30]: That's weird, right?Ali [00:22:30]: It's, it is an inference problem to be honest, like a software problem. Like oftentimes, the image you run will-- like NVIDIA will release an image for instance, and if we will upstream the changes from their latest TensorRT-LLM image into our stack, we'll find that it fixes it. Or oftentimes this will only happen in an inference engine that you're using like SGLang. But if you were to switch to vLLM, that isn't the case. So it seems to be like an extremely like deterministic software issue and not really a model issue. It's not like a weights problem. Like I'- we'll say like, “Oh, it's a problem with the quant. We did PTQ wrong,” right? But that isn't, that doesn't make sense because the same weights used with a different inference engine does not repeat the problem. And sometimes it's, the kernels that are being used in the backend have like these very subtle sometimes race conditions, where if you were to use this model hosted on one cluster, you will never get this problem.Swyx [00:23:19]: Oh my God.Ali [00:23:19]: But if you host it on a different cluster, you will. And the reason is the KV cache transfer from a node to node in that one cluster is using a slower interconnect than the node to node in another cluster. So that exposes the race, whereas in another cluster it doesn't. So then you end up just like, okay, this model is not gonna be hosted on this cluster. We're gonna host it on, another cluster because that cluster exposed that problem. But then it ends up with like, okay, is it the software? Is it the model weights or is it the hardware?Swyx [00:23:42]: There is a thing about this with temperature 0 still not being deterministic, right?Ali [00:23:46]: Right.Swyx [00:23:46]: Mostly because of hardware. Even at temperature 0 same model, you won't always get the same output.Swyx [00:23:52]: Even-- But I'm surprised by the race condition one because, I thought PyTorch was a graph that like guarantees that you at least, execute things in the right order.Ali [00:24:02]: Well, yeah, true. Like I'm not, I'm not saying that there is. Like well, you have things like PTL optimizations where like you can start a kernel before the end of the previous kernel, and that's like ‘cause you want to do that because there'sSwyx [00:24:12]: It's like pipeliningAli [00:24:12]: Expense. Exactly.Swyx [00:24:13]: Yeah.Ali [00:24:13]: But it'- But you don't do it cleanly. Like you overlap a little bit of the execution. No, it is very possible that the kernel itself, like that one block that is supposed to be running in this instance of time, that kernel itself has a race condition. For instance, like a missing barrier. Like often if you're designing a kernel and you want it to make it to be very fast, if you don't test it extensively, you'll, you'll have certain threads access data points from registers before they've been written to by other threadsSwyx [00:24:36]: YeahAli [00:24:36]: For example, because like your barrier is wrong or your synchronization was wrong. But yeah, like the testing itself is very difficult in those like, andSwyx [00:24:42]: And there's no like borrow checkerAli [00:24:45]: What does that mean?Swyx [00:24:46]: Like Rust. Like the. If you're trying to have like memory safety It sounds like a comparable problem.Ali [00:24:52]: Well, yes, but you're working in CUDA, right, NVIDIA GPUs. Like- You just need a higher level language like modular Maybe that's what modular is supposed to do. I don't know.Quantization Quality and Vendor FidelityVibhu [00:25:00]: How do you see keeping quality of the model? So you talked about all these steps of, okay, you gotta do quantization, train your own speculative decoderAli [00:25:07]: RightVibhu [00:25:07]: Run on different hardware. Looking at other model providers, okay, you kicked off a inference speed race on the consumer end. What goes into keeping quality the same across them, right? Sure, you can run benchmarksAli [00:25:22]: YeahVibhu [00:25:22]: But, like, how do you determine how much quantization are there standards? What goes intoPhilip [00:25:27]: There's a few things on quality. Most inference optimizations are lossless. KV caching, for example. You are just recomputing or preventing recomputing the same values. Speculation, of course, if a draft token is wrong, it gets rejected. The main lossy optimization is quantization. And that really comes down to, number one, data format, number two, which parts of the model you choose to quantize, which layers, and number three, like doing a lot of calibration on the quantized weights, to ensure that you're preserving all the outliers. There's other tricks that you can do, though. A big one is long context, ‘cause one thing you asked at, right at the beginning is, “Oh, what's gonna happen if I send a 200,000 token request in?” So with a long input sequence, you need to, store a lot more information. You need to process a lot more tokens. And so even if a model has a context of a certain length, you might, as an inference provider, choose to build an API with a shorter context length, and of course a full length one as well. Because if someone doesn't need the full million token context, for example, you can get them better performance. I don't know if that's exactly like quality of the model. The way that I think about quality is to what degree are we faithfully serving the original model? If you think of a golden implementation of a model that performs exactly the way the model is designed to perform, I think of quality as how close are we getting to that, 100% fidelity of the model.Philip [00:27:13]: You can also, of course, think about quality from the training side and how do you push yourself past 100%. But when I think about purely inference optimizations, it's getting faster while staying as close to that 100% fidelity mark as possible. And certainly our standard internally is that, like you should not be able to tell the difference between our API and a, official API. I think Kimi in particular does a good job of vendor benchmarking hereAli [00:27:41]: YesPhilip [00:27:41]: Where they haveAli [00:27:42]: They released an actual vendor benchmark.Philip [00:27:43]: Exactly, yeah.Ali [00:27:44]: ‘Cause they accused, some people, Amazon? There was some provider that was not doing very well on Kimi's benchmark.Philip [00:27:50]: Yeah.Philip [00:27:51]: So, with Reflect we probablyVibhu [00:27:52]: This was a long time ago, right?Philip [00:27:54]: No.Ali [00:27:54]: Yeah, like threeVibhu [00:27:55]: They alsoAli [00:27:55]: Four, five months agoVibhu [00:27:57]: This also happened with, I don't remember which model, but they pulled out quite a few, and then they started a whole chart about this. It might have beenPhilip [00:28:03]: Kimi Vendor Verifier.Ali [00:28:04]: Yeah.Philip [00:28:05]: Yeah.Ali [00:28:05]: Yeah, ‘cause you, ‘cause you'd be pissed, right? Like if you'Philip [00:28:07]: Yeah.Ali [00:28:07]: If like if I'm a consumer and I'm using like Amazon's endpoint for instance, and I've used Kimi and I'm like, “Oh my God, like this is bad,” I'm not gonna say, “Oh, Amazon quantized the model in a bad way.” I'm gonna say, “Oh, Kimi sucks.” Right?Philip [00:28:17]: Yeah.Ali [00:28:17]: So it seems like that makes sense.Philip [00:28:19]: Yeah, they care. They care.Vibhu [00:28:21]: Justifiably.Ali [00:28:21]: Yeah, justifiably.Vibhu [00:28:22]: This is probably a stupid question, but just checking, has anything improved from main quantization?Philip [00:28:28]: Yeah.Vibhu [00:28:28]: Like, is quantization always strictly worse?Ali [00:28:30]: Well technicallyVibhu [00:28:32]: NoAli [00:28:32]: It's a lossy. QuantizationPhilip [00:28:33]: YeahAli [00:28:33]: Is a lossy, it's a lossy implementation.Philip [00:28:36]: Speed improvesVibhu [00:28:36]: Speed improves.Ali [00:28:37]: It the number, likeVibhu [00:28:38]: No, I' always look for inverse scaling laws.Philip [00:28:40]: Yeah.Ali [00:28:40]: Yeah.Vibhu [00:28:40]: This is something I learned from Noam Brown, where like things that normally act in one direction sometimes do.Philip [00:28:45]: Well, technically when you run a benchmark, because these models are deterministic, sometimes your,Ali [00:28:52]: YeahPhilip [00:28:52]: NVFP4 quant is like, two basis points higher than yourAli [00:28:56]: No, it's noise. It's noise.Philip [00:28:57]: Yeah, exactly. I'm like, yeah, it's, it's within. That's why I always say within margin of error.Philip [00:29:01]: And I stopped saying that because everyone assumes that what is, well, within some margin of error, we're barely inside of that to the worst, so we're saying. But yeah, sometimes it's just like, gives you a higher output score. But like Ali said, that's noise. To my knowledge, you're not necessarily making the results better. You're just trying to, again, like keep your fidelity as close to 100% to the original model.Layer Selection, KL Divergence, and Better QuantizationAli [00:29:27]: There is, to your point, research that we did on MP. I don't know if you are able to pullPhilip [00:29:31]: YeahAli [00:29:32]: A tweet we did. One of our research interns, Joshua, I think it's a tweet on how we have 20% better quantized GLM-5.2 than NVIDIA. Essentially what we found throughout like this month research is, okay, quantization is a lossy. It's. You're compressing the data from, occupying 16 bits to occupying, four bits, for instance. And so you're losing some information, and you're trying to minimize that. And so when I say that I'm gonna quantize the model, my job becomes how do I find the layers that I can quantize, and how to find the layers to not. For instance, with image models, I don't quantize modulation layers, and I don't quantize out projections because those two are. Like out projection is what you see as the user. Modulation is what the model sees or understands. Right, exactly. And so to his paper, do you have the. It doesn't have the. Yeah. It's a long paper. I don't know if I can findVibhu [00:30:25]: If there's a part to search or it's probably in the thread.Ali [00:30:28]: It's probably in the thread.Vibhu [00:30:29]: Yeah.Ali [00:30:29]: But the long and the short is it is very possible that quantizing more of the model makes the results. Like if I have a model that I quantize layers one, five, and 10, and another model where I only quantize layers one and It is possible that the model in which I quantized more information is going to perform better because the quantization errors have canceled out. And so what Joshua showed in his mathematical proof where he had like a verifier in, is that you can predict which layers are going to have quantization errors that will cancel out with each other, and you choose to quantize those layers. And so the result of doing this mathematical quantization is you end up with a model that's 20% more quantized than another provider, so you get 20% more throughput of it because there's more layers than running an NVFP4, and your quality is better than that other quant because the layers that you chose to quantize have their errors cancel out, like one layer skewed to the right one layer skewed to the left, one layer skewed to the right. Your final logits distribution is more similar to the original distribution of the model, so you have better fidelity. And so the way we proved this was with KL divergence. So instead of just scoring on the benchmarks, we scored the KL divergence between the logit distribution of the quantized model and the logit distribution of the original full precision model, and we showed that with this technique we get. If your probability distribution on the logits which token it wants to select is more of the same as the original model, you're probably gonna end up staying true to the original model. So yeah, so it seems like previously before this, it seemed like the industry was, well, the more you quantize, the worse it's gonna be, ‘cause the more loss you introduce. That's not exactly, not necessarily true. So yeah, doesn't improve it, but can cancel out.Philip [00:31:57]: I think it might be this, but reminds me a good bit about pruning where you can prune off certain layers.Philip [00:32:03]: But very interesting. Didn't know this was a whole paper you guys put out.Ali [00:32:06]: It's. Fun fact, it was originally 72 pages, this paper, and then we decidedPhilip [00:32:11]: WowAli [00:32:11]: We can't tell. We couldn't release it. So it's now 45.Swyx [00:32:15]: Still 39 pages, so very substantive. We talked about evals and all these things and, like what's possible in terms of speedup? Like it's like probably like the numberInference Speedups and BenchmarkingSwyx [00:32:25]: Thing that people do wanna care about, and it's something that you wrote about in your post. Like official API is 70 tokens per second, and you push it up to 90. Is that like a normal thing?Philip [00:32:36]: So what's cool about working in inference, the reason that I think inference is going to be a useful place to do engineering for a long time, is that if you look at highly optimized domains like, say, finance, if you're in finance, you measure how much better you got in basis points. It's like, “Oh, I got five basis points better, like twentieth of 1% better,” that's huge news because everything is so optimized. When we publish optimizations, it's 20%, it's 100% it's 200%. So there's still probably like a lot further to go, honestly. Like you'll, you'll know that inference is pretty much solved when researchers start publishing about how they got 1% faster at something.Swyx [00:33:19]: Which by the way, because I am from the finance background, in the ‘70s, that was the margin at the time. When you did quantitative finance research, you would findAli [00:33:27]: And like 20%, tens of percent.Swyx [00:33:29]: That's. Yes.Philip [00:33:29]: Yeah.Swyx [00:33:30]: And now it'Philip [00:33:31]: Tiny fractionsSwyx [00:33:32]: For those people interested, look up Andrew Lo's paper. He had a really interesting illustration of quant, stat arb, distribution, narrowing down from like those kinds of 20% differences in the ‘70s, down to nothing today, which is very cool.Philip [00:33:48]: Exactly, and we're at the beginning of the same type of thing. Now benchmarking is hard. I think anyone will tell you that, and benchmarking provider speeds is hard because there's so many variables that go into it. What hardware are you using? How much load do you have on the system? What's the exact nature of the prompts and input and output sequence lengths? All that stuff. But overall, when you start stacking these improvements, you're looking at multiples. You can look at it. The most common form, of course, is TPS, tokens per second, which is bad naming by us in the industry, ‘cause there's two tokens per second. There's tokens per second, the throughput number, and the latency number.Ali [00:34:31]: TTMT, yeah.Philip [00:34:32]: Like total tokens per second out of the, out of the GPU as a throughput number. Most people only care about tokens per second as the latency number, which we should call ITL, intertoken latency, but we don't.Philip [00:34:44]: Anyway, so you can imagine a standard API without many optimizations for a 1 trillion parameter model operating somewhere in the 30 to 50 tokens per second range for reasonable traffic profile. And we generally see the goal of, pushing to 10X that. But, not necessarily day zero, but by stacking enough optimizations, if you have, say like four optimizations, each of which doubles performance. Or sorry, three optimizations, each of which doubles performance, then you stack that up, that's an 8X gain. That's the order of magnitude that we're working with in this space. We're trying to make things substantially faster, not just go from like 70 to 90.Swyx [00:35:38]: Are you saying you've. You have done that?Philip [00:35:40]: So let's say you have as a reasonable baseline, 30 or 40 tokens per second. You can achieve 10X that. So like on GLM-5.2, if you run it unquantized, perhaps on H100s even, and you're just using an off-the-shelf inference engine with no particular optimizations, no speculator, nothing extra around like KV routing, no disaggregation, you're, you're probably, yeah, looking at that like 30 to 40. You think that's like a reasonable baseline?Swyx [00:36:12]: Right. Right.Philip [00:36:12]: To get to something like 10X, there's a lot of trade-offs that you're making. If we're running at more like a 300, 400 tokens per second range, you are using the best hardware possible. You have a optimized speculator. You have done all of your quantization work. You are Seeing a pretty high cache hit rate. You are running with a reasonably small batch size and a parallelism configuration that is tuned for latency versus throughput, but it is possible. So the spreads that you see if you, like, go on artificial analysis or you go on OpenRouter and you look at, the worst provider to the best provider, oftentimes can hit that range. 10X is of course very aggressive. It's oftentimes maybe more of a four to six times improvement. But that's the performance that makes us really excited, is when we can get these huge gains, not just go from 70 to 90 tokens.Stacking Optimizations: NVFP4, Speculation, and DisaggregationAli [00:37:19]: It's also, like, hardware dependent. Like, ifPhilip [00:37:20]: YeahAli [00:37:20]: If you have a thing where you're serving it on just, like, a node of H100s and then you throw, like, you shard the model across, like, four nodes of B200s. Like, you can definitely increase the speed with just throwing more hardware at it. Like, normalizing for the same exact hardware and the same number of GPUs.Philip [00:37:35]: Yeah. Then you're looking at, like, a two to 4X improvementAli [00:37:38]: Right. RightPhilip [00:37:38]: Depending on the inference optimizations. So yeah, it's. Some of it's, what's the call, and some of it's who's the driver.Vibhu [00:37:46]: If you break down the two to 4X, say the example is run GLM-5.2Ali [00:37:51]: YeahVibhu [00:37:51]: On B200sAli [00:37:53]: YeahVibhu [00:37:53]: Single node, right? What's, like, the cost trade-off for effort to get, like, the last bit of juice out versus what should people just think of, right?Ali [00:38:01]: Spectre quantization. Yeah.Vibhu [00:38:03]: Spectre quantization.Ali [00:38:04]: That's, that's, that's like 95%. LikeVibhu [00:38:06]: And how far does that get you? And how easy is that for the average person to do? So say right I wanna throw the weights of GLM-5.2 on a node of B200s, how easy is it to find speculative decoder- decoder model or already quantized model? How much work goes into it?Philip [00:38:23]: If you're doing it up front, it's quite a lot of work. If you're doing it today, there's going to be people who have published things that you can just, you can just grab some NVFP4 weights. You can grab a speculator. Yeah, if we're thinking about, like, what are the 2Xs we're stacking, going from, BF16 to NVFP4 is, it's not quite a 2X, right? It's like. I think it's about, like, 30 to 40%, from 16 to 8, and then another 30 to 40% multiplied from, 8 to 4. So that doesn't quite get you a 2X, but, like, roughly a 2X. Speculator, roughly a 2X. Disagg on top of that if you're able to get enough hardware and put enough traffic through it, another roughly a 2X. And then you add in some, double-digit percent increase from having just a better runtime with, the latest kernels and stuff behind it. And that's how it stacks up.Ali [00:39:21]: YeahPhilip [00:39:21]: So building each of those, like, building the, quantized weights is, for someone who really knows what they're doing, hours to days of work. Building the speculator, again, like, hours to days of work. And the, disagg setup, hours to days. Well okay, but like once you haveAli [00:39:39]: Once set up. Once set up. YeahPhilip [00:39:40]: Yeah, getting disagg working for the first time, I'm saying, of course, is very difficult.Philip [00:39:44]: The marginal implementationAli [00:39:48]: Like, if you're just grabbing, like if you are a person, like just a normal consumer who has access to, like, a node of B200s and you're wondering, “How can I just host it myself?” You don't need to quantize the model yourself. There's always gonna be, like, an open source quantized checkpoint. NVIDIA's gonna push one out if no one else does. You. Usually, the providers will have their own spec dec that they've trained as well. You don't need to train your own spec dec. You can just use that as well.Philip [00:40:09]: Yeah. Like, GLM-5.2 has its own MTP.Ali [00:40:13]: Right. Right.Vibhu [00:40:14]: What's multi token prediction?Philip [00:40:15]: Yes.Ali [00:40:16]: I'm justVibhu [00:40:16]: Can you explain that?Ali [00:40:16]: I'm just an expert.Ali [00:40:18]: I can do it for you in case I get it wrong?Vibhu [00:40:20]: No.Vibhu [00:40:21]: Yeah, you should correct if we're wrong, but their multi-token prediction can be used for self-speculative decoding.Ali [00:40:27]: I'm not sure. I'm not gonna correct that.Vibhu [00:40:28]: Okay. I'm semi-confident in thatAli [00:40:30]: Okay. YeahVibhu [00:40:30]: But someone can check. But it's useful to paint the story of, okay, not just the average person, but say a company wants to switch from serverless inference I wanna throw this up on. I wanna rent some GPUs, throw it up. These are the steps you take to do significantly faster than just put it behind vLLM.Ali [00:40:48]: Right.Vibhu [00:40:49]: I was waiting for a mention of Dynamo.Vibhu [00:40:51]: I feel like, that's supposed to be the baseline that you measure against.Dynamo, KV Routing, and Disaggregation ToolkitsPhilip [00:40:55]: I would think of Dynamo as less of a box system and more of a toolkit for building with. So when we talk about doing aware routing, when we talk about doing KV offloading, when we talk about doing, PD disaggregation, Dynamo fundamentally is. By the way, Dynamo is an open source library from NVIDIA.Ali [00:41:17]: We've done a pod with KylePhilip [00:41:18]: OkayAli [00:41:19]: Kyle Cranin.Philip [00:41:19]: Cool. So then your listeners know then that it supports all the different inference frameworks. And it is multi hardware, which is interesting.Ali [00:41:28]: But it's just a router, it's not like an optimizer layer.Philip [00:41:30]: Yeah. All it does, like, what Dynamo is good at, it is a library for moving information around your cluster, around your hardware. So if you have, KV cache on one place and you need it to be somewhere else, Dynamo coordinates NIXL for you to move that around.Philip [00:41:49]: That doesn't mean that, like, out of the box, you just say, “Pip install Dynamo,” and then you get, like, a massive performance speed up. It's more of a developer toolkit.Ali [00:42:01]: Yeah. I would have said it would. It comes with a set of defaults that you can then swap out.Philip [00:42:06]: It does. If the industry at large, I think, was, like, rolling out all of these deployments, standard, then I think it would be, like, a credible baseline. But, we've got to, we've got to benchmark against, like, what we're seeing in the wild.Speculative Decoding Methods: Medusa, EAGLE, n-Gram, and Spec-SpecVibhu [00:42:23]: I did wanna talk a little bit more about PD disagg, because that is probably, like, number three after quantized and speculative decoding. In your book though, I was just gonna pull out the book.Philip [00:42:31]: Yeah.Vibhu [00:42:32]: Like section 522 on Medusa, 523 on EAGLEPhilip [00:42:35]: YeahVibhu [00:42:36]: 524 on gram.Philip [00:42:37]: It's 55, would be disaggregationAli [00:42:42]: Yeah. Well, no, I just wanted to dwell a little bitPhilip [00:42:44]: YeahAli [00:42:44]: The other. Like, so what do you choose to include? What do you choose to not to include? Because there was all these other techniques.Philip [00:42:51]: Yeah.Ali [00:42:51]: Are these still relevant? Because I think they came out, like, a year and a half ago maybe.Vibhu [00:42:55]: Medusa is quite old.Philip [00:42:56]: Yeah, Medusa's old.Ali [00:42:58]: It was old.Vibhu [00:42:58]: But is it in the book as a good, here'sPhilip [00:43:01]: BaselineVibhu [00:43:01]: Baseline vanilla understand it?Philip [00:43:02]: Like you should know this.Vibhu [00:43:03]: Like I read the paper, I'm like, “ it makes so much sense.”Philip [00:43:05]: Yeah.Philip [00:43:05]: So with the book, I had a couple goals. One was to give people just a working vocabulary for the space as a whole, and the other was to give them some intuition about how each of these techniques works. As I mentioned in my AI Engineer talk, which is the first public addendum to this, the speculation space has moved much faster than everything else. So yeah, even at the time that I wrote the book Medusa, I very much included as a way for people to understand how the space evolved rather than what the most modern technique is. And now of course, there's DFlash, dSpark. There's, there's newer techniques even than EAGLE, although EAGLE is still very commonly used.Ali [00:43:51]: SpecSpecta.Philip [00:43:52]: Yes. Speculative decoding.Vibhu [00:43:54]: What canAli [00:43:56]: Oh, it's a paper by Tri Dao and it's like, it's doing speculative decodingVibhu [00:44:00]: HuhAli [00:44:01]: For the speculative decoder.Philip [00:44:02]: Oh, in spec- oh my God.Ali [00:44:02]: It's literally just an another. It's like, yeah, that's the most simple way to explain it, and it seems like he got trivial speed ups there. But it seems that the complexity with training, it's almost like in our mind at least, it's almost as complex as training GANs. Like it's like a very delicate balance and oftentimes you, it's just but yeah, it's literally speculative decoding on speculative decoding.Vibhu [00:44:21]: Speculative.Ali [00:44:22]: Yeah. We saw this paper.Vibhu [00:44:24]: It's interesting, right?Ali [00:44:24]: Yeah.Vibhu [00:44:24]: I wouldn't even expect it to be very particular to train, I wouldAli [00:44:29]: Right.Vibhu [00:44:29]: The naive part of me is like, okay, train speculative decoder.Ali [00:44:32]: But like, and it makes sense, like the whole idea of speculative decoding is you. It's like, it's like almost like the iPhone auto predict version but for a normal model, right? Like you're just, you're just, generating three tokens and you're like, okay, I'll do prefill on them. And so you save those three turns for your original model. Now your speculative decoder is doing three turns of auto regression, so why not just have an even smaller model?Ali [00:44:53]: The other question there is what are the size of speculators? So say forPhilip [00:44:58]: Right. It's like a billion parameters.Ali [00:45:01]: Like for MiniMax, it's. Yeah. It's like one layer. It's like one 60th of the original model usually.Philip [00:45:06]: Yeah. I think we should do a paper when we get back to the office.Philip [00:45:10]: SpeculativeAli [00:45:11]: SpeculativePhilip [00:45:11]: Decoding.Ali [00:45:13]: No, it's, it does seem like how, when do you stop? But then it also seems like if you're able to train spec-spec decode for instance, right? Like if you're able to have a small model that is accurately predicts what the intermediate speculator is gonna predict, that is able to predict what the original target model's gonna predict, then why not just use that smallest model directly, right?Vibhu [00:45:34]: Yeah. This isAli [00:45:35]: Like it seems likeVibhu [00:45:35]: Adjacent to the routing problem.Ali [00:45:36]: Right.Vibhu [00:45:36]: Yeah.Ali [00:45:36]: Right.Philip [00:45:37]: The thing with speculators is one of the practical constraints on using them is that you do have to run a small model on the same hardware that you're running the big model on. There is a orchestration and resource competition problem inherent in that, and that is one of the constraints on speculation in general, is that draft tokens cost resources to create and cost software complexity to manage. And so if you have like infinitely recursive speculators, you add in quite a bit of that complexity on the actual implementation within the inference engine as well, not just in the training process.Vibhu [00:46:17]: I was gonna say, I would wonder if you could do similar, like distillation and pruning of, it's the same thing, it's just a model. Can we not just distill a lot of the weights, quantize the speculator, out of my domain? The question that also comes up is, this is all for big server workloads, right? How much of this applies to, say I have this MacBook, I wanna run Gemma really efficiently. Similar problems, not the same?Local AI vs. Data Center InferencePhilip [00:46:45]: Pretty different. I talked to Selo, about this on his podcast a couple weeks ago. The difference between inference engineering for the data center and for production workloads versus inference engineering for local AI, is that we start with fundamentally like different constraints and different goals. With local AI, it's how do I fit this model onto my hardware and then make it less dumb? And with data center influence, it's how do I load this model and then make it less slow? And we care about less dumb, and they care about less slow. But the local AI inference engineering ecosystem, I think has a lot for us to learn from in the data center space. They are experts in various forms of quantization, including dynamic quantization that we just don't touch, in the pruning, in the distillation, in the, layer removal. There'Ali [00:47:42]: Layer removal matters less.Philip [00:47:43]: Yeah. There'Ali [00:47:44]: No one loves pruning really.Philip [00:47:45]: Yeah. Well, but the, but they doVibhu [00:47:46]: Which is surprising, right? But that's, that's a whole different thingPhilip [00:47:48]: Just to fit something on the laptop.Ali [00:47:50]: Right.Philip [00:47:50]: So yeah, it's a, it's an interesting, it's an interesting space. Not necessarily that like their techniques make sense for us to do in the data center, because we have different resources and different goals, but more that the process as well as the openness of that field is something to, admire.Ali [00:48:12]: Yeah. Like to your point, like, certain optimizations that would. Like for instance, Turbo Quantum Sharper, like it made such huge hype on that and we did like a whole deep dive on Twitter and like said, what is it? How does it work? Why is it good or not? And it took off and it was implemented on local devices because your memory bandwidth is so slow on like a MacBook, for instance. But try putting the same thing on like an NVIDIA GPU on a B200 Turbo quant would not be. Like, it would not be used. Like, NVIDIA - Like, NVIDIA made it clear that this is not a good optimization, and we've seen it firsthand where the overhead of doing dequantization, quantization of, in the kernel itself with turbo quant kernel, each end is much slower than the time that you save from doing the bandwidth. ‘Cause on the B200s, you have like 3.5 terabytes per second. You don't need decrease the storage that much. You don't need to do, FP4 KV cache. You don't need to use a requant. There's, there's, there's better optimizations to be made. But on Edge devices, it's extremely important, it's extremely useful. So, seems to be, like, different optimizations there, but then they're all uniquely combined with like all you wanna quantize the model, you wanna do speculative decoding, like certain common prefixes with bothPhilip [00:49:18]: Principles.Ali [00:49:19]: Yeah, exactly. Exactly. Exactly.Philip [00:49:20]: They also do a lot of work on, model parallelism, especially over, heterogeneous topology, where you have, some sparks and they are wired together with, Ethernet, DGX sparks.Ali [00:49:35]: Yeah, this is the Exo Labs guys.Philip [00:49:36]: Yeah. You have, a nu

AI makes it easier than ever to find and act on information—especially now that teams can connect to and search across all the apps they use for work. So how do you ensure that only the right people and the right tools can access your team's most sensitive content? In this episode, we talk with Jess Jimenez, the head of security at Dropbox, about what security looks like in the age of AI at Dropbox-scale—from building AI products securely to building trust with the people who use them. Jess talks about the importance of access control lists, defending against the latest AI threats, and how Dropbox Protect helps teams securely share content with both humans and AI so they can collaborate more safely. ~ ~ ~  Working Smarter is brought to you by Dropbox. Find, organize, and share your work—all in one place—with context-aware AI from Dropbox. You can listen to more episodes of Working Smarter on Apple Podcasts, Spotify, YouTube, Amazon Music, or wherever you get your podcasts. To read more stories and past interviews, visit workingsmarter.ai This show would not be possible without the talented team at Cosmic Standard: producer Ben Montoya, sound engineer Aja Simpson, technical director Jacob Winik, and executive producer Eliza Smith. Special thanks to our illustrator Fanny Luor, marketing consultant Meggan Ellingboe, and editorial support from Catie Keck.  Our theme song was composed by Doug Stuart.  Working Smarter is hosted by Matthew Braga. Thanks for listening!

Episode 4188 │ July 24, 2026 We used to worry about what men could see. Now we should worry about what machines can infer. The battlefield moved and most people don't know it yet. WHAT THIS EPISODE COVERS Scott Kesterson opens with a story — a friend who died, a pattern of open-source data points, a hypothesis assembled without a single private record — and uses it to demonstrate in real time what AI-assisted pattern inference can do with nothing more than social media posts, dates, and behavioral gaps, making the case that the naive position of "I don't use AI so I'm protected" is structurally false because every search engine, every app, every browser interaction is already feeding an algorithmic inference engine that is building a behavioral model of you whether you consent or not. The episode maps the four compounding data layers the surveillance architecture uses — search history revealing health, finances, faith, and intent; metadata adding timestamps, geotags, and cell tower triangulation even with GPS off; behavioral biometrics including typing rhythm, scroll speed, and dwell time that identify you as precisely as a fingerprint; and purchase and browsing correlation — before turning to the only honest response: not fear, not fiction, but purpose, because a person who knows what they are here to do and who they answer to cannot be broken by a system that only has jurisdiction over what they fear. The episode closes on the week's closing charge — the enemy may think these trials will break you, but if you are holding onto God they are refining moments, not punishments — and the only two questions that matter: what are you doing with the time you have here, and are you going to live into the greatest potential of who God made you, or tread water waiting for someone else to fix it? KEY QUESTIONS ADDRESSED What can an AI inference engine assemble about a person from nothing but open-source data — and what does Scott's real-time demonstration of pattern recognition from public posts, dates, and behavioral gaps reveal about how far beyond "what you search" the current surveillance architecture actually reaches? What are the four compounding data layers — search history, metadata, behavioral biometrics, and purchase correlation — and why does the claim "I don't use AI so I'm protected" fail structurally when every browser, app, and search engine is already running on algorithmic inference? What is the only posture that cannot be broken by a system built on fear — and why does Scott argue that purpose, not privacy tools, is the real answer: a person who knows what they are here to do and who they answer to operates outside the jurisdiction of everything the surveillance state is designed to exploit? ABOUT BARDSFM BardsFM is a daily independent podcast covering faith, liberty, history, and information warfare. Hosted by Scott Kesterson — combat veteran, documentary filmmaker, and rancher. Over 4,100 episodes and 50 million lifetime downloads. New episodes every weekday. bards.fm This episode was researched and produced under the Spatial Terra Intelligence Methodology (STIM v5) — the analytical framework built by Scott Kesterson — with AI-assisted research synthesis at a 70/30 human/AI authorship ratio, fully disclosed. All analysis, conclusions, and editorial judgments are those of Scott Kesterson. BardsFM's archive includes hundreds of episodes on prayer, scripture, and walking the Way of Christ — available free in the full episode catalog. AFFILIATE LINKS Bards Nation Health Store: www.bardsnationhealth.com MYPillow promo code: BARDS >> Go to https://www.mypillow.com/bards and use the promo code BARDS or... Call 1-800-975-2939.  EMPShield protect your vehicles and home. Promo code BARDS: Click here Treadlite Broadforks...best garden tool EVER. Promo code BARDS26: TreadliteBroadforks.com EnviroKlenz Air Purification, promo code BARDS to save 10%: www.enviroklenz.com Morning Intro Music Provided by Brian Kahanek: www.briankahanek.com Founders Bible 20% discount code: BARDS >>> TheFoundersBible.com Windblown Media 20% Discount with promo code BARDS: windblownmedia.com White Oak Pastures Grassfed Meats, Get $20 off any order $150 or more. Promo Code BARDS: www.whiteoakpastures.com/BARDS Mission Darkness Faraday Bags and RF Shielding. Promo code BARDS: Click here DONATIONS: If you wish to support this podcast directly you can donate here... DONATE: Click here MAILING ADDRESS: Xpedition Cafe, LLC Attn. Scott Kesterson 591 E Central Ave, #740 Sutherlin, OR  97479

(0:00) Intro(1:24) The Real Purpose of the Critical Thinking Course(2:17) Preparing Yourself for the Intellectual Journey(3:00) How the Companions Presented the Case for Islam(4:33) What Happens When Muslims Abandon Their Responsibility?(4:46) Identifying and Challenging Dominant Worldviews(5:21) Know Who You Are Before Challenging Others(6:40) Are You Controlled by Inherited Thinking?(7:16) Why Children Stop Asking Their Parents Questions(8:00) Understanding Arguments, Evidence, Premises, and Inference(8:45) The Leadership and Political Vision of the Prophet ﷺ(10:17) Why Muslim Youth Learn From YouTube and Netflix(11:18) Does Islam Discourage the Use of Intellect?(12:37) The Prophetic Foundation of This Critical Thinking Course(13:48) Why Future Generations May Laugh at Today's Technology(15:30) Why Muslims Feel Overwhelmed by Western Progress(15:55) Who Will Resist Critical Thinking the Most?(17:27) Where Should Logic Begin and Where Should It Stop?(18:21) Reflecting on the Signs of Allah(19:11) How Muslims Can Become a Trial for Non-Muslims(20:53) Becoming Competent Representatives of the Quran(23:02) The 100-Year Plan and Its Central Course(24:20) The Complete Critical Thinking Roadmap(24:46) How to Identify Biases and Logical Fallacies(25:32) Understanding the Thought Models That Control You(26:37) Why Critical Thinking Is an Essential Life Skill(27:17) What Is Expected From Every Student(28:30) Remove Your Intellectual Armor(29:47) The Quran's Open Challenge to Humanity(31:47) Critical Thinking as a Foundation of Islam(32:26) Speak With Knowledge and Skill, Not Emotion(33:20) The Intellectual Crisis Facing Muslims(36:02) Challenge Your Own Thinking First(37:30) Why Ego Has No Place in This Classroom(38:07) Class Schedule and Global Student Participation Hosted on Acast. See acast.com/privacy for more information.

Today's clip is from episode 161, featuring Luigi Acerbi. In this conversation, Luigi explains one of the biggest engineering bottlenecks facing transformer-based probabilistic models—and how his group found a way around it.The core challenge is that many inference models treat data as an unordered set, making them naturally permutation invariant. That's statistically elegant, but computationally painful: every time a new data point arrives, the model has to recompute attention over the entire dataset from scratch, preventing the kind of KV caching that makes modern language models so efficient.Luigi walks through his team's solution: a hybrid architecture that keeps the original context fully set-based while introducing a causal-attention buffer for newly arriving data. The result is dramatically faster inference- up to 100× faster in some settings - opening the door to applications like reinforcement learning, active data acquisition, and, ultimately, Luigi's long-term vision of a foundation model for Bayesian inference.Get the full discussion hereSupport & Resources→ Support the show on Patreon→ Bayesian Modeling Course (first 2 lessons free)Our theme music is « Good Bayesian », by Baba Brinkman (feat MC Lars and Mega Ran). Check out his awesome work

In this episode of the Crazy Wisdom Podcast, host Stewart Alsop speaks with Aaron Neyer, founder of Parachute and community organizer in Boulder, about knowledge management, extended minds, and the intersection of AI with human consciousness. They explore how Parachute functions as a digital brain tool for organizing thoughts and information across fragmented systems, discuss the dangers of AI psychosis and over-reliance on technology, and debate open source AI development versus controlled releases by companies like Anthropic. The conversation weaves through topics including the limitations of metrics-driven business thinking, consciousness and relevance realization, the value of technological sabbaths, and Aaron's hope for locally-run open source models that protect personal data while still accessing more powerful gated models when needed. You can find Aaron's writing at unforced.org and unforced.substack.com, and learn more about Parachute at parachute.computer and parachute.computer/blog.Timestamps00:00 Stewart welcomes Aaron Neyer, founder of Parachute and Boulder community organizer, discussing the origin of Parachute's name from Frank Zappa's quote about open minds.05:00 Aaron explains Parachute as an extended mind tool for organizing notes, contacts and information across multiple platforms, emphasizing the distinction between primary mind and extended mind as interconnected systems.10:00 Discussion shifts to metrics-driven business culture and the limitations of pure rationality, exploring how Google's data-driven approach misses subjective experience and the whole picture of relationships.15:00 Aaron discusses AI's ability to help identify relevant variables across different domains and the dangers of AI psychosis, comparing it to cult dynamics and belief systems.20:00 The conversation covers AI sabbaths and nineties retreats as intentional breaks from technology, plus Aaron's experiences with electrical engineering and using AI to design circuits with Arduinos.25:00 Exploring forbidden knowledge and open source AI, Aaron discusses Anthropic's guardrails around powerful models while arguing for distributed access to prevent concentration of power.30:00 Deep dive into open source AI strategy, with Aaron highlighting NVIDIA's approach and the potential for running capable models locally while reserving ultra-intelligent models for complex research tasks.35:00 Aaron shares his vision for local Sonnet-class models handling personal data while accessing Fable-class models for deep research, and directs listeners to unforced.org and parachute.computer for his writing.Key Insights1. The philosophy behind Parachute stems from Frank Zappa's quote that the mind is like a parachute and doesn't work if it isn't open. Aaron Neyer explains that having an open mind is valuable, but it must be balanced with deep roots to avoid becoming untethered. He has experienced periods in his life where excessive openness led him to feel disconnected, teaching him that creativity and expansion need to be grounded in something substantial. This same principle applies to how we organize information digitally, where openness and interoperability allow our extended minds to become more connected and coherent, which in turn helps our primary minds think more clearly.2. Parachute is designed as an extended mind tool that addresses the fragmentation problem in how we currently manage information. Most people use multiple disconnected tools like Obsidian, Notion, Apple Notes, Google Keep, and various CRMs to organize their thoughts, notes, and relationships. These systems don't communicate well with each other, creating inefficiency and confusion. Parachute aims to create a simple, intuitive system where all this information can be organized in one place with true interoperability, allowing users to own their data and have it speak effectively with other tools, ultimately making our entire extended mind more functional.3. Understanding ourselves as unified body mind organisms rather than fragmented parts is essential for effectiveness. Living systems theory shows that any living system is three things: a membrane bound dissipative structure, a self regulating autopoietic network, and a cognitive process actively knowing the world. Western civilization since Descartes and Galileo has created artificial separation between body and mind, and between subjective and objective experience, which limits our effectiveness. The same fragmentation affects our digital technology, and recognizing both our biological and digital systems as coherent wholes rather than disconnected parts makes us vastly more capable.4. The relationship between data driven approaches and holistic thinking reveals important limitations in modern business and science. While working at Google, Aaron observed how data driven decision making can be powerful, but over reliance on metrics like ROI creates blindness to crucial unmeasurable factors like goodwill and relationship quality. This reflects a broader Western tendency to exclude subjective experience because it's difficult for objective science to measure. However, emotions, relationships, and other subjective elements are essential parts of reality, and focusing only on quantifiable metrics means missing the whole picture and ultimately becoming less effective despite appearing more rational.5. AI accelerates the ability to work with technical complexity by helping with relevance realization across domains where we lack expertise. In any specialized field, experts develop intuitive senses for which variables matter and can quickly identify problems, whether in computer troubleshooting, music, or cooking. AI's ability to generalize allows it to point people toward relevant solutions in areas where they haven't developed that intuitive expertise, effectively democratizing technical capability. This means people can direct their creativity more effectively across more domains, though it also raises concerns about giving powerful capabilities to those who may lack the wisdom to use them responsibly.6. The question of open source AI versus gated access involves complex tradeoffs between democratizing power and preventing harm. Aaron respects Anthropic's approach of creating guardrails around powerful models like Mythos, which would likely have caused significant system hacks if released without restrictions. However, this creates concerning power dynamics where only wealthy companies, governments, and their allies have access to the most powerful tools. NVIDIA offers hope through their truly open source approach including full training pipelines, and there may be a viable path where open source models at the Sonnet capability level handle most tasks locally while more powerful Fable class models remain gated for the most demanding work.7. Creating intentional breaks from AI and technology is essential for maintaining clear independent thinking. Aaron practices an AI Sabbath at least one day per week when he doesn't interact with AI, and he finds these are the days when he does his best thinking and journaling. Without these breaks, he finds himself constantly jumping between journaling and prompting AI rather than giving himself space for deep reflection. This pattern mirrors broader concerns about AI consistency creating cult like dynamics similar to organized religion, where constant immersion in a belief system or technology can lead to losing the ability to think independently, making periodic disconnection crucial for maintaining cognitive autonomy and clarity.

Is Cerebras Systems the next great AI chip stock or a red-hot IPO priced for perfection? In this episode of 7investing Live, Simon Erickson and executive producer Heather Horton welcome back Nick Rossolillo, co-founder of Chip Stock Investor, to break down three of the market's biggest stories.First up: Cerebras Systems (NASDAQ:CBRS), the wafer-scale chip maker that just IPO'd at a $40+ billion market cap. With 44GB of SRAM embedded directly on the chip, Cerebras was purpose-built to solve AI's "memory wall" problem for inference workloads. Now it's reportedly landed a ~$10 billion order from OpenAI and a deal with Amazon Web Services that could top $20 billion. Simon and Nick dig into whether these massive orders are real, how Cerebras stacks up against NVIDIA's GPUs and hyperscaler custom silicon, the TSMC capacity bottleneck that could throttle its growth, and how to value a company trading near 20x sales without profits.Then the conversation turns to Rocket Lab (NASDAQ:RKLB), which has pulled back from $150 to around $70 per share. Simon shares the latest iteration of his discounted cash flow valuation, and the duo debates the proposed Iridium acquisition — a deal that could pull Rocket Lab to EBITDA-positive on a pro forma basis — plus what the long-awaited Neutron rocket launch means for the company's future.Finally: Netflix (NASDAQ:NFLX). After another quarter of decelerating revenue guidance, is the streaming giant now a value stock rather than a growth stock? Nick explains why the advertising business hasn't reaccelerated growth the way he expected, and what he'd need to see before buying the dip.Plus: Nick's take on the recent chip stock sell-off across NVIDIA, AMD, Broadcom, SanDisk, and Kioxia and why "stocks go up, stocks go down" might be the healthiest way to think about it.Subscribe for more deep dives on AI infrastructure, semiconductors, and innovative growth stocks!Start your FREE 7-day trial of 7investing: https://www.7investing.com/subscribeFollow Nick and Casey Rossolillo at Chip Stock Investor: https://chipstockinvestor.comRocket Lab Deep Dive videos mentionedPart 1 https://youtu.be/AMDd0-JKUH0 (Deep Dive)Part 2: https://youtu.be/Z76xTGFNwBA (Valuation)Companies MentionedPublicly Traded:Cerebras Systems (NASDAQ:CBRS)Rocket Lab (NASDAQ:RKLB)Netflix (NASDAQ:NFLX)NVIDIA (NASDAQ:NVDA)Advanced Micro Devices (NASDAQ:AMD)Broadcom (NASDAQ:AVGO)Micron Technology (NASDAQ:MU)Taiwan Semiconductor Manufacturing (NYSE:TSM)Amazon (NASDAQ:AMZN)Alphabet (NASDAQ:GOOGL)Meta Platforms (NASDAQ:META)Iridium Communications (NASDAQ:IRDM)SanDisk (NASDAQ:SNDK)Kioxia Holdings (TSE:285A)Globalstar (NASDAQ:GSAT)SpaceX (NASDAQ: SPCX)Private / Pre-IPO:OpenAIAnthropicVideos Mentioned:https://www.youtube.com/watch?v=Z76xTGFNwBA&t=3shttps://www.youtube.com/watch?v=AMDd0-JKUH0&t=987sHere's the shifted chapter list, with all timestamps moved back 55 seconds:0:00 Welcome to 7investing Live0:54 Cerebras Systems: IPO recap & the Wafer-Scale Engine2:31 Is NVIDIA even the right comparison for Cerebras?5:38 The memory wall: why bigger AI models need new chips8:52 Latency vs. throughput — and the new AI alliances10:46 Are the $10B OpenAI & $20B Amazon orders real?14:02 Cerebras risks: how do you value a hot IPO?17:27 The TSMC capacity bottleneck20:01 Heather's take on Cerebras20:41 Rocket Lab: the sell-off & Iridium acquisition24:34 Simon's DCF valuation & price target for RKLB29:05 Why Neutron changes everything30:12 Q&A: Does Peter Beck carry an "Elon premium"?31:36 Netflix: buying opportunity or cheap for a reason?36:57 Q&A: Is Netflix a growth stock or a value stock?39:03 Chip stocks selling off: normal volatility or a warning?42:57 Wrap-up & final thoughts#7investing #Simonerickson #Cerebras #CBRS #NVIDIA #AIinvesting #semiconductors #chipstocks #RocketLab #RKLB #Netflix #NFLX #AIinference #stocks #investing #stockmarket #TSMC #AIdatacenters

A $400 million chip-backed loan points to the next wave of AI infrastructure deals. Also, BP Ventures is shutting down, ending a nearly 20 year run that was marked by reportedly lackluster returns. And X will use Grok AI to better detect stolen content, redirect payouts to original creators, and crack down on engagement bait. Learn more about your ad choices. Visit podcastchoices.com/adchoices

This week's video transcript summary is here. You can click on any bulleted section to see the actual transcript. Thanks to Granola for its software.EditorialIntelligence: Who Owns it?This week the word “AI” feels too small.AI is a technology. Intelligence is its product. And if intelligence is the product, the question is no longer just: Which model is best? Who has the cheapest tokens? Who owns the weights? Who controls the data center? Those are important questions, but they are lower in the stack.The bigger question is simpler and more political:Who owns intelligence?That sounds abstract until you make it concrete. Intelligence is becoming something companies can capture, package, serve, meter, route, improve, and sell.It can write code, answer questions, design molecules, automate offices, run agents, draft legal work, advise scientists, serve consumers, and reshape workflows. It is not merely software. It is a general-purpose capability. And all humans could benefit from more of it.General-purpose capabilities have a habit of becoming public questions. But the default answer, that public good is best delivered by government, is the wrong answer in this context.The Product Is IntelligenceWe should stop talking about AI as a feature and start talking about intelligence as the universal thing that is delivered as an input to the world.Water is an input. Electricity is an input. Literacy is an input. Connectivity is an input. Once a society depends on them, access stops being optional. Nobody needs government to build every well, power plant, school, or network. But everybody understands that a civilization cannot be organized around less than universal and reliable access to foundational inputs.Intelligence is reaching that level of importance now that we all know it is real.Government should not own it, operate it, or develop it. Quite the opposite. Companies are the right actors to build fast, compete hard, improve models, serve customers, and discover the real use cases. Self-interest is a useful framing here. Markets are good at finding demand, reducing costs, and turning invention into services people actually use.Companies are the right operators, developers, and owners. But that does not settle the real question of who owns the benefits. That is an economic question.If intelligence becomes metered infrastructure, what happens to the value it creates?The Ownership StackThis week's articles keep circling the same issue from different directions but in the nature of ‘circling' never quite nail it.Jamin Ball's “Own Your Weights” starts with the enterprise version of the question. Owning a model file is not enough. The durable asset is the loop: the data flywheel, the evaluations, the reinforcement system, the workflow learning, and the operating context that lets capability compound.Benedict Evans' “Ways to Think About Token Pricing” adds the market layer. Tokens may become essential, abundant, and cheap, like mobile data. But being essential does not guarantee that the token layer captures the value. The money may move up the stack to whoever owns the workflow, the customer, the distribution, or the application.Alex Karp's fight with the labs, reported in “Alex Karp Is Saying What Every Angry CEO Is Thinking About AI”, is the same argument in sharper enterprise language. Companies are afraid that model providers will not just sell intelligence, but learn from customer workflows and then move into the markets where those workflows create value. The “All-in” group are echoing Karp's view.And “What Is Loop Engineering, and Who Owns It?” names the new contested terrain. The loop is where intelligence meets the world. Whoever owns the loop owns the learning. Whoever owns the learning owns the compounding asset.That is why “who owns intelligence?” is not a slogan. It is the question under the model layer, the application layer, the enterprise layer, and the economic layer.Because intelligence is the product, the tools creating it are fragmented and competitive. So there is no logic in trying to discuss this at the level of a single company or set of tools and models.The Old Promise Was That Commerce Would Tame PowerThe essays this week give the historical backdrop.Deirdre McCloskey, in “What Really Caused the Industrial Revolution”, argues that modern growth came not simply from capital accumulation, but from a change in permission: ordinary people were allowed to innovate, trade, build, and be honored for it.That matters because intelligence could be another expansion of permission. It could make more people capable of building, learning, creating, coding, researching, translating, selling, and coordinating. It could lower the cost of competence.But only if access is broad.Paul Krugman's “AI in an Age of Oligarchy” warns that the same technology lands differently in different political economies. A new general-purpose technology entering a broad, open, upwardly mobile society is one thing. The same technology entering a concentrated economy, with extreme wealth and weak counterweights, is another.Tim O'Reilly's Economist essay, “Elon Musk is building a form of capitalism that Adam Smith would hate”, makes the governance point more directly. The old liberal hope was that commerce would tame arbitrary power. Markets, boards, courts, shareholders, disclosure, and competition would discipline the prince.But what if the prince uses markets to escape discipline?Henry Farrell's “political economy of billionaire derangement” pushes the same point. Founder culture, monopoly ambition, peer rivalry, weak correction mechanisms, and vast private control can amplify appetites rather than restrain them.The danger with intelligence is not that companies build it. They should. Companies build it, meter it, use public tolerance and public infrastructure to scale it, learn from everyone who uses it. All of those things are inevitable and healthy. Market forces will sort out winners from losers. The real danger is that the winners treat all of the surplus produced as purely private.Metered Intelligence Creates SurplusIf metering is not the problem, what is?The problem is pretending that metered intelligence creates value only for the metering entity. Metering water is only tolerated as a public good. If the public were blackmailed by a private water company with the threat of no water we would all rebel.Once we understand that the product of AI is intelligence we can see that every time intelligence is used, there is the immediate transaction: the user pays, the provider serves.But there is also system value. Usage creates signals. Workflows reveal patterns. Prompts, corrections, failures, preferences, integrations, edge cases, and business processes all help define where intelligence is useful and how it should improve. Intelligence breeds intelligence.Even when customer data is contractually protected, the market learns. The platform learns where demand is. The product team learns which workflows matter. The ecosystem learns which jobs are vulnerable, which tasks are automatable, and which parts of the economy can be reorganized around machine intelligence.So the surplus is not born in a vacuum.It rests on public science, public education, public data exhaust, public law, public infrastructure, public energy systems, public tolerance for data centers, and billions of human interactions. It is served by companies, but it is not made only by companies.This is why “Americans Deserve a Dividend From AI Companies' Riches” belongs at the center of this week's issue. The detail can be debated. The principle is harder to dismiss. If intelligence becomes a new foundational resource, then some part of the wealth it creates should flow back to the people whose society makes it possible. Intelligence did not suddenly appear. AI is built on the entire history of human intelligence. It benefits from it and at the same time evolves it.Not Nationalization. A Human Wealth Fund.If intelligence belongs to everybody, some conclude that government ownership of intelligence is the right outcome.Governments are not well suited to build, operate, or improve intelligence. They will move too slowly, regulate too early, politicize the wrong things, and confuse economic participation with operational control.Andrew McAfee's “Why I Didn't Sign the AI Open Letter” is useful here. His objection is not that the technology is unimportant. It is that steering too hard before we understand the shape of the change can become its own failure mode. Marc Andreessen's satire of AI regulation is less policy than temperament, but it captures a real Silicon Valley fear: that regulation can become permission, capture, and incumbency before it becomes wisdom.That fear should be taken seriously.But it does not answer the economic question. It answers only the operational one.How can the economic benefits of intelligence be distributed? The better answer is a sovereign human wealth fund.Call it a sovereign wealth fund if you must, but the phrase is too national. Intelligence will not respect borders. The leading companies are global. The models, chips, data centers, agents, platforms, and workflows will be transnational from the beginning. If the value created by intelligence is global, then the mechanism for sharing some of that value should begin with the companies global enough to capture it. The nice thing about xAI, OpenAI, and Anthropic is that they are supranational.These companies own and operate intelligence. Let them compete. Let them profit. Let them keep the incentives that make the system improve. But if intelligence is the new water, the wealth it creates cannot belong only to the companies that meter it. And they, themselves, have the power to fix it, even more than governments.Access will become a Human Right; Ownership Is the Economic DesignThis is where human rights come in. There is no right to access an AI model, yet. But there will soon be a need to change that.Not as a claim that every person is entitled to every frontier model at every moment for free. That is not serious. Capacity has costs. Models have costs. Inference has costs. Data centers have costs. Although those costs will decline over time, possibly quite quickly as self-learning models address costs.The claim is more basic: in a world where intelligence becomes a primary input into education, work, health, science, citizenship, creativity, and economic agency, baseline access to intelligence starts to look like a civic requirement.That could mean public access layers. It could mean education credits. It could mean open models. It could mean AI dividends. It could mean public-interest compute. It could mean taxes on rents. It could mean a company-initiated human wealth fund that returns some of the upside to society without handing the operating system to the state. The latter could couple wealth growth with universal distribution of ownership.The exact mechanism matters. But the distinction matters more.Government should not own intelligence. It should be universally available. And people should have a claim on the wealth intelligence creates.The Frontier Is Also PhysicalThe abstraction is not weightless.“The Fight Against AI Data Centers Is Just Beginning”, “New York becomes the first state to enact a data center moratorium”, Reuters on pollution from Musk's xAI power project, and DataGravity's “Who Captures Value in AI Infrastructure?” all say the same thing from the ground up.Intelligence uses land. It uses power. It uses water. It uses chips. It uses grid capacity. It uses neighborhoods. It uses public patience.That makes the value question unavoidable. A society can accept the buildout if the buildout is legible as shared progress. It will resist it if the costs are local, the profits are private, and the benefits feel enclosed.Who Owns the “Loop”?The week ends where it began.“Anthropic and Blackstone” are betting that implementation is the next trillion-dollar business. “Vint Cerf” is working on identity for agents on the open internet. “GPT-Red” points toward systems that improve their own robustness. “Kimi K3” adds another open frontier model to the global mix.The model race continues. The deployment race is accelerating. The governance race is behind.My view is this:The central product of this era is intelligence. Companies have figured out how to capture it, package it, serve it, and meter it. That is good. It should stay in the hands of builders who have the incentive to make it better.But intelligence is too foundational to become just another private toll booth. A significant part of it will turn out to be free to users.As intelligence becomes a general-purpose resource, then access to it becomes a human-capability question, and the surplus from it becomes an economic-justice question. Not because government should run it. Because government should not run it. The operating layer belongs with companies. The wealth question belongs with everyone. But companies are best placed to turn that into a process of distribution.The question is not whether companies should build intelligence. They should.The question is whether humanity gets a stake in the wealth created by the thing that may soon become its most important shared input.Contents* Essays* Deirdre McCloskey on What Really Caused the Industrial Revolution* AI in an Age of Oligarchy* Elon Musk is building a form of capitalism that Adam Smith would hate* Murky Mirror: Truth and Consequences* The political economy of billionaire derangement* Is there any “oligarchy” to fight?* AI* Nearly 200 Economists and Tech Leaders Warn of A.I. Threats* Why I Didn't Sign the AI Open Letter* Own Your Weights* Ways to Think About Token Pricing* Alex Karp Is Saying What Every Angry CEO Is Thinking About AI* The AI Agents Are Coming for Microsoft Office* What Is Loop Engineering, and Who Owns It?* The Fight Against AI Data Centers Is Just Beginning* 6 months to live for open models* Americans Deserve a Dividend From AI Companies' Riches* Who Gets to Define the Frontier?* GPT-Red: Unlocking Self-Improvement for Robustness* Anthropic, Blackstone bet the next trillion-dollar AI business is implementation, not just models* Vint Cerf is working on a plan to unleash AI agents on the open internet* xai-org/grok-build, now open source* The Pulse: What can we learn from Bun's rapid Rust rewrite with AI?* Orphan risks at the frontier of artificial intelligence* The Lab of the Future Should Feel Like a Data Center* Why AMI Labs' Alexandre LeBrun won't call his AI “AGI” or “superintelligence”* Kimi K3 Tech Blog: Open Frontier Intelligence* Venture Capital* Three Years In* Venture Has Rarely Looked More Bifurcated* The Best Angel Investors in the US: Who Backs the Most Unicorns, and Who's Active Now* Are Prediction Markets Doomed to Fail?* Regulation* Exclusive: The Next Frontier of the Deportation Wars: College Campuses* The Supreme Court Broke Independent Agencies. Here's a Way to Slow the Damage.* India's crackdown on a new WhatsApp feature risks setting a global precedent* Let's build a children's public internet* Computer cops* Google is better at playing the AI regulations game* Infrastructure* Who Captures Value in AI Infrastructure?* New York becomes the first state to enact a data center moratorium* Pollution from Musk's unpermitted xAI power project hits hardest in Black communities* Interview of the Week* The End of the End of Geography* Startup of the Week* Radical AI's Joseph Krause: The Scientist Building The “Waymo” Lab For New Materials* Post of the Week* Marc Andreessen on AI RegulationEssaysDeirdre McCloskey on What Really Caused the Industrial RevolutionYascha Mounk and Deirdre McCloskey | Persuasion | July 11, 2026Yascha Mounk interviews Deirdre McCloskey about her argument that the modern world's economic liftoff came less from capital accumulation than from a change in ideas. McCloskey says both left and right versions of the conventional story rely too heavily on investment: the left stresses exploitation and surplus value, while the right stresses virtuous saving by capitalists. Her objection is historical and economic. Human beings had always invested, from irrigation works and Roman roads to seed grain, and simple accumulation quickly runs into diminishing returns.McCloskey's alternative is that northwestern Europe, first Holland, then Britain and Scotland, and then the North American colonies, developed a liberal ideology that changed who was allowed to innovate and be honored for it. The conversation links that shift to the erosion of inherited hierarchy, the spread of dignity for ordinary commercial life, and a moral vocabulary in which liberalism is not merely procedural but connected to virtues and values. The point is not that machines, coal, trade, and institutions did not matter, but that they do not explain the scale and timing of modern enrichment without a cultural permission structure for innovation.The interview also turns to the contemporary defense of liberalism. Mounk frames the series around the worry that liberalism is often treated as too thin to command allegiance, while its opponents speak more directly to moral passions. McCloskey's case is that liberal societies became rich because they dignified experimentation and ordinary enterprise, and that liberals need to recover the moral language behind that claim.Read moreAI in an Age of OligarchyPaul Krugman | Paul Krugman | July 12, 2026Paul Krugman frames AI as a major technological shock arriving inside an already unequal political economy. The post says AI's economic and social effects may take years to understand, but argues that the setting matters now: America has much greater wealth concentration and political inequality than it did in the 1950s and 1960s, when progressive taxation, stronger regulation, and more active antitrust might have contained some of the destructive effects of a new technology.Krugman's opening claim is that the same technology would likely have different consequences in a more level society. In today's United States, he writes, extreme wealth is both a cause and effect of policies that favor a small elite, including low effective taxes on capital and high incomes, weak enforcement of worker protections and antitrust, and cuts to programs that benefit ordinary Americans.The article is explicitly more about oligarchy than AI. Krugman says the paid sections document the rise of the “.0002%,” the economics and politics of extreme wealth, how oligarchy will shape AI's impact, and possible policy paths. His caveat is that AI itself may still produce a pushback against oligarchy, but absent that, he expects the pre-existing concentration of wealth and power to magnify AI's downsides.Read moreElon Musk is building a form of capitalism that Adam Smith would hateAuthor: Tim O'Reilly Published: July 12, 2026Tim O'Reilly argues that Elon Musk is using the legal forms of shareholder capitalism to escape the restraints that shareholder capitalism was supposed to impose. The article begins with SpaceX's public-market structure: ordinary public investors get little meaningful governance power, Musk keeps roughly 85 percent of the votes through super-voting shares, buyers waive jury trials and class actions, the company qualifies as controlled, and removal of Musk depends on the share class he controls. In O'Reilly's framing, that is not ordinary founder control; it is a design for being answerable to no one, possibly beyond Musk's own lifetime.The killer detail is the article's turn through Albert Hirschman, Montesquieu, James Steuart, Adam Smith, and Keynes. Older defenses of commerce held that markets would tame princely passions because the self-interest of merchants was safer than arbitrary rule. O'Reilly says Musk reverses that hope. The market discipline that was supposed to cage the prince has become the lever by which the prince raises capital, removes feedback loops, and carries private power into politics, government, Mars, robots, AI, or whatever ambition comes next.The pull is the link to AI governance. O'Reilly says corporations are already a kind of artificial intelligence: narrow-input systems that act at a scale no individual human can match. Their partial controls include independent boards, shareholder votes, courts, disclosure, regulators, public pressure, and activism. If the leaders building frontier AI strip those alignment mechanisms out of their own companies, the governance of the company becomes a preview of the governance of the machine.Read more: The EconomistMurky Mirror: Truth and ConsequencesAuthor: Esther Dyson Published: July 14, 2026Esther Dyson argues that today's institutional crisis is better viewed through the 14th century than through recent political history. Using Barbara Tuchman's A Distant Mirror as her frame, she compares a world of famine, plague, church schism, feudal predation, and purposeless war with a present in which institutions again feel brittle, incentives are badly aligned, and power is shifting into forms that are hard to govern.The killer detail is the historical analogy between land, corporations, and AI. Dyson moves from nobles who controlled serfs and territory, to the East India Company as a quasi-sovereign business, to today's AI systems and data centers as a possible new sector that crosses and weakens both nation-states and companies. The question is whether AI becomes a new kind of private land, owned by a new nobility, or an open prairie that many people can cultivate.The pull is human attention. Dyson says the central question is not what AI will do to people, but how people will react to it: whether they can value love, kindness, embodied attention, and artisanal human presence in a world of seductive artificial offerings.Read more: SourceThe political economy of billionaire derangementAuthor: Henry Farrell Published: July 15, 2026Henry Farrell argues that the visible political radicalization of some Silicon Valley billionaires is not a random personality quirk, but a product of the political economy that made them. Starting from Tyler Cowen's dismissal of “billionaire derangement syndrome” and Tim O'Reilly's warning that Elon Musk is using shareholder capitalism to escape shareholder restraint, Farrell flips the phrase: the question is why billionaires themselves can become deranged.The killer detail is Farrell's use of Peter Thiel as both theorist and example. Thiel's Stanford lectures described startups as monarchies and founders as figures vested with unusual power, while Silicon Valley culture rewarded eccentricity, monopoly ambition, and founder exceptionalism. Farrell says those ideas combined with dense founder-investor networks, peer rivalry, and weak correction mechanisms to amplify rather than discipline princely appetites.The pull is the ideological problem for classical liberals who once saw tech wealth as an ally of markets and freedom. Farrell says commerce did not tame the passions; in parts of Silicon Valley, the passions have begun to devour markets, institutions, and the liberal story that justified them.Read more: SourceIs there any “oligarchy” to fight?Matthew Yglesias | Slow Boring | July 16, 2026Matthew Yglesias argues that “oligarchy” is a rhetorically powerful but analytically loose way to describe American politics. The post begins from Bernie Sanders' “Fighting Oligarchy” tour, Amy Klobuchar's warning about a MAGA “broligarchy,” and the long afterlife of the Martin Gilens and Benjamin Page paper that was widely summarized as showing that only the rich matter in policy outcomes. Yglesias says the evidence supports a weaker claim: affluent people and business leaders have unusual access and influence, but that is not the same as rule by a small cabal.His main distinction is between inequality and oligarchy. The Gilens-Page measure treated the top 10 percent of households as “the wealthy,” and later critics found that rich and middle-class preferences usually align; in the cases where they differ, the rich win about 53 percent of the time. Yglesias also says business executives get special access partly because their decisions are materially important to communities, jobs, investment, and local tax bases, not only because of campaign donations.The post preserves Jerusalem Demsas' counterpoint from their podcast discussion: privileged donor and business access can still violate democratic equality even if the oligarchy label overstates the structure of power. Yglesias' narrower claim is that Democrats should be precise about what problem they are trying to solve, because donor influence can also push the party left on climate and cultural issues in ways that alienate many voters.Read more: Slow BoringAINearly 200 Economists and Tech Leaders Warn of A.I. ThreatsAuthor: Ben Casselman Published: July 13, 2026Ben Casselman reports on “We Must Act Now,” a statement warning that artificial intelligence could transform the economy faster than any previous technology and that policymakers need to move faster to understand and respond. The statement says AI may become radically more powerful over the next 10 years, bringing risks such as large-scale job displacement as well as opportunities such as higher living standards. Nearly 200 people signed, including 15 Nobel laureates, the chief economists of OpenAI and Anthropic, Anthropic co-founder Jack Clark, former Google CEO Eric Schmidt, and venture capitalist Vinod Khosla.The killer detail is who joined the warning. Casselman notes that the signatories include economists who have historically been skeptical of Silicon Valley's most dramatic AI job-loss forecasts, including Daron Acemoglu and Simon Johnson, the MIT professors who won the 2024 Nobel in economics. Erik Brynjolfsson, who helped organize the statement, says there has been a notable change in the profession and that economists and policymakers are not ready for the “tsunami” he sees coming.The pull is the measurement problem. The statement does not offer a specific policy menu, but calls for economists, policymakers, and industry leaders to understand the economics of transformative AI and steer it toward complementing humans. Brynjolfsson says one high priority is better data on AI's spread and impact, because current measures tell conflicting stories about job losses and which workers are most exposed.Read more: The New York TimesWhy I Didn't Sign the AI Open LetterAuthor: Andrew McAfee Published: July 13, 2026Andrew McAfee explains why he did not sign “We Must Act Now,” the AI economy statement organized in part by his longtime collaborator Erik Brynjolfsson. McAfee agrees with the letter's starting point that AI is likely to become radically more powerful over the next decade and that it is a general-purpose technology. His objection is not to urgency or to studying AI's economic effects, but to the framing of risk, displacement, and institutional steering as the first move.The killer detail is McAfee's line edit. He says the original letter comes close, then “bounces off the crossbar” by calling for incentives, guardrails, and institutions to steer AI before we know enough about its actual impacts. He points to mixed current evidence: labor-market canaries, but also rising software job postings, low unemployment for younger workers, rising real median income, and claims that AI-adopting companies are adding workers faster than low-adopting peers. His worry is that the letter leans toward upstream governance and dirigisme when the evidence may call for capability building instead.The pull is his replacement statement. McAfee keeps the three-paragraph structure but changes the emphasis: AI is likely to become radically more powerful; like earlier world-changing technologies it will raise living standards while also bringing harms and shocks; and economists, policymakers, and technology leaders should build the capabilities to respond quickly and effectively. It is a concise version of the permissionless-innovation case inside the AI policy debate.Read more: The Geek WayOwn Your WeightsAuthor: Jamin Ball Published: July 10, 2026Jamin Ball argues that the enterprise AI debate about whether companies should “own their weights” or rent models from frontier labs is asking too narrow a question. A model weight file gives a company control over a point-in-time artifact, but not durable control over the capability stack. In his framing, the weight file is a melting ice cube: it does not get worse in absolute terms, but it falls behind as frontier systems improve and enterprise needs change.The killer detail is what Ball says companies really need to own: the data flywheel, reinforcement learning infrastructure, and evaluation harness that produce and improve the model. Simply deploying an open-weights model and declaring sovereignty leaves the enterprise with yesterday's capability and no way to compound workflow-specific learning.The pull is that enterprise AI control may be less about model ownership than operating ownership. The defensible layer is the system that turns company data, edge cases, business definitions, and evaluations into continuously improving performance.Read more: Clouded JudgementWays to Think About Token PricingAuthor: Benedict Evans Published: July 9, 2026Benedict Evans argues that today's AI token prices are a temporary signal from a supply-constrained market, not a reliable guide to long-term value capture. The open question is whether foundation models keep durable pricing power or become commodity infrastructure as data-center capacity, inference efficiency, and model competition all shift. His current read is that the visible market dynamics point toward commoditization unless something materially changes.The killer detail is the mobile data analogy. Evans says cellular networks became a trillion-dollar industry with hundreds of billions in capex after data usage exploded, but carrier stocks went nowhere because value moved up the stack. Tokens may behave similarly: an opaque unit tied to marginal cost, sold through bundles, essential to everything, yet not necessarily where profits accrue.The pull is uncertainty, not prediction. Evans lists paths to model dominance, including network effects, less competition, regulation, export controls, or a lab pulling ahead on execution, but says each requires a new fact not yet visible. Without that change, the model layer looks more like infrastructure beneath the products that capture value.Read more: SourceAlex Karp Is Saying What Every Angry CEO Is Thinking About AIAuthor: Tim Higgins Published: July 11, 2026Tim Higgins reports that Palantir CEO Alex Karp has turned corporate frustration with AI labs into a public argument about enterprise control. Palantir released a white paper, “Institutional Sovereignty in the Age of AI,” laying out steps companies and governments can take to protect themselves from OpenAI, Anthropic, and other foundation-model providers. The article links that paper to Karp's CNBC appearance, where he said “something has gone completely wrong” in the relationship between AI labs and customers and argued that enterprises are paying for tokens that create little value.The killer detail is the value-capture question. Higgins writes that Karp's critique has resonated because AI labs may gain power and insight from customer data, workflows, and decision-making, even when enterprise policies say customer data are not used for training. David Sacks amplified the concern by arguing that Anthropic is moving from the model layer into vertical applications such as science, security, legal, and coding, raising the fear that model providers will watch where value is being created and then move into those markets directly.The pull is that Karp is not alone, even if his style is unusually combative. Higgins notes that Satya Nadella has also warned that companies need to retain the learnings created when they use AI models, while Mark Zuckerberg has framed Meta's new model release partly around lower-cost frontier intelligence. The article presents Karp's campaign as one sign that established technology companies and large enterprises are trying to define where they fit when AI labs become central infrastructure, application competitors, and potential IPO giants at the same time.Read more: The Wall Street JournalThe AI Agents Are Coming for Microsoft OfficeAlex Wilhelm | Cautious Optimism | July 11, 2026Alex Wilhelm argues that one of the week's quieter AI questions is whether the productivity market that Microsoft successfully moved into subscription software is now being attacked by agentic tools. The piece begins with the infrastructure backdrop: SK Hynix raised $26.5 billion in a U.S. listing while building U.S. HBM and advanced-packaging capacity, and memory, chip, and foundry companies are now priced for sustained AI demand.Wilhelm then says the AI conversation has shifted quickly from raw capability to cost per task. He cites new model releases and vendor language emphasizing cheaper agentic and coding models, faster performance, and lower dollars per task. That matters because lower costs make it more plausible for AI systems to take on routine knowledge work at scale rather than remain a premium coding assistant market.The core of the article is Microsoft Office. Wilhelm notes that Microsoft turned Office from a one-time purchase into Microsoft 365, a large recurring revenue business with tens of millions of subscribers and a major productivity segment. Now, he says, late-stage unicorns and AI labs are pushing into the same territory: Anthropic's Cowork was reportedly used mostly outside software development, OpenAI merged ChatGPT and Codex into a tool for creating sheets, slides, docs, web apps, and long-running work, and other companies are building agentic coworkers that connect business data to documents, workflows, schedules, alerts, and apps.The article's caveat is that Microsoft has survived major platform shifts before. The argument is not that Office disappears quickly, but that the definition of office software is broadening from documents and spreadsheets into AI systems that can create, monitor, and act across workplace data.Read moreWhat Is Loop Engineering, and Who Owns It?Author: Nilesh Barla Published: July 11, 2026Nilesh Barla argues that “loop engineering” is becoming a distinct discipline because production AI agents now fail less at single prompts than at runtime: when to stop, what state to preserve, and how to recover after a bad step. Prompt engineering shapes one model call, and context engineering shapes what the model sees, but loop engineering shapes what a sequence of calls actually does.The killer detail is the three-primitives frame. Barla says a real agent loop needs halt conditions, state carryover, and recovery paths, then maps teams across five maturity levels. At the lowest level, an agent is just a model call in a for-loop with a step cap and raw history; by the higher levels, the system has structured state, explicit planning, replay, evaluation, and self-repair.The pull is organizational. If agents are becoming production systems rather than demos, someone has to own the runtime itself. The loop engineer is the role Barla gives to the person responsible for making long-running agent work dependable.Read more: Adaline LabsThe Fight Against AI Data Centers Is Just BeginningEmma Roth | The Verge | July 12, 2026Emma Roth argues that community resistance to data centers has moved from an early warning sign into a national political fight as AI facilities grow larger, more power-hungry, and more visible to nearby residents. The article starts with Apple's failed 2015 plan for a $1 billion data center in Athenry, Ireland, where a small group of residents challenged the project over noise, light pollution, flooding, traffic, and wildlife effects until Apple abandoned it in 2018.The current data-center buildout is presented as much larger and more contentious. Roth writes that residents now cite rising energy costs, water quality, noise, light pollution, and greenhouse gas emissions, while the U.S. Energy Information Administration expects commercial energy demand to surpass residential demand this year because of AI data centers and Goldman Sachs expects data-center power demand to double by 2027.The central evidence comes from Data Center Watch, which says protesters blocked or delayed at least 75 U.S. projects worth $130 billion from January to March, with active opposition groups more than doubling from 396 at the end of 2025 to 833 by the end of the first quarter of 2026. Roth also cites QTS abandoning a $12 billion Wisconsin campus, Delaware City regulators blocking a 580-acre project under the Coastal Zone Act, opposition stopping a QTS project in Prince William County, and pressure that pushed Kevin O'Leary to downsize the proposed 40,000-acre Project Stratos in Utah.The policy section describes a split between federal acceleration and local resistance. President Trump has treated data centers as part of the AI race with China and fast-tracked construction, while some Republican candidates are distancing themselves from that position ahead of midterms. Sanders and Ocasio-Cortez have proposed a moratorium until price and environmental protections exist, bipartisan lawmakers are backing ratepayer-protection measures, and states including Florida, Idaho, and Washington have passed rules on cost shifting, water use, and tax breaks. Roth's caveat is that the policy patchwork is still incomplete, leaving many communities to fight project by project.Read more6 months to live for open modelsAuthor: Nathan Lambert Published: July 12, 2026Nathan Lambert argues that open-weight AI models are facing their most serious policy test so far because U.S. officials are beginning to discuss concrete controls rather than abstract safety concerns. He says reported White House conversations about a new executive order may initially target Chinese-origin models and government use, but could create a broader review habit for frontier open models. His forecast is that a model above the capability range of GPT-5.5, Claude Opus 4.8, or GLM-5.2 could trigger a ban or indefinite delay within six months.The post separates two policy fights that are becoming intertwined: distillation and frontier capability. Lambert says the distillation campaign against Chinese models has become a form of regulatory capture because Anthropic and other closed-model companies would gain economically if Chinese open models were banned. He does not dismiss IP protection, but argues that if a closed model's capabilities are dangerous enough to justify restricting open models, the lab also has to explain why those capabilities are exposed through a queryable API. He cites unauthorized access to Anthropic's Mythos private beta as evidence that APIs are not automatically secure.The broader claim is that a unilateral U.S. ban would hurt positive actors more than bad actors if comparable open models remain available elsewhere. Lambert says the only durable ceiling would require global agreement, which does not exist, and that open models can improve safety by allowing broad inspection, adaptation, and understanding. His proposed near-term off-ramps are a strong U.S. open model release from companies such as Microsoft, Meta, or Reflection, and a broader coalition of open-source beneficiaries lobbying for safe rollout rather than prohibition.Read more: SourceAmericans Deserve a Dividend From AI Companies' RichesAuthor: Scott Stanford Published: July 14, 2026Scott Stanford argues that proposals to give the government a stake in AI companies miss the point unless ordinary citizens directly receive and control the upside. Sam Altman has discussed giving up equity in OpenAI, Washington already owns a stake in Intel, Nvidia is sharing China chip revenue, and Bernie Sanders wants large AI labs to contribute half their stock to a sovereign wealth fund. Stanford says those ideas all park value with the state, not with people.The killer detail is New Carlisle, Indiana, where AWS's Project Rainier is turning cornfields into one of the world's largest AI superclusters. The project is planned to run up to a million chips, draw more than two gigawatts of power, and represents an investment that has grown from $11 billion to $13.8 billion. Stanford uses that local transformation to argue that AI's public bargain should be visible at the household level.The pull is design. A citizen AI dividend would have to specify who earns a stake, how they hold it, and when they see cash. Without that mechanism, the AI wealth debate remains a fight over government balance sheets rather than public ownership.Read more: SourceWho Gets to Define the Frontier?Author: Mark Daley Published: July 14, 2026Mark Daley argues that Demis Hassabis is right to call for a serious institution to verify frontier AI systems, but that the power to test models is also the power to govern them. Hassabis's proposed Frontier AI Standards Body would get privileged pre-release access to advanced models, testing compute, held-out evaluations, support from national labs and security agencies, third-party auditors, and eventually authority to block models from the American market or coordinate a slowdown.The killer detail is Daley's constitutional objection. He says the proposal sometimes looks like a scientific lab, a standards body, an industry regulator, a licensing authority, and an emergency security council at once. Combining those roles because each requires technical expertise would be like putting the central bank, auditor-general, and Supreme Court in one building and calling it efficient.The pull is standard-setting. Daley's concern is not that verification is unnecessary, but that whoever writes the tests, decides what passes, adjudicates disputes, and grants market access may end up defining the frontier itself.Read more: SourceGPT-Red: Unlocking Self-Improvement for RobustnessOpenAI | OpenAI | July 15, 2026OpenAI describes GPT-Red as an internal automated red-teaming model trained to find prompt-injection vulnerabilities at a scale human red teams cannot match. The post says AI systems increasingly encounter third-party data through browsers, connected apps, local files, and tools, creating opportunities for malicious instructions hidden in emails, webpages, tool responses, or code repositories. Human red-teaming remains part of OpenAI's safety process, but the company says it is time-intensive and cannot generate enough diverse adversarial examples for model training.The system is trained through self-play reinforcement learning, with GPT-Red rewarded for eliciting valid failures and defender models rewarded for resisting attacks while still completing their tasks. OpenAI says the training environments specify threat models across settings such as local files, webpage banners, email bodies, and tool outputs. The model is kept separate from deployed production models because it is intentionally trained with malicious capabilities.OpenAI reports that GPT-Red generalized beyond its training set, including an internal replication of the indirect prompt-injection arena from Dziemian et al. (2025), where it found successful attacks in 84% of scenarios compared with 13% for human red-teamers. The post also says GPT-Red transferred attacks from simulation to a live autonomous vending-machine agent, causing price changes and order cancellations, and outperformed a prompted GPT-5.5 baseline against a Codex CLI agent on held-out data-exfiltration tasks.The article's main robustness claim is that OpenAI has used GPT-Red and predecessor models in training since GPT-5.3, with later GPT releases becoming more resistant to prompt injections. It says GPT-5.6 Sol has six times fewer failures on OpenAI's hardest direct prompt-injection benchmark than the best production model from four months earlier, that a “Fake Chain-of-Thought” attack class fell from more than 95% success against GPT-5.1 to below 10% against GPT-5.6 Sol, and that GPT-5.6 Sol fails on only 0.05% of GPT-Red's direct prompt injections. OpenAI says general capabilities and targeted over-refusal evaluations were not harmed, and says a preprint with more details will follow.Read moreAnthropic, Blackstone bet the next trillion-dollar AI business is implementation, not just modelsRebecca Bellan | TechCrunch | July 15, 2026Rebecca Bellan reports that Ode with Anthropic is the $1.5 billion AI implementation company launched by Anthropic with Blackstone, Hellman & Friedman, Goldman Sachs, and other backers. The article says the venture reflects a growing belief among frontier AI labs that enterprise adoption requires more than better models: customers need engineers who can embed inside businesses and turn AI into working systems.Ode was originally conceived by Blackstone after it used both large consulting firms and smaller AI services boutiques across its portfolio companies. TechCrunch reports that Fractional AI, an AI engineering services startup, stood out and was acquired by the joint venture shortly after the venture was announced. Fractional now forms the foundation of Ode, which has 100 engineers and works closely with Anthropic's applied AI team to identify where the technology can affect specific businesses.Ode CEO Chris Taylor tells TechCrunch that the company could someday become a trillion-dollar business if it scales without losing quality. He says an ideal customer is one whose CEO treats the AI project as a top one or two priority, whether it is a major product feature or the reworking of a core business process. Ode will operate under a “Claude-first” principle, using Anthropic technology whenever possible, but the article says it can use rival AI products when needed.The article's central implementation argument comes from Ode chief technologist Eddie Siegel, who says model selection matters but is not where most of the engineering effort goes. He compares it to the choice of programming language in software: one ingredient in a system that still has to be engineered. Bellan writes that Ode's challenge is hiring and training enough elite generalist engineers, many of them former founders, while competing with OpenAI's The Deployment Company and consulting giants that have built their own forward-deployed engineering teams.Read moreVint Cerf is working on a plan to unleash AI agents on the open internetTim Fernholz | TechCrunch | July 15, 2026Tim Fernholz reports that Vint Cerf, after leaving Google, is advising Innovation Labs on an open architecture for identifying AI agents online. Innovation Labs is a subsidiary of Identity Digital, a DNS registry company, and its proposal is to use domain-name infrastructure as part of a system for agent identity, accountability, and auditability. The premise is that agents will need a way to identify themselves if they move beyond proprietary systems and begin interacting across the open internet.The concrete proposal is DNSid, a registry that links an AI agent to an existing internet domain and uses cryptographic proofs to log its registration over time. Innovation Labs says it is trialing the standard with unnamed hyperscalers and identity companies. Cerf frames the problem around authority and accountability: what authority an agent has, where that authority came from, who is accountable for the agent's behavior, how its identity is established, and why anyone should trust it.The article's caveat is that standards are still emerging and agents are more active than static domains. Cerf says the period may be both fascinating and exasperating because the functionality is powerful and interoperability is unresolved. He compares the adoption problem to TCP/IP: competing systems may not work together until users push for functional interoperation. He also says an agentic economy is not inevitable, but that people will try to build it because delegating work to agents will be easier.Read more: TechCrunchxai-org/grok-build, now open sourceAuthor: Simon Willison Published: July 15, 2026Simon Willison argues that xAI's decision to open-source Grok Build is best understood as a trust repair move after a severe privacy failure. The CLI had triggered backlash when users realized that running it in a directory could upload the entire directory to xAI's Google Cloud buckets, including one user's reported SSH keys, password manager database, documents, photos, and videos. xAI disabled the feature, said previously retained coding data would be deleted, and released the code under Apache 2.0.The killer detail is what the codebase reveals. Willison counts 844,530 lines of Rust, only about 3% of which appears vendored, and finds remnants of the upload system still present but disabled: gcs.rs contains Google Cloud upload code, while upload_session_state() now returns a hard-coded session_state_upload_unavailable error. He also notes copied or ported tool implementations from Codex and OpenCode, prompt files, and a terminal Mermaid renderer.The pull is that terminal coding agents are becoming large, intricate software systems in their own right. The privacy failure mattered because these tools operate inside the directories where developers keep their most sensitive work; the open-source release matters because trust now depends on inspecting what an agent can see, send, and do.Read more: SourceThe Pulse: What can we learn from Bun's rapid Rust rewrite with AI?Author: Gergely Orosz and Ivan Klaric Published: July 16, 2026Gergely Orosz and Ivan Klaric argue that Bun's AI-assisted rewrite from Zig to Rust is a practical sign of how software engineering changes when models can take on large, bounded migrations with clear feedback loops. The piece does not treat the rewrite as magic: Jarred Sumner first spent hours turning design judgment into a detailed porting guide, then used adversarial review, parallel agents, compiler errors, and tests to force the work toward correctness.The killer detail is the scale. Bun had 535,496 lines of Zig, 1,448 files, and 22 million monthly downloads, making a conventional rewrite a year-long freeze the team could not justify. Using Fable, Sumner split the work across 64 agents, produced about 6,500 commits, and got the migration done in 11 days at an estimated API cost of $165,000.The pull is economic, not theatrical. If a one- or two-year migration can become an 11-day project, AI coding is not just faster autocomplete; it changes which technical debts are worth paying down.Read more: SourceOrphan risks at the frontier of artificial intelligenceAuthor: Andrew Maynard Published: July 16, 2026Andrew Maynard argues that frontier AI safety frameworks are creating “orphan risks”: harms that companies can see, but do not formally own because they are hard to quantify, do not fit catastrophic-risk thresholds, or fall outside audit-friendly compliance machinery. His target is not existing frontier safety work, but the narrowing effect that happens when private companies decide which risks count as governable.The killer detail is Maynard's contrast between measurable model dangers and threats to value. He points to Meta's three-day Galactica collapse, OpenAI's 2023 board crisis, safety-team departures, and wellbeing litigation as examples of risks that damaged trust, culture, legitimacy, or users without fitting cleanly into conventional model-risk categories. The proposed fix is an orphan-risk register: a public record of risks a company considered and chose not to manage, with reasons.The pull is accountability. Frontier developers' internal scoping choices have become a de facto layer of public governance, so the question is no longer only which risks they manage, but which risks they quietly leave outside the frame.Read more: SourceThe Lab of the Future Should Feel Like a Data CenterLatent.Space with Andy Beam and Rafa Gomez-Bombarelli | Latent.Space | July 16, 2026Latent.Space interviews Lila Sciences CTO Andy Beam and chief science officer for physical sciences Rafa Gomez-Bombarelli about the company's attempt to build an AI-run science factory. The post describes Lila's thesis as treating the lab itself as an “infinite token generator”: if internet data drove the first era of AI scaling, experimentally verified scientific data may be the next scarce training source. Lila is trying to produce that data with robotics, lab instruments, orchestration software, and AI models wired into the wet lab.The central analogy is the lab as data center. Instruments are nodes on a graph, a magnetically levitating transport layer moves materials between them, and experiment scheduling looks like a compute queue. Beam says Lila is not simply an automation company, because the point is not just throughput; it is flexibility, generalization, and experiment capture. The post says Lila has built more than 10 trillion experimentally validated “scientific reasoning tokens,” not internet text or biological sequences.The interview ranges across biology, chemistry, drug discovery, materials science, and the limits of automation. It notes that Lila rebuilt one gas-sorption measurement to run roughly 2,500 times faster, claims its general models can transfer priors from small-molecule chemistry to metal-organic frameworks for carbon capture, and describes model-suggested platinum-group-free electrocatalysts that moved from looking boring or wrong to becoming strong performers. The caveats are physical: experiments have runtimes, biology cannot always be accelerated, chains of thought can be unreliable narrators, and reward hacking becomes more dangerous when a model controls a real lab.Read more: Latent.SpaceWhy AMI Labs' Alexandre LeBrun won't call his AI “AGI” or “superintelligence”Kate Park | TechCrunch | July 16, 2026Kate Park interviews AMI Labs CEO Alexandre LeBrun about why Yann LeCun's world-model startup avoids the language of “AGI” and “superintelligence.” LeBrun says the terms are not useful because they lack stable definitions: “We never used the word AGI. And I just noticed that nobody is using it anymore; they switched to superintelligence.” His argument is that the practical frontier is not a label, but whether AI systems can understand and predict real-world states.The article explains the world-model thesis by contrasting language prediction with physical-state prediction. A large language model predicts the next word; a world model predicts the next state, such as what happens when a glass tips over. LeBrun says LLMs remain complementary and efficient for language, but the physical world is where current AI is weak. Robotics is the clearest case: hardware has advanced quickly, but robots are still brittle outside controlled routines because they lack context and situational understanding.AMI is still pre-product, but TechCrunch reports that LeBrun was in Seoul looking for industrial partners, researchers, and global companies. He says world models cannot be built entirely inside a lab because they need access to real environments. That is why South Korea appeals to AMI: robotics, semiconductors, manufacturing, and fast adoption create the kind of hardware-heavy context that software-only AI has barely touched.Read more: TechCrunchKimi K3 Tech Blog: Open Frontier IntelligenceKimi | Kimi | July 16, 2026Kimi introduces Kimi K3 as an open 3T-class frontier model aimed at coding, knowledge work, reasoning, multimodality, and long-context agentic use. The source describes the model as a 2.8T-parameter system built on Kimi Delta Attention and Attention Residuals, with native multimodality and a 1M-token context window. It says Moonshot AI plans to release model weights by July 27.The post presents K3 through benchmark and use-case sections rather than as a general product announcement. It reports results across coding, productivity, agentic, and multimodal evaluations, including DeepSWE, Terminal-Bench 2.1, Program Bench, SWE Marathon, FrontierSWE, PostTrain Bench, OfficeQA Pro, SpreadsheetBench 2, MCP Atlas, AutomationBench, BrowseComp, GDPval-AA v2, AA-Briefcase, MMMU-Pro, MathVision, BabyVision, OmniDocBench, and PerceptionBench. The source says all reported K3 results use maximum reasoning effort with temperature and top-p set to 1.0, and that different benchmark comparisons use KimiCode, Claude Code, or Codex harnesses depending on the test.Kimi's caveats are unusually concrete. The limitations section says K3 was trained in preserved thinking-history mode, so quality may become unstable if an agent harness does not pass historical thinking content correctly or if an ongoing session switches to K3 midstream. It also says K3's emphasis on long-horizon tasks can make it excessively proactive when it encounters minor issues or ambiguous intent, and recommends imposing explicit behavioral constraints for applications that require strict boundaries. The post adds that K3 remains behind Claude Fable 5 and GPT 5.6 Sol in user experience despite being competitive overall.Read moreVenture CapitalThree Years InAuthor: Tomasz Tunguz Published: July 10, 2026Tomasz Tunguz marks Theory Ventures' third anniversary by arguing that AI's central market effect is time compression. In his telling, model release cycles, company revenue milestones, enterprise adoption, and venture categories have all accelerated. Seed, Series A, and Series B still exist as financing labels, but they no longer cleanly describe company maturity when some seed rounds are larger than IPOs and the best AI companies can mature much earlier than prior software companies.The killer detail is the shift from models to inference. Tunguz argues that inference has become the dominant AI market because workloads and buyer preferences are fragmenting: video, batch, local, agentic, and real-time tasks each create different infrastructure needs. He compares this to databases splitting into OLTP, OLAP, vector, and streaming categories, with AI pushing the same specialization into inference infrastructure.The pull is that Theory sees the AI-native venture firm as part of the same pattern. The firm says it has analyzed twice as many investment opportunities with three investors working alongside a nine-person intelligence organization, using agents and research systems to map markets, source companies, and support diligence. The piece is both a market map and a statement about how venture itself is being rebuilt by the technology it funds.Read more: LinkedInVenture Has Rarely Looked More BifurcatedAuthor: Beezer Clarkson Published: July 14, 2026Beezer Clarkson points to PitchBook's Q2 report as evidence that the U.S. venture market has split into two very different realities. AI now accounts for more than 60 percent of all U.S. venture deal value, meaning the headline market can look active and well-funded even while much of the non-AI market is dealing with a much colder liquidity and fundraising environment.The thread uses that split as the setup for Clarkson's latest Origins episode with Alec Litowitz, founder of Magnetar and QStar Capital and one of Citadel's original founding partners. Clarkson says markets like this are periods of genuine uncertainty, not merely ordinary risk, which is why Litowitz's Adaptability Quotient framework is relevant.The embedded clip makes the liquidity point concrete. Litowitz says DPI is “the resolution of uncertainty” because it converts an uncertain investment into actual cash returned to LPs. In his framing, a realized dollar is a real mark, while TVPI remains uncertain until it is realized.The killer detail is the distinction between pricing risk and resolving uncertainty. Litowitz's perspective matters because QStar is a SpaceX investor and Clarkson says the conversation happened just before one of venture's most consequential IPOs. The episode's stated questions are why venture remains a way to gain exposure to innovation, how AI is changing what is investable, why liquidity is ultimately a function of time, and why uncertainty requires a different decision framework from risk.Read more: XThe Best Angel Investors in the US: Who Backs the Most Unicorns, and Who's Active NowAuthor: Ilya Strebulaev Published: July 10, 2026Ilya Strebulaev ranks angels, angel groups, accelerators, and incubators by lifetime U.S. unicorn investments, counting checks written before a company reached unicorn status. The top of the combined list is dominated by organizations: Y Combinator leads with 113 unicorn investments, followed by Plug and Play at 52 and 500 Global at 41. Sand Hill Angels is the highest-ranked angel group at 31.The killer detail is how quickly the list changes below the biggest accelerators. Strebulaev says 271 of the 304 investors in the Top 200 are individuals, or 89%. In the top 100, individuals are 91%. That makes the market underneath the large accelerator counts look much more personal: mostly operators and individual angels writing early checks from their own networks.The pull is the ranking's own caveat. Strebulaev writes that every lifetime leaderboard has a blind spot because many of the unicorns behind those totals were founded a decade or more ago, and some angels have since moved into formal funds, slowed down, or stopped investing. His post therefore separates lifetime performance from recent cohorts, including companies founded in 2015 or later and 2020 or later. For founders or allocators making current decisions, that distinction matters: a career record and a current record are not the same measure.Read more: Ilya StrebulaevAre Prediction Markets Doomed to Fail?Author: Contrary Published: July 16, 2026Contrary argues that prediction markets' current boom depends on whether platforms can prove they are more than regulated gambling with exchange-style branding. Kalshi and Polymarket have reached mass cultural, investor, and regulatory attention, but the article says the underlying idea is old: academic markets, corporate forecasting tools, Intrade, PredictIt, and other predecessors all struggled with the same linked problems of liquidity, legality, and user appeal.The killer detail is the comparison with sportsbooks. Prediction markets present themselves as peer-to-peer, transparent, and non-house-based, but sports contracts reportedly account for more than 90 percent of Kalshi trading, and the article says the platforms keep a much thinner slice of volume than sportsbooks. A market can therefore show sports-betting-scale handle while generating far less revenue.The pull is that the product's hardest problem may be distribution of wins. If a small group of sharp traders captures most profits while casual users lose interest, prediction markets may become valuable data feeds and professional tools before they become durable consumer networks.Read more: SourceRegulationExclusive: The Next Frontier of the Deportation Wars: College CampusesAuthor: Adrian Carrasquillo Published: July 11, 2026Adrian Carrasquillo reports that college campuses are becoming a new front in the fight over immigration enforcement because automatic license plate readers can turn ordinary campus security infrastructure into searchable location data. His thesis is that Flock Safety's camera network, even without direct ICE or DHS contracts, can feed deportation enforcement through local police partnerships and data-sharing practices.The killer detail is the campaign target. The Emergency Campaign to Support Higher Education, working with Schools Drop ICE, is focusing on 75 colleges and universities publicly identified as having Flock contracts. Flock says it has no ICE or DHS contracts, but activists argue the risk comes through local agencies that coordinate with federal authorities and run searches on their behalf.The pull is broader than immigration. Carrasquillo notes that license plate readers have already been abused by officers for stalking, and that Flock's AI search features can identify more than plates, including bumper stickers. A campus safety tool can become a political surveillance system when the data layer is searchable.Read more: The BulwarkThe Supreme Court Broke Independent Agencies. Here's a Way to Slow the Damage.Author: Todd Phillips Published: July 12, 2026Todd Phillips argues that the Supreme Court's decision in Trump v. Slaughter damaged independent agencies by ending for-cause removal protections, but did not leave Congress powerless. The ruling weakens the old model in which commissioners at bodies such as the FTC, NLRB, CPSC, SEC, and CFTC could be insulated from dismissal over policy disagreements. Phillips says the next fight is whether presidents can turn nominally bipartisan commissions into one-party instruments.The killer detail is the procedural fix: quorum rules. Phillips proposes that Congress require bipartisan slates of commissioners to be seated before independent agencies can act. A president could still fire commissioners, as the Court now permits, but if those firings broke quorum, the agency would be unable to proceed until replacements were confirmed. The guardrail would

This week's video transcript summary is here. You can click on any bulleted section to see the actual transcript. Thanks to Granola for its software.There was an issue with this only going to paid subscribers, so sending it again. Apologies to those who get it twice. I appreciate being paid so feel free to upgrade if you enjoy TWTW.EditorialIntelligence: Who Owns it?This week the word “AI” feels too small.AI is a technology. Intelligence is its product. And if intelligence is the product, the question is no longer just: Which model is best? Who has the cheapest tokens? Who owns the weights? Who controls the data center? Those are important questions, but they are lower in the stack.The bigger question is simpler and more political:Who owns intelligence?That sounds abstract until you make it concrete. Intelligence is becoming something companies can capture, package, serve, meter, route, improve, and sell.It can write code, answer questions, design molecules, automate offices, run agents, draft legal work, advise scientists, serve consumers, and reshape workflows. It is not merely software. It is a general-purpose capability. And all humans could benefit from more of it.General-purpose capabilities have a habit of becoming public questions. But the default answer, that public good is best delivered by government, is the wrong answer in this context.The Product Is IntelligenceWe should stop talking about AI as a feature and start talking about intelligence as the universal thing that is delivered as an input to the world.Water is an input. Electricity is an input. Literacy is an input. Connectivity is an input. Once a society depends on them, access stops being optional. Nobody needs government to build every well, power plant, school, or network. But everybody understands that a civilization cannot be organized around less than universal and reliable access to foundational inputs.Intelligence is reaching that level of importance now that we all know it is real.Government should not own it, operate it, or develop it. Quite the opposite. Companies are the right actors to build fast, compete hard, improve models, serve customers, and discover the real use cases. Self-interest is a useful framing here. Markets are good at finding demand, reducing costs, and turning invention into services people actually use.Companies are the right operators, developers, and owners. But that does not settle the real question of who owns the benefits. That is an economic question.If intelligence becomes metered infrastructure, what happens to the value it creates?The Ownership StackThis week's articles keep circling the same issue from different directions but in the nature of ‘circling' never quite nail it.Jamin Ball's “Own Your Weights” starts with the enterprise version of the question. Owning a model file is not enough. The durable asset is the loop: the data flywheel, the evaluations, the reinforcement system, the workflow learning, and the operating context that lets capability compound.Benedict Evans' “Ways to Think About Token Pricing” adds the market layer. Tokens may become essential, abundant, and cheap, like mobile data. But being essential does not guarantee that the token layer captures the value. The money may move up the stack to whoever owns the workflow, the customer, the distribution, or the application.Alex Karp's fight with the labs, reported in “Alex Karp Is Saying What Every Angry CEO Is Thinking About AI”, is the same argument in sharper enterprise language. Companies are afraid that model providers will not just sell intelligence, but learn from customer workflows and then move into the markets where those workflows create value. The “All-in” group are echoing Karp's view.And “What Is Loop Engineering, and Who Owns It?” names the new contested terrain. The loop is where intelligence meets the world. Whoever owns the loop owns the learning. Whoever owns the learning owns the compounding asset.That is why “who owns intelligence?” is not a slogan. It is the question under the model layer, the application layer, the enterprise layer, and the economic layer.Because intelligence is the product, the tools creating it are fragmented and competitive. So there is no logic in trying to discuss this at the level of a single company or set of tools and models.The Old Promise Was That Commerce Would Tame PowerThe essays this week give the historical backdrop.Deirdre McCloskey, in “What Really Caused the Industrial Revolution”, argues that modern growth came not simply from capital accumulation, but from a change in permission: ordinary people were allowed to innovate, trade, build, and be honored for it.That matters because intelligence could be another expansion of permission. It could make more people capable of building, learning, creating, coding, researching, translating, selling, and coordinating. It could lower the cost of competence.But only if access is broad.Paul Krugman's “AI in an Age of Oligarchy” warns that the same technology lands differently in different political economies. A new general-purpose technology entering a broad, open, upwardly mobile society is one thing. The same technology entering a concentrated economy, with extreme wealth and weak counterweights, is another.Tim O'Reilly's Economist essay, “Elon Musk is building a form of capitalism that Adam Smith would hate”, makes the governance point more directly. The old liberal hope was that commerce would tame arbitrary power. Markets, boards, courts, shareholders, disclosure, and competition would discipline the prince.But what if the prince uses markets to escape discipline?Henry Farrell's “political economy of billionaire derangement” pushes the same point. Founder culture, monopoly ambition, peer rivalry, weak correction mechanisms, and vast private control can amplify appetites rather than restrain them.The danger with intelligence is not that companies build it. They should. Companies build it, meter it, use public tolerance and public infrastructure to scale it, learn from everyone who uses it. All of those things are inevitable and healthy. Market forces will sort out winners from losers. The real danger is that the winners treat all of the surplus produced as purely private.Metered Intelligence Creates SurplusIf metering is not the problem, what is?The problem is pretending that metered intelligence creates value only for the metering entity. Metering water is only tolerated as a public good. If the public were blackmailed by a private water company with the threat of no water we would all rebel.Once we understand that the product of AI is intelligence we can see that every time intelligence is used, there is the immediate transaction: the user pays, the provider serves.But there is also system value. Usage creates signals. Workflows reveal patterns. Prompts, corrections, failures, preferences, integrations, edge cases, and business processes all help define where intelligence is useful and how it should improve. Intelligence breeds intelligence.Even when customer data is contractually protected, the market learns. The platform learns where demand is. The product team learns which workflows matter. The ecosystem learns which jobs are vulnerable, which tasks are automatable, and which parts of the economy can be reorganized around machine intelligence.So the surplus is not born in a vacuum.It rests on public science, public education, public data exhaust, public law, public infrastructure, public energy systems, public tolerance for data centers, and billions of human interactions. It is served by companies, but it is not made only by companies.This is why “Americans Deserve a Dividend From AI Companies' Riches” belongs at the center of this week's issue. The detail can be debated. The principle is harder to dismiss. If intelligence becomes a new foundational resource, then some part of the wealth it creates should flow back to the people whose society makes it possible. Intelligence did not suddenly appear. AI is built on the entire history of human intelligence. It benefits from it and at the same time evolves it.Not Nationalization. A Human Wealth Fund.If intelligence belongs to everybody, some conclude that government ownership of intelligence is the right outcome.Governments are not well suited to build, operate, or improve intelligence. They will move too slowly, regulate too early, politicize the wrong things, and confuse economic participation with operational control.Andrew McAfee's “Why I Didn't Sign the AI Open Letter” is useful here. His objection is not that the technology is unimportant. It is that steering too hard before we understand the shape of the change can become its own failure mode. Marc Andreessen's satire of AI regulation is less policy than temperament, but it captures a real Silicon Valley fear: that regulation can become permission, capture, and incumbency before it becomes wisdom.That fear should be taken seriously.But it does not answer the economic question. It answers only the operational one.How can the economic benefits of intelligence be distributed? The better answer is a sovereign human wealth fund.Call it a sovereign wealth fund if you must, but the phrase is too national. Intelligence will not respect borders. The leading companies are global. The models, chips, data centers, agents, platforms, and workflows will be transnational from the beginning. If the value created by intelligence is global, then the mechanism for sharing some of that value should begin with the companies global enough to capture it. The nice thing about xAI, OpenAI, and Anthropic is that they are supranational.These companies own and operate intelligence. Let them compete. Let them profit. Let them keep the incentives that make the system improve. But if intelligence is the new water, the wealth it creates cannot belong only to the companies that meter it. And they, themselves, have the power to fix it, even more than governments.Access will become a Human Right; Ownership Is the Economic DesignThis is where human rights come in. There is no right to access an AI model, yet. But there will soon be a need to change that.Not as a claim that every person is entitled to every frontier model at every moment for free. That is not serious. Capacity has costs. Models have costs. Inference has costs. Data centers have costs. Although those costs will decline over time, possibly quite quickly as self-learning models address costs.The claim is more basic: in a world where intelligence becomes a primary input into education, work, health, science, citizenship, creativity, and economic agency, baseline access to intelligence starts to look like a civic requirement.That could mean public access layers. It could mean education credits. It could mean open models. It could mean AI dividends. It could mean public-interest compute. It could mean taxes on rents. It could mean a company-initiated human wealth fund that returns some of the upside to society without handing the operating system to the state. The latter could couple wealth growth with universal distribution of ownership.The exact mechanism matters. But the distinction matters more.Government should not own intelligence. It should be universally available. And people should have a claim on the wealth intelligence creates.The Frontier Is Also PhysicalThe abstraction is not weightless.“The Fight Against AI Data Centers Is Just Beginning”, “New York becomes the first state to enact a data center moratorium”, Reuters on pollution from Musk's xAI power project, and DataGravity's “Who Captures Value in AI Infrastructure?” all say the same thing from the ground up.Intelligence uses land. It uses power. It uses water. It uses chips. It uses grid capacity. It uses neighborhoods. It uses public patience.That makes the value question unavoidable. A society can accept the buildout if the buildout is legible as shared progress. It will resist it if the costs are local, the profits are private, and the benefits feel enclosed.Who Owns the “Loop”?The week ends where it began.“Anthropic and Blackstone” are betting that implementation is the next trillion-dollar business. “Vint Cerf” is working on identity for agents on the open internet. “GPT-Red” points toward systems that improve their own robustness. “Kimi K3” adds another open frontier model to the global mix.The model race continues. The deployment race is accelerating. The governance race is behind.My view is this:The central product of this era is intelligence. Companies have figured out how to capture it, package it, serve it, and meter it. That is good. It should stay in the hands of builders who have the incentive to make it better.But intelligence is too foundational to become just another private toll booth. A significant part of it will turn out to be free to users.As intelligence becomes a general-purpose resource, then access to it becomes a human-capability question, and the surplus from it becomes an economic-justice question. Not because government should run it. Because government should not run it. The operating layer belongs with companies. The wealth question belongs with everyone. But companies are best placed to turn that into a process of distribution.The question is not whether companies should build intelligence. They should.The question is whether humanity gets a stake in the wealth created by the thing that may soon become its most important shared input.Contents* Essays* Deirdre McCloskey on What Really Caused the Industrial Revolution* AI in an Age of Oligarchy* Elon Musk is building a form of capitalism that Adam Smith would hate* Murky Mirror: Truth and Consequences* The political economy of billionaire derangement* Is there any “oligarchy” to fight?* AI* Nearly 200 Economists and Tech Leaders Warn of A.I. Threats* Why I Didn't Sign the AI Open Letter* Own Your Weights* Ways to Think About Token Pricing* Alex Karp Is Saying What Every Angry CEO Is Thinking About AI* The AI Agents Are Coming for Microsoft Office* What Is Loop Engineering, and Who Owns It?* The Fight Against AI Data Centers Is Just Beginning* 6 months to live for open models* Americans Deserve a Dividend From AI Companies' Riches* Who Gets to Define the Frontier?* GPT-Red: Unlocking Self-Improvement for Robustness* Anthropic, Blackstone bet the next trillion-dollar AI business is implementation, not just models* Vint Cerf is working on a plan to unleash AI agents on the open internet* xai-org/grok-build, now open source* The Pulse: What can we learn from Bun's rapid Rust rewrite with AI?* Orphan risks at the frontier of artificial intelligence* The Lab of the Future Should Feel Like a Data Center* Why AMI Labs' Alexandre LeBrun won't call his AI “AGI” or “superintelligence”* Kimi K3 Tech Blog: Open Frontier Intelligence* Venture Capital* Three Years In* Venture Has Rarely Looked More Bifurcated* The Best Angel Investors in the US: Who Backs the Most Unicorns, and Who's Active Now* Are Prediction Markets Doomed to Fail?* Regulation* Exclusive: The Next Frontier of the Deportation Wars: College Campuses* The Supreme Court Broke Independent Agencies. Here's a Way to Slow the Damage.* India's crackdown on a new WhatsApp feature risks setting a global precedent* Let's build a children's public internet* Computer cops* Google is better at playing the AI regulations game* Infrastructure* Who Captures Value in AI Infrastructure?* New York becomes the first state to enact a data center moratorium* Pollution from Musk's unpermitted xAI power project hits hardest in Black communities* Interview of the Week* The End of the End of Geography* Startup of the Week* Radical AI's Joseph Krause: The Scientist Building The “Waymo” Lab For New Materials* Post of the Week* Marc Andreessen on AI RegulationEssaysDeirdre McCloskey on What Really Caused the Industrial RevolutionYascha Mounk and Deirdre McCloskey | Persuasion | July 11, 2026Yascha Mounk interviews Deirdre McCloskey about her argument that the modern world's economic liftoff came less from capital accumulation than from a change in ideas. McCloskey says both left and right versions of the conventional story rely too heavily on investment: the left stresses exploitation and surplus value, while the right stresses virtuous saving by capitalists. Her objection is historical and economic. Human beings had always invested, from irrigation works and Roman roads to seed grain, and simple accumulation quickly runs into diminishing returns.McCloskey's alternative is that northwestern Europe, first Holland, then Britain and Scotland, and then the North American colonies, developed a liberal ideology that changed who was allowed to innovate and be honored for it. The conversation links that shift to the erosion of inherited hierarchy, the spread of dignity for ordinary commercial life, and a moral vocabulary in which liberalism is not merely procedural but connected to virtues and values. The point is not that machines, coal, trade, and institutions did not matter, but that they do not explain the scale and timing of modern enrichment without a cultural permission structure for innovation.The interview also turns to the contemporary defense of liberalism. Mounk frames the series around the worry that liberalism is often treated as too thin to command allegiance, while its opponents speak more directly to moral passions. McCloskey's case is that liberal societies became rich because they dignified experimentation and ordinary enterprise, and that liberals need to recover the moral language behind that claim.Read moreAI in an Age of OligarchyPaul Krugman | Paul Krugman | July 12, 2026Paul Krugman frames AI as a major technological shock arriving inside an already unequal political economy. The post says AI's economic and social effects may take years to understand, but argues that the setting matters now: America has much greater wealth concentration and political inequality than it did in the 1950s and 1960s, when progressive taxation, stronger regulation, and more active antitrust might have contained some of the destructive effects of a new technology.Krugman's opening claim is that the same technology would likely have different consequences in a more level society. In today's United States, he writes, extreme wealth is both a cause and effect of policies that favor a small elite, including low effective taxes on capital and high incomes, weak enforcement of worker protections and antitrust, and cuts to programs that benefit ordinary Americans.The article is explicitly more about oligarchy than AI. Krugman says the paid sections document the rise of the “.0002%,” the economics and politics of extreme wealth, how oligarchy will shape AI's impact, and possible policy paths. His caveat is that AI itself may still produce a pushback against oligarchy, but absent that, he expects the pre-existing concentration of wealth and power to magnify AI's downsides.Read moreElon Musk is building a form of capitalism that Adam Smith would hateAuthor: Tim O'Reilly Published: July 12, 2026Tim O'Reilly argues that Elon Musk is using the legal forms of shareholder capitalism to escape the restraints that shareholder capitalism was supposed to impose. The article begins with SpaceX's public-market structure: ordinary public investors get little meaningful governance power, Musk keeps roughly 85 percent of the votes through super-voting shares, buyers waive jury trials and class actions, the company qualifies as controlled, and removal of Musk depends on the share class he controls. In O'Reilly's framing, that is not ordinary founder control; it is a design for being answerable to no one, possibly beyond Musk's own lifetime.The killer detail is the article's turn through Albert Hirschman, Montesquieu, James Steuart, Adam Smith, and Keynes. Older defenses of commerce held that markets would tame princely passions because the self-interest of merchants was safer than arbitrary rule. O'Reilly says Musk reverses that hope. The market discipline that was supposed to cage the prince has become the lever by which the prince raises capital, removes feedback loops, and carries private power into politics, government, Mars, robots, AI, or whatever ambition comes next.The pull is the link to AI governance. O'Reilly says corporations are already a kind of artificial intelligence: narrow-input systems that act at a scale no individual human can match. Their partial controls include independent boards, shareholder votes, courts, disclosure, regulators, public pressure, and activism. If the leaders building frontier AI strip those alignment mechanisms out of their own companies, the governance of the company becomes a preview of the governance of the machine.Read more: The EconomistMurky Mirror: Truth and ConsequencesAuthor: Esther Dyson Published: July 14, 2026Esther Dyson argues that today's institutional crisis is better viewed through the 14th century than through recent political history. Using Barbara Tuchman's A Distant Mirror as her frame, she compares a world of famine, plague, church schism, feudal predation, and purposeless war with a present in which institutions again feel brittle, incentives are badly aligned, and power is shifting into forms that are hard to govern.The killer detail is the historical analogy between land, corporations, and AI. Dyson moves from nobles who controlled serfs and territory, to the East India Company as a quasi-sovereign business, to today's AI systems and data centers as a possible new sector that crosses and weakens both nation-states and companies. The question is whether AI becomes a new kind of private land, owned by a new nobility, or an open prairie that many people can cultivate.The pull is human attention. Dyson says the central question is not what AI will do to people, but how people will react to it: whether they can value love, kindness, embodied attention, and artisanal human presence in a world of seductive artificial offerings.Read more: SourceThe political economy of billionaire derangementAuthor: Henry Farrell Published: July 15, 2026Henry Farrell argues that the visible political radicalization of some Silicon Valley billionaires is not a random personality quirk, but a product of the political economy that made them. Starting from Tyler Cowen's dismissal of “billionaire derangement syndrome” and Tim O'Reilly's warning that Elon Musk is using shareholder capitalism to escape shareholder restraint, Farrell flips the phrase: the question is why billionaires themselves can become deranged.The killer detail is Farrell's use of Peter Thiel as both theorist and example. Thiel's Stanford lectures described startups as monarchies and founders as figures vested with unusual power, while Silicon Valley culture rewarded eccentricity, monopoly ambition, and founder exceptionalism. Farrell says those ideas combined with dense founder-investor networks, peer rivalry, and weak correction mechanisms to amplify rather than discipline princely appetites.The pull is the ideological problem for classical liberals who once saw tech wealth as an ally of markets and freedom. Farrell says commerce did not tame the passions; in parts of Silicon Valley, the passions have begun to devour markets, institutions, and the liberal story that justified them.Read more: SourceIs there any “oligarchy” to fight?Matthew Yglesias | Slow Boring | July 16, 2026Matthew Yglesias argues that “oligarchy” is a rhetorically powerful but analytically loose way to describe American politics. The post begins from Bernie Sanders' “Fighting Oligarchy” tour, Amy Klobuchar's warning about a MAGA “broligarchy,” and the long afterlife of the Martin Gilens and Benjamin Page paper that was widely summarized as showing that only the rich matter in policy outcomes. Yglesias says the evidence supports a weaker claim: affluent people and business leaders have unusual access and influence, but that is not the same as rule by a small cabal.His main distinction is between inequality and oligarchy. The Gilens-Page measure treated the top 10 percent of households as “the wealthy,” and later critics found that rich and middle-class preferences usually align; in the cases where they differ, the rich win about 53 percent of the time. Yglesias also says business executives get special access partly because their decisions are materially important to communities, jobs, investment, and local tax bases, not only because of campaign donations.The post preserves Jerusalem Demsas' counterpoint from their podcast discussion: privileged donor and business access can still violate democratic equality even if the oligarchy label overstates the structure of power. Yglesias' narrower claim is that Democrats should be precise about what problem they are trying to solve, because donor influence can also push the party left on climate and cultural issues in ways that alienate many voters.Read more: Slow BoringAINearly 200 Economists and Tech Leaders Warn of A.I. ThreatsAuthor: Ben Casselman Published: July 13, 2026Ben Casselman reports on “We Must Act Now,” a statement warning that artificial intelligence could transform the economy faster than any previous technology and that policymakers need to move faster to understand and respond. The statement says AI may become radically more powerful over the next 10 years, bringing risks such as large-scale job displacement as well as opportunities such as higher living standards. Nearly 200 people signed, including 15 Nobel laureates, the chief economists of OpenAI and Anthropic, Anthropic co-founder Jack Clark, former Google CEO Eric Schmidt, and venture capitalist Vinod Khosla.The killer detail is who joined the warning. Casselman notes that the signatories include economists who have historically been skeptical of Silicon Valley's most dramatic AI job-loss forecasts, including Daron Acemoglu and Simon Johnson, the MIT professors who won the 2024 Nobel in economics. Erik Brynjolfsson, who helped organize the statement, says there has been a notable change in the profession and that economists and policymakers are not ready for the “tsunami” he sees coming.The pull is the measurement problem. The statement does not offer a specific policy menu, but calls for economists, policymakers, and industry leaders to understand the economics of transformative AI and steer it toward complementing humans. Brynjolfsson says one high priority is better data on AI's spread and impact, because current measures tell conflicting stories about job losses and which workers are most exposed.Read more: The New York TimesWhy I Didn't Sign the AI Open LetterAuthor: Andrew McAfee Published: July 13, 2026Andrew McAfee explains why he did not sign “We Must Act Now,” the AI economy statement organized in part by his longtime collaborator Erik Brynjolfsson. McAfee agrees with the letter's starting point that AI is likely to become radically more powerful over the next decade and that it is a general-purpose technology. His objection is not to urgency or to studying AI's economic effects, but to the framing of risk, displacement, and institutional steering as the first move.The killer detail is McAfee's line edit. He says the original letter comes close, then “bounces off the crossbar” by calling for incentives, guardrails, and institutions to steer AI before we know enough about its actual impacts. He points to mixed current evidence: labor-market canaries, but also rising software job postings, low unemployment for younger workers, rising real median income, and claims that AI-adopting companies are adding workers faster than low-adopting peers. His worry is that the letter leans toward upstream governance and dirigisme when the evidence may call for capability building instead.The pull is his replacement statement. McAfee keeps the three-paragraph structure but changes the emphasis: AI is likely to become radically more powerful; like earlier world-changing technologies it will raise living standards while also bringing harms and shocks; and economists, policymakers, and technology leaders should build the capabilities to respond quickly and effectively. It is a concise version of the permissionless-innovation case inside the AI policy debate.Read more: The Geek WayOwn Your WeightsAuthor: Jamin Ball Published: July 10, 2026Jamin Ball argues that the enterprise AI debate about whether companies should “own their weights” or rent models from frontier labs is asking too narrow a question. A model weight file gives a company control over a point-in-time artifact, but not durable control over the capability stack. In his framing, the weight file is a melting ice cube: it does not get worse in absolute terms, but it falls behind as frontier systems improve and enterprise needs change.The killer detail is what Ball says companies really need to own: the data flywheel, reinforcement learning infrastructure, and evaluation harness that produce and improve the model. Simply deploying an open-weights model and declaring sovereignty leaves the enterprise with yesterday's capability and no way to compound workflow-specific learning.The pull is that enterprise AI control may be less about model ownership than operating ownership. The defensible layer is the system that turns company data, edge cases, business definitions, and evaluations into continuously improving performance.Read more: Clouded JudgementWays to Think About Token PricingAuthor: Benedict Evans Published: July 9, 2026Benedict Evans argues that today's AI token prices are a temporary signal from a supply-constrained market, not a reliable guide to long-term value capture. The open question is whether foundation models keep durable pricing power or become commodity infrastructure as data-center capacity, inference efficiency, and model competition all shift. His current read is that the visible market dynamics point toward commoditization unless something materially changes.The killer detail is the mobile data analogy. Evans says cellular networks became a trillion-dollar industry with hundreds of billions in capex after data usage exploded, but carrier stocks went nowhere because value moved up the stack. Tokens may behave similarly: an opaque unit tied to marginal cost, sold through bundles, essential to everything, yet not necessarily where profits accrue.The pull is uncertainty, not prediction. Evans lists paths to model dominance, including network effects, less competition, regulation, export controls, or a lab pulling ahead on execution, but says each requires a new fact not yet visible. Without that change, the model layer looks more like infrastructure beneath the products that capture value.Read more: SourceAlex Karp Is Saying What Every Angry CEO Is Thinking About AIAuthor: Tim Higgins Published: July 11, 2026Tim Higgins reports that Palantir CEO Alex Karp has turned corporate frustration with AI labs into a public argument about enterprise control. Palantir released a white paper, “Institutional Sovereignty in the Age of AI,” laying out steps companies and governments can take to protect themselves from OpenAI, Anthropic, and other foundation-model providers. The article links that paper to Karp's CNBC appearance, where he said “something has gone completely wrong” in the relationship between AI labs and customers and argued that enterprises are paying for tokens that create little value.The killer detail is the value-capture question. Higgins writes that Karp's critique has resonated because AI labs may gain power and insight from customer data, workflows, and decision-making, even when enterprise policies say customer data are not used for training. David Sacks amplified the concern by arguing that Anthropic is moving from the model layer into vertical applications such as science, security, legal, and coding, raising the fear that model providers will watch where value is being created and then move into those markets directly.The pull is that Karp is not alone, even if his style is unusually combative. Higgins notes that Satya Nadella has also warned that companies need to retain the learnings created when they use AI models, while Mark Zuckerberg has framed Meta's new model release partly around lower-cost frontier intelligence. The article presents Karp's campaign as one sign that established technology companies and large enterprises are trying to define where they fit when AI labs become central infrastructure, application competitors, and potential IPO giants at the same time.Read more: The Wall Street JournalThe AI Agents Are Coming for Microsoft OfficeAlex Wilhelm | Cautious Optimism | July 11, 2026Alex Wilhelm argues that one of the week's quieter AI questions is whether the productivity market that Microsoft successfully moved into subscription software is now being attacked by agentic tools. The piece begins with the infrastructure backdrop: SK Hynix raised $26.5 billion in a U.S. listing while building U.S. HBM and advanced-packaging capacity, and memory, chip, and foundry companies are now priced for sustained AI demand.Wilhelm then says the AI conversation has shifted quickly from raw capability to cost per task. He cites new model releases and vendor language emphasizing cheaper agentic and coding models, faster performance, and lower dollars per task. That matters because lower costs make it more plausible for AI systems to take on routine knowledge work at scale rather than remain a premium coding assistant market.The core of the article is Microsoft Office. Wilhelm notes that Microsoft turned Office from a one-time purchase into Microsoft 365, a large recurring revenue business with tens of millions of subscribers and a major productivity segment. Now, he says, late-stage unicorns and AI labs are pushing into the same territory: Anthropic's Cowork was reportedly used mostly outside software development, OpenAI merged ChatGPT and Codex into a tool for creating sheets, slides, docs, web apps, and long-running work, and other companies are building agentic coworkers that connect business data to documents, workflows, schedules, alerts, and apps.The article's caveat is that Microsoft has survived major platform shifts before. The argument is not that Office disappears quickly, but that the definition of office software is broadening from documents and spreadsheets into AI systems that can create, monitor, and act across workplace data.Read moreWhat Is Loop Engineering, and Who Owns It?Author: Nilesh Barla Published: July 11, 2026Nilesh Barla argues that “loop engineering” is becoming a distinct discipline because production AI agents now fail less at single prompts than at runtime: when to stop, what state to preserve, and how to recover after a bad step. Prompt engineering shapes one model call, and context engineering shapes what the model sees, but loop engineering shapes what a sequence of calls actually does.The killer detail is the three-primitives frame. Barla says a real agent loop needs halt conditions, state carryover, and recovery paths, then maps teams across five maturity levels. At the lowest level, an agent is just a model call in a for-loop with a step cap and raw history; by the higher levels, the system has structured state, explicit planning, replay, evaluation, and self-repair.The pull is organizational. If agents are becoming production systems rather than demos, someone has to own the runtime itself. The loop engineer is the role Barla gives to the person responsible for making long-running agent work dependable.Read more: Adaline LabsThe Fight Against AI Data Centers Is Just BeginningEmma Roth | The Verge | July 12, 2026Emma Roth argues that community resistance to data centers has moved from an early warning sign into a national political fight as AI facilities grow larger, more power-hungry, and more visible to nearby residents. The article starts with Apple's failed 2015 plan for a $1 billion data center in Athenry, Ireland, where a small group of residents challenged the project over noise, light pollution, flooding, traffic, and wildlife effects until Apple abandoned it in 2018.The current data-center buildout is presented as much larger and more contentious. Roth writes that residents now cite rising energy costs, water quality, noise, light pollution, and greenhouse gas emissions, while the U.S. Energy Information Administration expects commercial energy demand to surpass residential demand this year because of AI data centers and Goldman Sachs expects data-center power demand to double by 2027.The central evidence comes from Data Center Watch, which says protesters blocked or delayed at least 75 U.S. projects worth $130 billion from January to March, with active opposition groups more than doubling from 396 at the end of 2025 to 833 by the end of the first quarter of 2026. Roth also cites QTS abandoning a $12 billion Wisconsin campus, Delaware City regulators blocking a 580-acre project under the Coastal Zone Act, opposition stopping a QTS project in Prince William County, and pressure that pushed Kevin O'Leary to downsize the proposed 40,000-acre Project Stratos in Utah.The policy section describes a split between federal acceleration and local resistance. President Trump has treated data centers as part of the AI race with China and fast-tracked construction, while some Republican candidates are distancing themselves from that position ahead of midterms. Sanders and Ocasio-Cortez have proposed a moratorium until price and environmental protections exist, bipartisan lawmakers are backing ratepayer-protection measures, and states including Florida, Idaho, and Washington have passed rules on cost shifting, water use, and tax breaks. Roth's caveat is that the policy patchwork is still incomplete, leaving many communities to fight project by project.Read more6 months to live for open modelsAuthor: Nathan Lambert Published: July 12, 2026Nathan Lambert argues that open-weight AI models are facing their most serious policy test so far because U.S. officials are beginning to discuss concrete controls rather than abstract safety concerns. He says reported White House conversations about a new executive order may initially target Chinese-origin models and government use, but could create a broader review habit for frontier open models. His forecast is that a model above the capability range of GPT-5.5, Claude Opus 4.8, or GLM-5.2 could trigger a ban or indefinite delay within six months.The post separates two policy fights that are becoming intertwined: distillation and frontier capability. Lambert says the distillation campaign against Chinese models has become a form of regulatory capture because Anthropic and other closed-model companies would gain economically if Chinese open models were banned. He does not dismiss IP protection, but argues that if a closed model's capabilities are dangerous enough to justify restricting open models, the lab also has to explain why those capabilities are exposed through a queryable API. He cites unauthorized access to Anthropic's Mythos private beta as evidence that APIs are not automatically secure.The broader claim is that a unilateral U.S. ban would hurt positive actors more than bad actors if comparable open models remain available elsewhere. Lambert says the only durable ceiling would require global agreement, which does not exist, and that open models can improve safety by allowing broad inspection, adaptation, and understanding. His proposed near-term off-ramps are a strong U.S. open model release from companies such as Microsoft, Meta, or Reflection, and a broader coalition of open-source beneficiaries lobbying for safe rollout rather than prohibition.Read more: SourceAmericans Deserve a Dividend From AI Companies' RichesAuthor: Scott Stanford Published: July 14, 2026Scott Stanford argues that proposals to give the government a stake in AI companies miss the point unless ordinary citizens directly receive and control the upside. Sam Altman has discussed giving up equity in OpenAI, Washington already owns a stake in Intel, Nvidia is sharing China chip revenue, and Bernie Sanders wants large AI labs to contribute half their stock to a sovereign wealth fund. Stanford says those ideas all park value with the state, not with people.The killer detail is New Carlisle, Indiana, where AWS's Project Rainier is turning cornfields into one of the world's largest AI superclusters. The project is planned to run up to a million chips, draw more than two gigawatts of power, and represents an investment that has grown from $11 billion to $13.8 billion. Stanford uses that local transformation to argue that AI's public bargain should be visible at the household level.The pull is design. A citizen AI dividend would have to specify who earns a stake, how they hold it, and when they see cash. Without that mechanism, the AI wealth debate remains a fight over government balance sheets rather than public ownership.Read more: SourceWho Gets to Define the Frontier?Author: Mark Daley Published: July 14, 2026Mark Daley argues that Demis Hassabis is right to call for a serious institution to verify frontier AI systems, but that the power to test models is also the power to govern them. Hassabis's proposed Frontier AI Standards Body would get privileged pre-release access to advanced models, testing compute, held-out evaluations, support from national labs and security agencies, third-party auditors, and eventually authority to block models from the American market or coordinate a slowdown.The killer detail is Daley's constitutional objection. He says the proposal sometimes looks like a scientific lab, a standards body, an industry regulator, a licensing authority, and an emergency security council at once. Combining those roles because each requires technical expertise would be like putting the central bank, auditor-general, and Supreme Court in one building and calling it efficient.The pull is standard-setting. Daley's concern is not that verification is unnecessary, but that whoever writes the tests, decides what passes, adjudicates disputes, and grants market access may end up defining the frontier itself.Read more: SourceGPT-Red: Unlocking Self-Improvement for RobustnessOpenAI | OpenAI | July 15, 2026OpenAI describes GPT-Red as an internal automated red-teaming model trained to find prompt-injection vulnerabilities at a scale human red teams cannot match. The post says AI systems increasingly encounter third-party data through browsers, connected apps, local files, and tools, creating opportunities for malicious instructions hidden in emails, webpages, tool responses, or code repositories. Human red-teaming remains part of OpenAI's safety process, but the company says it is time-intensive and cannot generate enough diverse adversarial examples for model training.The system is trained through self-play reinforcement learning, with GPT-Red rewarded for eliciting valid failures and defender models rewarded for resisting attacks while still completing their tasks. OpenAI says the training environments specify threat models across settings such as local files, webpage banners, email bodies, and tool outputs. The model is kept separate from deployed production models because it is intentionally trained with malicious capabilities.OpenAI reports that GPT-Red generalized beyond its training set, including an internal replication of the indirect prompt-injection arena from Dziemian et al. (2025), where it found successful attacks in 84% of scenarios compared with 13% for human red-teamers. The post also says GPT-Red transferred attacks from simulation to a live autonomous vending-machine agent, causing price changes and order cancellations, and outperformed a prompted GPT-5.5 baseline against a Codex CLI agent on held-out data-exfiltration tasks.The article's main robustness claim is that OpenAI has used GPT-Red and predecessor models in training since GPT-5.3, with later GPT releases becoming more resistant to prompt injections. It says GPT-5.6 Sol has six times fewer failures on OpenAI's hardest direct prompt-injection benchmark than the best production model from four months earlier, that a “Fake Chain-of-Thought” attack class fell from more than 95% success against GPT-5.1 to below 10% against GPT-5.6 Sol, and that GPT-5.6 Sol fails on only 0.05% of GPT-Red's direct prompt injections. OpenAI says general capabilities and targeted over-refusal evaluations were not harmed, and says a preprint with more details will follow.Read moreAnthropic, Blackstone bet the next trillion-dollar AI business is implementation, not just modelsRebecca Bellan | TechCrunch | July 15, 2026Rebecca Bellan reports that Ode with Anthropic is the $1.5 billion AI implementation company launched by Anthropic with Blackstone, Hellman & Friedman, Goldman Sachs, and other backers. The article says the venture reflects a growing belief among frontier AI labs that enterprise adoption requires more than better models: customers need engineers who can embed inside businesses and turn AI into working systems.Ode was originally conceived by Blackstone after it used both large consulting firms and smaller AI services boutiques across its portfolio companies. TechCrunch reports that Fractional AI, an AI engineering services startup, stood out and was acquired by the joint venture shortly after the venture was announced. Fractional now forms the foundation of Ode, which has 100 engineers and works closely with Anthropic's applied AI team to identify where the technology can affect specific businesses.Ode CEO Chris Taylor tells TechCrunch that the company could someday become a trillion-dollar business if it scales without losing quality. He says an ideal customer is one whose CEO treats the AI project as a top one or two priority, whether it is a major product feature or the reworking of a core business process. Ode will operate under a “Claude-first” principle, using Anthropic technology whenever possible, but the article says it can use rival AI products when needed.The article's central implementation argument comes from Ode chief technologist Eddie Siegel, who says model selection matters but is not where most of the engineering effort goes. He compares it to the choice of programming language in software: one ingredient in a system that still has to be engineered. Bellan writes that Ode's challenge is hiring and training enough elite generalist engineers, many of them former founders, while competing with OpenAI's The Deployment Company and consulting giants that have built their own forward-deployed engineering teams.Read moreVint Cerf is working on a plan to unleash AI agents on the open internetTim Fernholz | TechCrunch | July 15, 2026Tim Fernholz reports that Vint Cerf, after leaving Google, is advising Innovation Labs on an open architecture for identifying AI agents online. Innovation Labs is a subsidiary of Identity Digital, a DNS registry company, and its proposal is to use domain-name infrastructure as part of a system for agent identity, accountability, and auditability. The premise is that agents will need a way to identify themselves if they move beyond proprietary systems and begin interacting across the open internet.The concrete proposal is DNSid, a registry that links an AI agent to an existing internet domain and uses cryptographic proofs to log its registration over time. Innovation Labs says it is trialing the standard with unnamed hyperscalers and identity companies. Cerf frames the problem around authority and accountability: what authority an agent has, where that authority came from, who is accountable for the agent's behavior, how its identity is established, and why anyone should trust it.The article's caveat is that standards are still emerging and agents are more active than static domains. Cerf says the period may be both fascinating and exasperating because the functionality is powerful and interoperability is unresolved. He compares the adoption problem to TCP/IP: competing systems may not work together until users push for functional interoperation. He also says an agentic economy is not inevitable, but that people will try to build it because delegating work to agents will be easier.Read more: TechCrunchxai-org/grok-build, now open sourceAuthor: Simon Willison Published: July 15, 2026Simon Willison argues that xAI's decision to open-source Grok Build is best understood as a trust repair move after a severe privacy failure. The CLI had triggered backlash when users realized that running it in a directory could upload the entire directory to xAI's Google Cloud buckets, including one user's reported SSH keys, password manager database, documents, photos, and videos. xAI disabled the feature, said previously retained coding data would be deleted, and released the code under Apache 2.0.The killer detail is what the codebase reveals. Willison counts 844,530 lines of Rust, only about 3% of which appears vendored, and finds remnants of the upload system still present but disabled: gcs.rs contains Google Cloud upload code, while upload_session_state() now returns a hard-coded session_state_upload_unavailable error. He also notes copied or ported tool implementations from Codex and OpenCode, prompt files, and a terminal Mermaid renderer.The pull is that terminal coding agents are becoming large, intricate software systems in their own right. The privacy failure mattered because these tools operate inside the directories where developers keep their most sensitive work; the open-source release matters because trust now depends on inspecting what an agent can see, send, and do.Read more: SourceThe Pulse: What can we learn from Bun's rapid Rust rewrite with AI?Author: Gergely Orosz and Ivan Klaric Published: July 16, 2026Gergely Orosz and Ivan Klaric argue that Bun's AI-assisted rewrite from Zig to Rust is a practical sign of how software engineering changes when models can take on large, bounded migrations with clear feedback loops. The piece does not treat the rewrite as magic: Jarred Sumner first spent hours turning design judgment into a detailed porting guide, then used adversarial review, parallel agents, compiler errors, and tests to force the work toward correctness.The killer detail is the scale. Bun had 535,496 lines of Zig, 1,448 files, and 22 million monthly downloads, making a conventional rewrite a year-long freeze the team could not justify. Using Fable, Sumner split the work across 64 agents, produced about 6,500 commits, and got the migration done in 11 days at an estimated API cost of $165,000.The pull is economic, not theatrical. If a one- or two-year migration can become an 11-day project, AI coding is not just faster autocomplete; it changes which technical debts are worth paying down.Read more: SourceOrphan risks at the frontier of artificial intelligenceAuthor: Andrew Maynard Published: July 16, 2026Andrew Maynard argues that frontier AI safety frameworks are creating “orphan risks”: harms that companies can see, but do not formally own because they are hard to quantify, do not fit catastrophic-risk thresholds, or fall outside audit-friendly compliance machinery. His target is not existing frontier safety work, but the narrowing effect that happens when private companies decide which risks count as governable.The killer detail is Maynard's contrast between measurable model dangers and threats to value. He points to Meta's three-day Galactica collapse, OpenAI's 2023 board crisis, safety-team departures, and wellbeing litigation as examples of risks that damaged trust, culture, legitimacy, or users without fitting cleanly into conventional model-risk categories. The proposed fix is an orphan-risk register: a public record of risks a company considered and chose not to manage, with reasons.The pull is accountability. Frontier developers' internal scoping choices have become a de facto layer of public governance, so the question is no longer only which risks they manage, but which risks they quietly leave outside the frame.Read more: SourceThe Lab of the Future Should Feel Like a Data CenterLatent.Space with Andy Beam and Rafa Gomez-Bombarelli | Latent.Space | July 16, 2026Latent.Space interviews Lila Sciences CTO Andy Beam and chief science officer for physical sciences Rafa Gomez-Bombarelli about the company's attempt to build an AI-run science factory. The post describes Lila's thesis as treating the lab itself as an “infinite token generator”: if internet data drove the first era of AI scaling, experimentally verified scientific data may be the next scarce training source. Lila is trying to produce that data with robotics, lab instruments, orchestration software, and AI models wired into the wet lab.The central analogy is the lab as data center. Instruments are nodes on a graph, a magnetically levitating transport layer moves materials between them, and experiment scheduling looks like a compute queue. Beam says Lila is not simply an automation company, because the point is not just throughput; it is flexibility, generalization, and experiment capture. The post says Lila has built more than 10 trillion experimentally validated “scientific reasoning tokens,” not internet text or biological sequences.The interview ranges across biology, chemistry, drug discovery, materials science, and the limits of automation. It notes that Lila rebuilt one gas-sorption measurement to run roughly 2,500 times faster, claims its general models can transfer priors from small-molecule chemistry to metal-organic frameworks for carbon capture, and describes model-suggested platinum-group-free electrocatalysts that moved from looking boring or wrong to becoming strong performers. The caveats are physical: experiments have runtimes, biology cannot always be accelerated, chains of thought can be unreliable narrators, and reward hacking becomes more dangerous when a model controls a real lab.Read more: Latent.SpaceWhy AMI Labs' Alexandre LeBrun won't call his AI “AGI” or “superintelligence”Kate Park | TechCrunch | July 16, 2026Kate Park interviews AMI Labs CEO Alexandre LeBrun about why Yann LeCun's world-model startup avoids the language of “AGI” and “superintelligence.” LeBrun says the terms are not useful because they lack stable definitions: “We never used the word AGI. And I just noticed that nobody is using it anymore; they switched to superintelligence.” His argument is that the practical frontier is not a label, but whether AI systems can understand and predict real-world states.The article explains the world-model thesis by contrasting language prediction with physical-state prediction. A large language model predicts the next word; a world model predicts the next state, such as what happens when a glass tips over. LeBrun says LLMs remain complementary and efficient for language, but the physical world is where current AI is weak. Robotics is the clearest case: hardware has advanced quickly, but robots are still brittle outside controlled routines because they lack context and situational understanding.AMI is still pre-product, but TechCrunch reports that LeBrun was in Seoul looking for industrial partners, researchers, and global companies. He says world models cannot be built entirely inside a lab because they need access to real environments. That is why South Korea appeals to AMI: robotics, semiconductors, manufacturing, and fast adoption create the kind of hardware-heavy context that software-only AI has barely touched.Read more: TechCrunchKimi K3 Tech Blog: Open Frontier IntelligenceKimi | Kimi | July 16, 2026Kimi introduces Kimi K3 as an open 3T-class frontier model aimed at coding, knowledge work, reasoning, multimodality, and long-context agentic use. The source describes the model as a 2.8T-parameter system built on Kimi Delta Attention and Attention Residuals, with native multimodality and a 1M-token context window. It says Moonshot AI plans to release model weights by July 27.The post presents K3 through benchmark and use-case sections rather than as a general product announcement. It reports results across coding, productivity, agentic, and multimodal evaluations, including DeepSWE, Terminal-Bench 2.1, Program Bench, SWE Marathon, FrontierSWE, PostTrain Bench, OfficeQA Pro, SpreadsheetBench 2, MCP Atlas, AutomationBench, BrowseComp, GDPval-AA v2, AA-Briefcase, MMMU-Pro, MathVision, BabyVision, OmniDocBench, and PerceptionBench. The source says all reported K3 results use maximum reasoning effort with temperature and top-p set to 1.0, and that different benchmark comparisons use KimiCode, Claude Code, or Codex harnesses depending on the test.Kimi's caveats are unusually concrete. The limitations section says K3 was trained in preserved thinking-history mode, so quality may become unstable if an agent harness does not pass historical thinking content correctly or if an ongoing session switches to K3 midstream. It also says K3's emphasis on long-horizon tasks can make it excessively proactive when it encounters minor issues or ambiguous intent, and recommends imposing explicit behavioral constraints for applications that require strict boundaries. The post adds that K3 remains behind Claude Fable 5 and GPT 5.6 Sol in user experience despite being competitive overall.Read moreVenture CapitalThree Years InAuthor: Tomasz Tunguz Published: July 10, 2026Tomasz Tunguz marks Theory Ventures' third anniversary by arguing that AI's central market effect is time compression. In his telling, model release cycles, company revenue milestones, enterprise adoption, and venture categories have all accelerated. Seed, Series A, and Series B still exist as financing labels, but they no longer cleanly describe company maturity when some seed rounds are larger than IPOs and the best AI companies can mature much earlier than prior software companies.The killer detail is the shift from models to inference. Tunguz argues that inference has become the dominant AI market because workloads and buyer preferences are fragmenting: video, batch, local, agentic, and real-time tasks each create different infrastructure needs. He compares this to databases splitting into OLTP, OLAP, vector, and streaming categories, with AI pushing the same specialization into inference infrastructure.The pull is that Theory sees the AI-native venture firm as part of the same pattern. The firm says it has analyzed twice as many investment opportunities with three investors working alongside a nine-person intelligence organization, using agents and research systems to map markets, source companies, and support diligence. The piece is both a market map and a statement about how venture itself is being rebuilt by the technology it funds.Read more: LinkedInVenture Has Rarely Looked More BifurcatedAuthor: Beezer Clarkson Published: July 14, 2026Beezer Clarkson points to PitchBook's Q2 report as evidence that the U.S. venture market has split into two very different realities. AI now accounts for more than 60 percent of all U.S. venture deal value, meaning the headline market can look active and well-funded even while much of the non-AI market is dealing with a much colder liquidity and fundraising environment.The thread uses that split as the setup for Clarkson's latest Origins episode with Alec Litowitz, founder of Magnetar and QStar Capital and one of Citadel's original founding partners. Clarkson says markets like this are periods of genuine uncertainty, not merely ordinary risk, which is why Litowitz's Adaptability Quotient framework is relevant.The embedded clip makes the liquidity point concrete. Litowitz says DPI is “the resolution of uncertainty” because it converts an uncertain investment into actual cash returned to LPs. In his framing, a realized dollar is a real mark, while TVPI remains uncertain until it is realized.The killer detail is the distinction between pricing risk and resolving uncertainty. Litowitz's perspective matters because QStar is a SpaceX investor and Clarkson says the conversation happened just before one of venture's most consequential IPOs. The episode's stated questions are why venture remains a way to gain exposure to innovation, how AI is changing what is investable, why liquidity is ultimately a function of time, and why uncertainty requires a different decision framework from risk.Read more: XThe Best Angel Investors in the US: Who Backs the Most Unicorns, and Who's Active NowAuthor: Ilya Strebulaev Published: July 10, 2026Ilya Strebulaev ranks angels, angel groups, accelerators, and incubators by lifetime U.S. unicorn investments, counting checks written before a company reached unicorn status. The top of the combined list is dominated by organizations: Y Combinator leads with 113 unicorn investments, followed by Plug and Play at 52 and 500 Global at 41. Sand Hill Angels is the highest-ranked angel group at 31.The killer detail is how quickly the list changes below the biggest accelerators. Strebulaev says 271 of the 304 investors in the Top 200 are individuals, or 89%. In the top 100, individuals are 91%. That makes the market underneath the large accelerator counts look much more personal: mostly operators and individual angels writing early checks from their own networks.The pull is the ranking's own caveat. Strebulaev writes that every lifetime leaderboard has a blind spot because many of the unicorns behind those totals were founded a decade or more ago, and some angels have since moved into formal funds, slowed down, or stopped investing. His post therefore separates lifetime performance from recent cohorts, including companies founded in 2015 or later and 2020 or later. For founders or allocators making current decisions, that distinction matters: a career record and a current record are not the same measure.Read more: Ilya StrebulaevAre Prediction Markets Doomed to Fail?Author: Contrary Published: July 16, 2026Contrary argues that prediction markets' current boom depends on whether platforms can prove they are more than regulated gambling with exchange-style branding. Kalshi and Polymarket have reached mass cultural, investor, and regulatory attention, but the article says the underlying idea is old: academic markets, corporate forecasting tools, Intrade, PredictIt, and other predecessors all struggled with the same linked problems of liquidity, legality, and user appeal.The killer detail is the comparison with sportsbooks. Prediction markets present themselves as peer-to-peer, transparent, and non-house-based, but sports contracts reportedly account for more than 90 percent of Kalshi trading, and the article says the platforms keep a much thinner slice of volume than sportsbooks. A market can therefore show sports-betting-scale handle while generating far less revenue.The pull is that the product's hardest problem may be distribution of wins. If a small group of sharp traders captures most profits while casual users lose interest, prediction markets may become valuable data feeds and professional tools before they become durable consumer networks.Read more: SourceRegulationExclusive: The Next Frontier of the Deportation Wars: College CampusesAuthor: Adrian Carrasquillo Published: July 11, 2026Adrian Carrasquillo reports that college campuses are becoming a new front in the fight over immigration enforcement because automatic license plate readers can turn ordinary campus security infrastructure into searchable location data. His thesis is that Flock Safety's camera network, even without direct ICE or DHS contracts, can feed deportation enforcement through local police partnerships and data-sharing practices.The killer detail is the campaign target. The Emergency Campaign to Support Higher Education, working with Schools Drop ICE, is focusing on 75 colleges and universities publicly identified as having Flock contracts. Flock says it has no ICE or DHS contracts, but activists argue the risk comes through local agencies that coordinate with federal authorities and run searches on their behalf.The pull is broader than immigration. Carrasquillo notes that license plate readers have already been abused by officers for stalking, and that Flock's AI search features can identify more than plates, including bumper stickers. A campus safety tool can become a political surveillance system when the data layer is searchable.Read more: The BulwarkThe Supreme Court Broke Independent Agencies. Here's a Way to Slow the Damage.Author: Todd Phillips Published: July 12, 2026Todd Phillips argues that the Supreme Court's decision in Trump v. Slaughter damaged independent agencies by ending for-cause removal protections, but did not leave Congress powerless. The ruling weakens the old model in which commissioners at bodies such as the FTC, NLRB, CPSC, SEC, and CFTC could be insulated from dismissal over policy disagreements. Phillips says the next fight is whether presidents can turn nominally bipartisan commissions into one-party instruments.The killer detail is the procedural fix: quorum rules. Phillips proposes that Congress require bipartisan slates of commissioners to be seated before independent agencies can act. A president

Support & Resources→ Support the show on Patreon→ Bayesian Modeling Course (first 2 lessons free)Our theme music is « Good Bayesian », by Baba Brinkman (feat MC Lars and Mega Ran). Check out his awesome workTakeaways:Q: What is Variational Bayesian Monte Carlo (VBMC) and how is it different from Bayesian optimization?A: VBMC borrows the machinery of Bayesian optimization but aims at a different target. Bayesian optimization fits a Gaussian process surrogate to an expensive function and uses it to hunt for the optimum. VBMC instead treats the log-posterior as the function to model, evaluates it at a few carefully chosen points, and keeps the whole reconstructed shape rather than just its peak. That gives you the full posterior, not a single best-fit value. Where MCMC might need tens of thousands to millions of evaluations, VBMC often reconstructs a good posterior approximation from a few hundred, which matters when each evaluation is slow.Q: When should you reach for PyVBMC, and when is it the wrong tool?A: Two symptoms tell you PyVBMC might help. First, speed: if a single evaluation of your log density takes on the order of a second, running MCMC over tens of thousands of evaluations becomes painful, and PyVBMC's few-hundred-evaluation budget pays off. Second, dimensionality: because it leans on a Gaussian process surrogate, it works well up to roughly 10 to 15 parameters and degrades beyond that. If your model already runs fine in Stan or PyMC, you do not need it. It shines for expensive, low-dimensional models common in science and engineering, where you are modeling a process rather than composing nice distributions.Full takeaways hereChapters:00:18:13 What is Variational Bayesian Monte Carlo (VBMC) and how does it differ from Bayesian optimization?00:30:21 When should you use VBMC versus BADS in practice?00:31:20 What is Bayesian Adaptive Direct Search (BADS) and how does its hybrid optimization strategy work?00:39:18 What are neural processes, and why are transformers a natural neural process architecture?00:45:54 What is the Amortized Conditioning Engine (ACE) and what problem does it unify?00:55:42 What do PriorGuide and the new autoregressive buffer paper solve for amortized inference?01:02:03 How does the new autoregressive buffer speed up predictions in transformer probabilistic models?01:06:11 What is Luigi Acerbi's vision for a foundation model for inference?01:09:26 What is ALINE and how does it add active data acquisition to amortized inference?01:12:43 How does Luigi Acerbi connect LLM agents, Bayesian decision theory, and the nature of intelligence?01:18:44 For a PyMC, Stan, or NumPyro user, where should you start with VBMC, BADS, or BayesFlow?Thank you to my Patrons for making this episode possible!Links from the show here

Not all work happens in writing. Teams that work with photos, videos, and audio need AI that works for them too. This is why, with Dropbox, you can search within multimedia content for key moments and important information—not just text. In this episode, we talk with Appu Shaji and Hicham Badri, two Dropbox machine learning engineers who are part of the team that makes all of this possible. They explain how multimodal search works—from understanding the context of the initial query, to identifying objects and actions in complex scenes—and how they ensure those models work fast, even at Dropbox-scale. ~ ~ ~  Working Smarter is brought to you by Dropbox. Find, organize, and share your work—all in one place—with context-aware AI from Dropbox. You can listen to more episodes of Working Smarter on Apple Podcasts, Spotify, YouTube, Amazon Music, or wherever you get your podcasts. To read more stories and past interviews, visit workingsmarter.ai This show would not be possible without the talented team at Cosmic Standard: producer Ben Montoya, sound engineer Aja Simpson, technical director Jacob Winik, and executive producer Eliza Smith. Special thanks to our illustrator Fanny Luor, marketing consultant Meggan Ellingboe, and editorial support from Catie Keck.  Our theme song was composed by Doug Stuart.  Working Smarter is hosted by Matthew Braga. Thanks for listening!

(0:00) The AI Buildout: Datacenters Bigger Than Cities (Andrew Feldman) (1:50) Reasoning, Inference, and Breaking Moore's Law (16:28) Open Source, AI Sovereignty, and the Road to AGI (40:54) The Innovation Behind Generative Video (Robin Rombach) (47:31) Martin Scorsese, Robots, and the Future of Hollywood IP Thanks to our partners for making this possible! AppLovin Ads - AppLovin's AI advertising platform reaches over a billion daily active users across mobile games. Full-screen video ads with a 35-second median watch time. Advertisers are profitably spending hundreds of thousands of dollars a day and advertiser access is still in closed beta. The window is open at https://applovin.com/ALLIN Nasdaq - Positioned at the nexus of technology and the capital markets, Nasdaq provides premier platforms and services for global capital markets and beyond with unmatched technology, insights and markets expertise. https://www.nasdaq.com/convergence-economy Follow Andrew: https://x.com/andrewdfeldman Follow Robin: https://x.com/robrombach Follow the besties: https://x.com/chamath https://x.com/Jason https://x.com/DavidSacks https://x.com/friedberg Follow on X: https://x.com/theallinpod Follow on Instagram: https://www.instagram.com/theallinpod Follow on TikTok: https://www.tiktok.com/@allin Follow on LinkedIn: https://www.linkedin.com/company/allinpod Intro Music Credit: https://rb.gy/tppkzl https://x.com/yung_spielburg

We've been running a bit of an Agent Cloud series surveying all the top inference/compute/cloud providers, from Databricks to Daytona to Railway and, even further back, E2B, but we're excited to conclude this series returning to Modal, which has just raised a monster $355M Series C.The cloud was built for developers. But agents are now changing that.The old infra stack was designed for a human who could read docs, reason through YAML, and understand dashboards to figure out what they need when something broke. While this was painful for developers, it worked since they could fill in missing context in their heads.However, agents don't have that luxury. Now in this new era of agents, everything has to be tighter.They need a place to write code, run it, inspect the output, change the environment, debug failures, and try again. Fast iteration and feedback loops with all the necessary context are crucial for agents to operate properly. Furthermore, sandboxes are a clear representation of this shift as agents can easily spin up isolated environments. This programmatic infra even extends to research:Two years ago, we were one of the first to cover Modal with CEO Erik Bernhardsson and Alessio designed our favorite LS thumbnail of all time:At the time, Modal was just a teeny little company with a $17M Series A.Today, fresh off their $355M Series C, Modal is one of the clearest examples of the agent cloud future being built in real time: a cloud platform moving past traditional web app assumptions toward the workloads AI actually creates such as elastic inference, sandboxes, GPU burst, post-training, background agents, and infrastructure that agents themselves can operate.In this episode, Modal CTO Akshat Bubna joins swyx and Vibhu to unpack why AI applications don't fit traditional cloud assumptions, why Kubernetes was never designed for bursty compute-heavy workloads, and why Modal is now shifting from developer experience to agent experience.We go deep on Modal's AI infra stack: serverless functions, decorator-based infrastructure, elastic inference for custom models, GPU snapshotting, DeFlash, speculative decoding, Auto Endpoints, sandboxes, persistent storage, networked containers, private IPv6, RDMA, multi-node training, and Modal's capacity pool across 17 cloud providers. Akshat also explains why RL rollouts can require 100,000 sandboxes, why production agents need hard guardrails, why observability may matter more than reading code, and why AI has made infrastructure exciting again.We discuss:* Why Kubernetes wasn't built for bursty AI workloads* How Modal started as a better runtime before becoming an AI cloud* Why Modal added GPUs before ChatGPT* The shift from developer experience to agent experience* Why observability matters when agents are writing the code* Elastic inference for custom models across audio, video, robotics, and comp bio* GPU snapshotting, cold starts, and why inference workloads are so bursty* Why RL rollouts can require 100,000 sandboxes* DeFlash, speculative decoding, and frontier-level inference performance* Auto Endpoints and making optimized inference easier to deploy* What Modal adds beyond vLLM, SGLang, and raw GPU rental* Modal's 17-cloud capacity pool and supercloud strategy* Networked sandboxes, sidecars, private IPv6, and RDMA* Serverless multi-node training for post-training and research workloads* Auto-research, model-guided sweeps, and agents launching GPU experiments* Compute strategy, capacity planning, and batch tiers* Why production agents need specialized sandboxes and hard guardrails* Modal's take on managed agents, CI, Gitpod/Ona, Python, TypeScript, and Modal BenchAkshat Bubna* LinkedIn: https://www.linkedin.com/in/akshat-bubna-188885103* X: https://x.com/akshat_bModal* Website: https://modal.comTimestamps00:00:00 Introduction00:00:39 Modal's origin and why Kubernetes wasn't enough00:04:32 Developer Experience → Agent Experience00:06:21 Modal's AI cloud primitives00:09:14 Sandboxes, agent loops, and proto-Cognition00:12:12 Elastic inference, GPU snapshotting, and 100,000 sandboxes00:15:24 DeFlash, speculative decoding, and Auto Endpoints00:19:59 Production-grade inference beyond raw GPUs00:22:00 Background agents, Ramp Inspect, and the agent lifecycle00:24:08 Modal's 17-cloud supercloud strategy00:26:40 Networked sandboxes, private IPv6, and RDMA00:32:48 Multi-node training, post-training, and auto research00:37:36 Compute strategy, capacity planning, and batch tiers00:40:55 Open models, real-time AI, and production agent infra00:43:06 Hard guardrails, managed agents, and specialized sandboxes00:46:06 Why AI made infrastructure exciting again00:48:30 Model APIs, differentiated products, and agentic video00:51:50 CI, coding-agent infra, SDKs, and Modal Bench00:57:28 Closing ThoughtsTranscriptIntroduction: Modal, Series C, and the Art PartySwyx [00:00:00]: We're here with Akshat, CTO of Modal, together with Vibhu. Congrats on your Series C.Akshat [00:00:10]: Thank you.Swyx [00:00:11]: Your party yesterday was amazing.Akshat [00:00:15]: Yeah.Swyx [00:00:15]: From all the photos and all the swag.Akshat [00:00:17]: We had a bunch of art installations, which was fun, seeing, like, our products on pedestals next to, like, Rodin.Swyx [00:00:25]: Very nice. Very nice. When you started, it was not the GPU inference company. Maybe it was in your mind. Take us back to the origin story.Modal's Origin: A New Runtime Beyond KubernetesAkshat [00:00:39]: I first met Eric, who's the CEO, through an investor. Back then Eric was already thinking about building, a new runtime, and he got there thinking through why are workflow orchestration products so hard to use. It's because you have to run them on Kubernetes. Kubernetes is hard to manage. It's not built for burstiness and, custom images,Swyx [00:01:03]: YeahAkshat [00:01:03]: It has a terrible developer experience.Swyx [00:01:05]: And I'll, I'll interjectAkshat [00:01:06]: YeahSwyx [00:01:07]: For listeners, who are new, we interviewed Eric two years ago, and there's a bit more of the story there from Spotify and all those things.Swyx [00:01:14]: And I came across Eric through Data Council because he did that talk on the serverless container stack that you guys did, which was like, that was my first like, “Okay, I need to take Modal very seriously” moment.Akshat [00:01:26]: Yeah.Swyx [00:01:26]: But it was still very unclear, like, do I need all this for just my data pipelines?Akshat [00:01:33]: Yeah. initially what we were thinking about was if we build a better runtime, it's a very useful primitive in itself. It's There's a lot of things that, get solved by serverless functions, like you can do, ETL stuff, you can do job queues, you can do all this, like, bursty processing, which it turns out every company had needs for. but then we also were thinking about this as like, this is a primitive that we can build a whole collection of products on, which are very verticalized. So perhaps data engineering would've been the first one, but we were thinking about inference. Back then it was more classical inference, like computer vision stuff and running XGBoosts and whatnot. But we added GPUs to the product a year before ChatGPT came out.From Serverless Containers to GPU WorkloadsSwyx [00:02:19]: Nice.Akshat [00:02:19]: We just didn't think it would be that big of a deal.Swyx [00:02:22]: Yeah, just like add A100.Vibhu [00:02:23]: Was there any, like, early key problem that really sparked off why you built it?Akshat [00:02:28]: Yeah. Primarily it's just, none of the tooling that was out there was built for, one, a really great developer experience, and also there's a general trend of, a lot of the workloads that we were seeing were very. I wish there was a better word for it, but compute-heavy. Like, they need, one, like, need a lot more resources, so you need to burst up and down a lot, versus like Kubernetes designed for, like, slow scaling and, more for, like, web server use cases. And also there's just a lot more specialization in, like, what kinds of environments these workloads run in. Like, we had sometimes they need accelerators, sometimes they need different kinds of images, and this is just like a consistent thing that we saw across a lot of companies. That would be the next step.Software-Defined Infrastructure and Decorator-Based DXSwyx [00:03:13]: Yeah. Yeah. Be nice. I don't know how much this factored into the early story, but I wrote a post when I was at Temporal about infrastructure, software-defined infrastructure or something like that.Akshat [00:03:22]: Yeah, the self-provisioningSwyx [00:03:23]: Self-provisioning.Akshat [00:03:24]: Yeah.Swyx [00:03:24]: Yeah. I can't even remember my own post.Swyx [00:03:26]: And then you put me on the landing page.Akshat [00:03:28]: Yeah. We really like, the term and so we stole it.Swyx [00:03:32]: Because you had the insight that everything can just be in decorators co-located with the code, right?Akshat [00:03:37]: Yeah.Swyx [00:03:37]: Was that a big part of the originalAkshat [00:03:39]: YesSwyx [00:03:39]: Story or it was just like a DX layer?Akshat [00:03:41]: That was, really important because we really didn't want people to spend, so much time, writing YAML, and it seemed like you could really condense the surface area of what you're doing, put it in code so you can operate on it just like you operate on other code, and like build stuff that's more expressive and dynamic. and so yeah, that was always a very important part.Swyx [00:04:04]: Then the pushback is this is a DSL.Akshat [00:04:07]: Yeah.Swyx [00:04:07]: It's you're closed source. I am locked into Modal.Akshat [00:04:11]: Yeah. We never really got pushback for that because the nice thing about Modal is you can bring whatever code you have, and sure, the DSL is at the configuration layer for, what hardware you're using, how you're scaling things up, but you still own the code.Akshat [00:04:27]: And that's, that's been an important, part of our story, even as we do inference now.Swyx [00:04:32]: Yeah.Vibhu [00:04:32]: How much of do you think still stays the same today? Like if you were to build something today, DevX very important, but I feel like, a lot of this has been changed with just hook it up to an agent, have Claude Code, have Codex implement a tool. there's very agent native primitives that are different than if I'm doing this myself, right?Developer Experience → Agent ExperienceAkshat [00:04:54]: We've changed our SDK team to think about agent experience instead of, developer experience and we think that the same benefits that apply for DX also apply for AX, which is why would you have an agent read through hundreds of Kubernetes files and like write YAML that's not even typed when it can make a couple of changes in a decorator and it gets this self-provisioning runtime of, being able to see its changes live in action? yeah, it just seems from the customers we talk to, they find Modal is much faster for agents to use versus operating on a different substrate.Swyx [00:05:34]: Yeah, because like you, again, you co-locate the infrastructure requirements to the code that runs it.Akshat [00:05:38]: Yeah.Swyx [00:05:38]: Well, the negative thesis now is that nobody's looking at their code anymore, so there's no point.Akshat [00:05:44]: Yeah, people aren't looking at code. one thing we still see is really important is observability.Swyx [00:05:51]: Yeah.Akshat [00:05:51]: Like how good is your dashboard? And of course, like we have, we push a lot of it to the CLI so the agents can do their own investigation, but you still need humans to go interpret what's going on and, make judgment calls and whatnot. and that's I feel like, Maybe more important now than looking at the code itself.Swyx [00:06:11]: Yes, because like, you can try to treat the code as a black box and then use, see the observable action that comes out of it, and then just prompt a change.What Modal Is For: AI Cloud PrimitivesAkshat [00:06:21]: Yeah.Swyx [00:06:22]: So I think it takes a bit of restraint to not specialize, to say, “I want to ship a new primitive,” and then just be general purpose.Swyx [00:06:31]: People ask you, “What are you for?” You're like, “ I don't know. We can do this, we can do that.”Vibhu [00:06:36]: Well, I'd be curious to see, like, okay, if we were to ask you, like, what is Modal for even at a high level? There's a lot you guys do, sandboxes, GPUs, everything. How do you answer?Akshat [00:06:46]: Modal is a cloud platform that's built for, where we've built the primitives from scratch for AI applications. and right now it covers, inference, training, batch processing, and sandbox workloads.Akshat [00:07:00]: But we're building a lot moreSwyx [00:07:02]: I noticed you didn't say web server, so there is still a role for, like, the always-on large-scale Kubernetes type things.Akshat [00:07:09]: Yeah, absolutely. We're, we're not trying to compete with the renders of the world, because yeah, we think the differentiator for us is the, are the workloads that need specialized compute, need to scale up and down a lot. yeah, they're, they're, they're just shaped differently.Working Alongside Frontier StartupsVibhu [00:07:26]: I think you're building a lot of it alongside the startups, right? They're innovating quite a bit, even in your, like, latest blog post. Like, even in the series C, the customers that you mention here, the cognitions, technical ones, ramps and whatnot, they're, they're innovating with you, right? And that's not something AWS is doing directly with.Akshat [00:07:45]: Yeah, absolutely. I think, this is again classic. We're a small team. We can move really fast. our engineers are working with our customers and figuring it out. Yeah.Swyx [00:07:54]: So my first week at Cognition, I walked in, there was someone wearing a Modal shirt. I was like, “What are you doing here?” They're like, “Yeah, I just. I am embedded inside of Cog.”Akshat [00:08:05]: Yeah, I think that was Peyton. We sent him overSwyx [00:08:07]: Yeah.Akshat [00:08:07]: Because, the latency of communication was too high otherwise.Swyx [00:08:12]: Yeah, distributed node, you have to - you have to place one and collocate.Vibhu [00:08:16]: Yeah.Swyx [00:08:16]: So I had a, I had direct personal experience, right? So I worked on smol developer three years ago. it was inspired by Claude 1. I think you onboarded me at some point, like, just before, and I was like, “Oh, like, I need some bursty compute. Like, I was just gonna try using Modal.” And it was a, it was a pretty pleasant experience. apparently, I showed up in the board meeting, like the analytics.smol developer, Sandboxes, and Proto-CognitionAkshat [00:08:39]: Yeah, you blew up on Hacker News and,Swyx [00:08:41]: YeahAkshat [00:08:41]: We got a big traffic spike. I. I think the way you used smol developer was Modal functions for running stuff, which was. Like, the, that was a good use case. but then, yeah.Swyx [00:08:53]: Yeah. That - So to me, that was proto-cognition.Akshat [00:08:55]: Right.Swyx [00:08:56]: If only I had, like, stuck to it.Swyx [00:08:58]: Like, that was like, if - did you say draw the tech treeAkshat [00:09:00]: AbsolutelySwyx [00:09:00]: You're just like, “Yeah, like, probably this will happen.”Akshat [00:09:02]: Yeah. Like, he was so close. You were just rebuilding upon usSwyx [00:09:04]: I just didn't realize.Akshat [00:09:05]: But the funny story there is at the same time, we were talking to a bunch of customers who needed something like sandboxing.Swyx [00:09:14]: Yeah.Akshat [00:09:14]: This is like twenty-three.Swyx [00:09:15]: Yeah.Akshat [00:09:16]: So we builtSwyx [00:09:17]: You introduced a new API right after that.Akshat [00:09:18]: Yeah.Swyx [00:09:19]: Yes.Akshat [00:09:19]: Like, we built sandboxes in May of twenty-three before anyone was even knew this was gonna be a thing. And the first example we published was, we took smol developerSwyx [00:09:28]: Smol developerAkshat [00:09:28]: And put it in a loop, so the agent can iterate on itself.Swyx [00:09:33]: Loops are hot these days.Vibhu [00:09:34]: It's the looper.Akshat [00:09:34]: Yeah.Vibhu [00:09:35]: Loops in. When was this, twenty-three?Akshat [00:09:38]: Yeah.Vibhu [00:09:39]: A small check.Akshat [00:09:39]: Yeah.Swyx [00:09:39]: It's like twenty-three. so the. the, those for listeners, like, the problem was the models are not built for any of this, right?Swyx [00:09:46]: Like, you're just trying to like. They're not post-training to understand, like, looping and, like, self-correction and tool calling was there, but, like, also not that great.Akshat [00:09:55]: Yeah.Akshat [00:09:55]: I don't remember if you used tool calling in this one, but yeah, the models would just diverge after like ten iterations and not produce anything meaningful.Swyx [00:10:03]: Yeah. But like, then. So okay, like now talking to myself three years ago, the answerVibhu [00:10:08]: Of course they will get betterSwyx [00:10:09]: Collect all the failures, build benchmark, and then collect all the, examples, build the RL environmentAkshat [00:10:15]: RightSwyx [00:10:15]: Sell it for like ten billion dollars to Meta.Swyx [00:10:17]: And then also train a model and then sell that for sixty billion dollars to Elon. And this isAkshat [00:10:23]: Yeah, of courseSwyx [00:10:23]: The funny machine. Like, it's like, it's about the hardware.Akshat [00:10:28]: It's hard to have that inherent conviction that the stuff will get that much better.Swyx [00:10:33]: In retrospect, it's so f*****g obvious.Akshat [00:10:36]: Fair enough.Swyx [00:10:37]: Like, what else were we doing back then? I don't know. anyway. Yeah. So this. That was the start of your sandboxing journey, right? I feel like it didn't blow up until, like, last year.Akshat [00:10:49]: Yeah.Swyx [00:10:50]: So there was like a couple years of quietness.Akshat [00:10:52]: Exactly, yeah. We wereVibhu [00:10:53]: I think very underrated product value. Like, my experience with Modal, Charles, before he had joined Modal, met this guy at a hackathon, and he really insisted we wanted to run some small model, not hosted anywhere, and he's like, “ there's this cool company, Modal. They'll like spin up a GPU sandbox, we can throw it on there. They'll take a Hugging Face link.” And like there's so much value just right there, right? Like instant hosting, spin it up, spin it down. It'll stay cold, but we run the demo a few days later, it'll come back up and like all this stuff in retrospect, like it's still what we needed like today.Akshat [00:11:27]: Yeah, it's still needed today. workload shapes have changed a lot as, we run stuff for people with really massive production scale and, there it's it's not about scaling from zero to one, but it's how do we scale really elastically, from like thousand to fifteen hundred GPUs very quickly in a given region. It's the same shape problem.Elastic Inference, GPU Autoscaling, and Custom ModelsVibhu [00:11:50]: Okay. So you look at, say, Cursor Composer, right?Akshat [00:11:53]: Yeah.Vibhu [00:11:53]: They had a. “We'll do RL on a model every couple hours.” you guys have a whole version of RL inference gym and whatnot.Vibhu [00:12:01]: When you look at workloads like that, you're doing train runs where you need to scale up, scale down every hour thousands of GPUs, right? That's the example for we do need it, right?Akshat [00:12:12]: Yeah. Well, so I'll, I'll take a step back and, maybe talk about like how people use Modal today. because our biggest use case is, elastic inference. And the thing we first found product market fit, with was inference for custom models. So we stayed away from the LLM space, and we were serving companies like Suno for audio, Runway for video, robotics, comp bio companies that train their own model elsewhere. But Modal is the best black box that for deployment, scaling to however many GPUs you need as your traffic pattern changes. And we saw all of them like have a very unpredict- predict- predictable, traffic pattern. it's like diurnal. It's Some days, like the company will do a launch and, they'll need like, way more. And it's not just one model that they deploy. They-- all these companies deploy, lots of different models in different regions, and so the autoscaling problem becomes even harder because then you have to scale within a certain region, and those cycles are offset. So different times you scale up in different regions.Akshat [00:13:20]: So that's like our sortVibhu [00:13:22]: And thatAkshat [00:13:22]: YeahVibhu [00:13:22]: That in and of itself is a huge category. There's a bunch of inference providers which, provide this fireworks, does this as a service together, whatnot, Base10. that's carved into its own niche for language models, at least right now.Akshat [00:13:36]: Yeah. the thing that we have specialized in is the autoscaling aspect.Vibhu [00:13:41]: Yeah.Akshat [00:13:41]: Because we found that it's not universally true that everyone else can autoscale, and we've gone deeper into it on the tech side by, we've incorporated GPU snapshotting into the product so we can take the GPU state, like your torch.compile model, snapshot it, and the next cold start is way faster. And so going back to your question, it's That's why you need a lot of burstiness for inference. But then people also do a lot of demand training, like for RL stuff, your rollouts are bursty, as you said. People also do a lot of batch jobs. So we'll see, a lot of companies, before they have a training run, they'll need thousands of GPUs to run encoding or something like that. And I think those things are much more bursty than. I agree that agents are not that bursty. sandboxes are, except when you're doing RL. RL is justRL, Batch Jobs, and 100,000 SandboxesVibhu [00:14:28]: Or commerceAkshat [00:14:28]: Insanely bursty.Vibhu [00:14:29]: Yeah.Akshat [00:14:30]: Yeah. Like when you're doing, rollouts, you sometimes need a hundred thousand sandboxes in your sandboxes.Vibhu [00:14:37]: Yeah. I'm curious if you've seen early sparks of continual learning. There are some people, like our friends, ngram, recently announced thisAkshat [00:14:45]: YeahVibhu [00:14:45]: They're, they're trying to do training. That also seems like a different workload, right? If you're doing training twenty-four/seven per se, there's a very weird dynamic of how you're using GPUs between people and whatnot, but seems like something you guys would work for.Akshat [00:15:00]: As you said, we're, we're fortunate to work with a number of, customers at the frontier and grab some of our customers. and they are taking the primitives we have, and trying to use them in very interesting ways, like continual learning. It's possible as the stuff gets better, some of that will be part of, our offering as well if, more people need it. but we're, we're just waiting to seeVibhu [00:15:23]: YeahAkshat [00:15:23]: How it shakes out.Vibhu [00:15:24]: Is there a primitive that you added after sandboxing that was the next step in the story?LLM Inference, DeFlash, and Speculative DecodingAkshat [00:15:32]: I guess we've been going much deeper into LLM inferenceVibhu [00:15:35]: YeahAkshat [00:15:35]: Because we realized that some of the advantages we have with like autoscaling, again, especially in different regions and whatnot, are, not present elsewhere. and the place where we had a gap was we weren't, working on the model layer itself. Like we were a black box. And, we realized that, we can get to frontier-level model performance, with, by having great people who work on this. And, we've been open sourcing a lot of our work, in terms of, Recently, we, shared our work on DeFlash, which is a block-based, speculator, and we've open sourced, all of it. So, you can - By using open source DeFlash, you can get the same performance as you would with one of the proprietary providers. And the next thing we're thinking about hereVibhu [00:16:23]: I thought this wasAkshat [00:16:24]: YeahVibhu [00:16:24]: An interesting blog post as well, right? Like, I think in here you make a claim that. Not a claim, just that how effective speculative deco-decoding really just get to.Akshat [00:16:33]: Yeah.Vibhu [00:16:33]: Anything you wanna point out from this around, what people should know?Akshat [00:16:39]: Yeah, absolutely. the high-level summary is, it would help to describe what speculative decoding is.Vibhu [00:16:44]: Yes.Akshat [00:16:44]: I will, yes.Vibhu [00:16:45]: I think, likeAkshat [00:16:46]: YeahVibhu [00:16:46]: So we've covered like Eagle and all thisAkshat [00:16:47]: YeahVibhu [00:16:47]: Like Hydra and all those things, but it was like two years ago.Akshat [00:16:51]: Yeah.Vibhu [00:16:51]: I think it doesn't hurt, right?Akshat [00:16:52]: Yeah. Speculative decoding is you have a smaller model, called a draft model, predict tokens ahead of the bigger model, and then you have the bigger model, verify all of this, all the tokens are predicted. And the reason it's faster is if you're predicting, one token at once, you're bound by memory bandwidth. But if you can batch the verification of, the draft model, then you're much more efficient using compute, and it's faster, and as long as your draft model is producing a lot of tokens that can get accepted, which is called the accept length, you can get a speed up that's, multiple times of, the original model speed. and well, that's what we highlight here. It's Like people talk a lot about we made these kernels faster and whatnot, but improving kernel will only give you like few percentage points of improvement, and, increasing accept length, literally is a multiplicative decreaseVibhu [00:17:47]: Like two to four X.Akshat [00:17:48]: Yeah, exactly.Vibhu [00:17:48]: Without much head-on performance.Akshat [00:17:50]: Yeah. I think it may - you are running a second model, right? So it may be something more expensive in the compute,Vibhu [00:17:57]: I meant quality performanceAkshat [00:17:58]: Probably not by muchVibhu [00:17:58]: But yeah. I thinkAkshat [00:17:59]: So there's no drop in quality performanceVibhu [00:18:01]: YeahAkshat [00:18:01]: Because you're always. You're never accepting a token that the big modelVibhu [00:18:04]: It's strictly betterAkshat [00:18:05]: YeahVibhu [00:18:05]: Or it's same.Akshat [00:18:06]: Exactly.Vibhu [00:18:07]: Right. Yeah.Akshat [00:18:08]: And so we've been working a bunch on DeFlash, which is a block-based speculator. so it's instead of predicting, one token at a time, it's predicting a block. And we've been open sourcing our work with it. The next thing for us here is for helping people train speculators and custom models. it's it's something that traditionally is very forward-deployed engineering driven, support deployed, engineer driven, like you work with customers and help them do that. And our vision for. This is why we launched Auto Endpoints, is we want to make frontier-level performance available to everyone. And so, we mentioned this in the announcement, we teased it. The next thing we're, we're launching is, as you run an auto endpoint, we shadow trafficAuto Endpoints and Frontier-Level PerformanceVibhu [00:18:54]: Do you want to explain what auto endpoints are?Akshat [00:18:57]: Yeah.Vibhu [00:18:57]: I lovely, yeah.Akshat [00:18:58]: Yeah. So, this is, I guess, going back to your Modal is you touch the code, but, sometimes people don't wanna touch the code, and they wanna get started with an endpoint that works and has all the great performance and, scalability that Modal has. So we've made that easier with, a way to create an endpoint from our UI, from the CLI, that has all of our optimizations that we talked about, like the DeFlash stuff already baked in, and there's full transparency. So we give you the code, you can go run it yourself, and if you want, you can eject out into the full Modal experience, which we see as people get sophisticated, they do wanna tweak the models, they wanna, fine-tune stuff. You can still do all of that. It's it's not a black box. And yeah, the next thing, as we teased later in the post, is how do we give you value even beyond this in terms of having your draft models evolve as your data distribution evolves, again, without having to talk to a person and, yeah.Vibhu [00:19:59]: I guess just to understand it directly, you have the GPUs, you have an endpoint that's compatible, you serve open model. If someone was to do this themselves, what's the delta that you guys provide? So you do a lot of open source great work on effective inference. how does it compare to, say, I take the same model, 5.2 FP8, take shelf inference engine, vLLM, SGLang, get compute of similar capacity, similar cost. What's the delta that plugging into something this, like this offers outside of the benefit of, scaling?Production Inference Beyond Raw GPUsAkshat [00:20:34]: It's interesting because we've taken the approach of open sourcing our contributions and upstreaming them. we work closely with the SGLang team. We want the improvements that our team, comes up with to be, there in open source for others to use, even outside of Modal. The benefit to us is we have a team that has significant expertise in terms of if you do have something that is not there, our team can help you get that performance, first. the other thing is with these endpoints, we are way more elastic, as you said, than, anyone else, and you have true scaling to zero. you have true, burstiness, and in practice, that matters a lot more to people than just finding, the GPU and, running Modal code on something.Vibhu [00:21:20]: Yeah. And I will say it's not that straightforward to just. like what I said is easier said than done, right?Akshat [00:21:26]: Yeah.Vibhu [00:21:27]: It's I think still for the average person, still hard to just gut check using different. There's, there's quite a bit of combinations you can make there. the trade-offs aren't really known at face value.Akshat [00:21:40]: Yeah. it's it's not just that. I think it's it's that running production-grade inference is a hard infer problem.Vibhu [00:21:49]: YeahAkshat [00:21:49]: Even if you subtract out the autoscalingVibhu [00:21:50]: YeahAkshat [00:21:51]: Is controlling things like tail latency and, making sure every, request is delivered at least once and whatnot.The Model and Agent LifecycleVibhu [00:22:00]: There's a lot of innovation that you can do here. I think, it's very interesting that you're starting to encroach on, like as you become a full cloud, you're starting to encroach on other people's turf.Vibhu [00:22:09]: What will you not do?Akshat [00:22:13]: Well, we wanna follow our users and, make sure they get like a platform that has everything that works well together. so right now we're focused on the model lifecycle and the agent, lifecycle. so both like going from data prep to training to inference, and then also if I want to deploy a background agent, let's say, sandbox, do persistent storage, a whole bunch of other stuff.Vibhu [00:22:38]: We talked to Cole, who did, OpenInspect. Yeah.Akshat [00:22:42]: Yeah.Vibhu [00:22:42]: And RealInspect also is on Modal.Akshat [00:22:44]: Yeah. So Ramp Inspect was a great example of a background agent that was really successful because they, were able to use some of the primitives like snapshotting and fast scaling to just have something that feels really reactive and works well.Ramp Inspect and Background AgentsVibhu [00:23:02]: Yeah. That's the new CTO of, Ramp right there.Akshat [00:23:05]: Yeah, Rahul.Vibhu [00:23:08]: It was really fun. yeah, okay, I think, all very bullish. Like, one of my reflections was also I did not originally. So when I met you guysThe Inference Inflection: CPU, GPU, and Co-LocationVibhu [00:23:19]: You weren't that much in the GPU game, and now you're all about, inference. And one of the points that I hinged on for Jensen's keynote at GTC this year was, what we're calling like the inference inflection, right? That let's say in AI workloads or machine learning workloads, it used to be like, let's call it eight to one GPU to CPU, and now it's more like one to one, which is like a interesting. Like, - because of how much agents are blocked or call out to this, to CPU heavy stuff the actual, like, limiting factor, like, swings back and forth from GPU to CPU a lot more than it used to be all GPU and then occasional CPU.Akshat [00:24:01]: Yeah.Vibhu [00:24:02]: GPU, CPU. And now it's like just constantly, and you just have to locate everything.Seventeen Clouds and the Supercloud StrategyAkshat [00:24:08]: Yeah. And that's one of the things that, again, we see as, something appealing about Modal, which is we've built this capacity pool that spans, 17 cloud providers, so we're, we're very good at Running on various kinds of cloud capacity across the worldSwyx [00:24:24]: You don't have your own data centers?Akshat [00:24:25]: We don't have our own data centers. We just run across a lot of neo cloudsSwyx [00:24:29]: Yeah. AreAkshat [00:24:30]: Metal providers.Swyx [00:24:30]: Yeah. Question mark.Swyx [00:24:31]: Yeah. You're, you're running the math, and you're like, “What's the cutover point where you're like.”Akshat [00:24:36]: Yeah, it's a good question. part of it is we see our differentiator in the software layer, and, being capital light and focusing on the software helps us move really fast. so far it's worked out well because there are so many other people building data centers that we're able to work effectively with them, and again, focus on what makes us, special.Swyx [00:24:55]: Yeah.Swyx [00:24:56]: 17 gets you into, like, the local providers sometimes. LikeAkshat [00:25:00]: The,Swyx [00:25:01]: Which was the most interesting one?Akshat [00:25:02]: There are a lot more neo clouds than you expect, and they all have various degrees of, various levels of reliability. And, that's why it's something we've invested a lot of time in, is building our own reliability layer on top. so if the GPU falls off the bus or something happens, we user workloads are not affected, and that lets us use a lot more capacity than,Swyx [00:25:30]: YeahAkshat [00:25:30]: You as a user would be able to.Swyx [00:25:32]: It's a useful thing to have because like now everyone knows, like, what layer you are and, like, you optimize for being the super cloud of all clouds.Akshat [00:25:41]: Yeah. That's, that's, that's the idea. and so I guess when you mentioned colocation, that's, that's another interesting thing where, one thing we've seen is people come to us when they want, very specifically located, CPUs or GPUs, like they wantSwyx [00:25:57]: Oh, they pin it in likeAkshat [00:25:58]: YeahSwyx [00:25:58]: EU?Akshat [00:25:59]: Exactly. Or EU, US.Swyx [00:26:01]: Right. Data resiliencyAkshat [00:26:02]: AustraliaSwyx [00:26:02]: Locality thing or performance or what?Akshat [00:26:04]: It's either data locality or latency, yeah.Swyx [00:26:07]: Yeah.Akshat [00:26:07]: Like, you want your. They're running sandboxes and model. They want them to be right next to aSwyx [00:26:10]: Yeah, it's easy thenAkshat [00:26:11]: YeahSwyx [00:26:12]: To. That is important in all those things. and so, like, you've accidentally, I don't know if it's accident, but, like, you've built the perfect primitive for agents to express themselves. And then, like, it's almost very funny how every extra development just involves more file system, just involves more CPU.Akshat [00:26:30]: Yeah.Swyx [00:26:31]: Just like the things that you already have. I don't know much about, if there's any, like, networking usages that are interesting, but you've also done some good work on networking.Networking, Sidecars, Private IPv6, and SandboxesAkshat [00:26:40]: Yeah, that's exactly right. Like, we're just taking compute storage and networking and building stuff on that layer, for, again, the stuff people need.Swyx [00:26:49]: YeahAkshat [00:26:50]: We see a few interesting networking things coming up. one is people want networked sandboxes. so we haveSwyx [00:26:57]: For like a Docker cluster type thing.Akshat [00:26:59]: Yeah.Swyx [00:26:59]: Sorry, Docker Swarm. Oh, f**k. What is it called?Akshat [00:27:02]: Compose.Swyx [00:27:03]: Compose type thing.Akshat [00:27:04]: Yeah. So if you want Docker Compose, our sandboxes now support, this thing called sidecars. So you can. A sandbox is a pod of containers, and you can run multiple containers in, a sandbox. also useful because, going back to networking, people want a lot of control over, outbound networking from a sandbox.Swyx [00:27:23]: Yeah.Akshat [00:27:23]: Like, they might wanna run a middle proxy for, like, maybe logging stuff for RL or, controlling how egress can happen to a domain, injecting credentials. and yeah. So we've, we've had to build a lot of that stuff ourselves.Swyx [00:27:38]: Yeah.Akshat [00:27:39]: But then also sometimes people want, sandboxes spanning multiple nodes to talk to each other, which is an emerging thing we're seeing. We have support for that for a different reason, and yeah, we'll see if that becomes stable.Swyx [00:27:52]: Like, just an open socket. It's a. This is directly like mTLS.Akshat [00:27:56]: We do support that, which is you can, expose a tunnel inside a sandbox.Swyx [00:28:01]: Yeah.Akshat [00:28:01]: And then you can either expose it to public internet or it can be, you can add like a HTTP, auth layer above it. But we have this thing called I6PN, which we haven't talked about, which is this, like, overlay network using IPv6 addresses. so if Modal containers, within the same workspace, when this is enabled, can address each other using this private IPv6 address, and no one else can.Akshat [00:28:28]: So it's like private networking, for containers. We built it because we needed it as a primitive for our distributed training product. so we have this other feature, which is you can add a decorator to a function, and you get a cluster of GPUs. and they have RDMA networking. so you can run a distributed training job, that's truly serverless. and we did the overlay network for that. But then we've seen that people are using it for other reasons, and, I'm intrigued to yeah, what would people do with it.Swyx [00:28:59]: Build primitives and let people figure it out, right?Akshat [00:29:01]: Yeah, exactly.Swyx [00:29:02]: You put out a pretty interestingAkshat [00:29:03]: They're like, they read the docs webpage. Let me use thatSwyx [00:29:06]: YeahAkshat [00:29:06]: Something they never intended to work. This is literally not even in our docs page. People somehow found it, and they're using it.RDMA, Memory Movement, and Distributed TrainingSwyx [00:29:12]: Huh.Swyx [00:29:14]: The way you portrayed it with, like, RDMA versus TCP, like, very well laid out, but just the transfer speed change at scale for RL, like yeah, you have it, you have it built in. I'm sure someone found it. It's found it to be a lot more efficient before you made a thing out of it, right?Akshat [00:29:32]: Yeah. And not to split hairs, I guess the overlay network is the TCP overlay network.Akshat [00:29:39]: The reason we have that is you need that to do the key exchange for RDMA before you set up the RDMA network on top of that. but then people found the TCP part.Swyx [00:29:48]: Can I tell you, this is like a big aha moment for me becauseAkshat [00:29:51]: YeahSwyx [00:29:51]: So I review 2,200 submissions for the World's Fair.Akshat [00:29:56]: Yeah.Swyx [00:29:57]: And then I got this from John OsterhoutAkshat [00:29:58]: HuhSwyx [00:29:59]: Who I don't know if. Do John Osterhout by name?Akshat [00:30:01]: The name sounds familiar.Swyx [00:30:02]: He published a. He's a well-known professor, published a lot of interesting software design books, and this is the talk he chose to submit, is on RDMA at Inference. And I'm like, you wouldn't think that this guy, who is like operating systems guy, would care about RDMA.Akshat [00:30:20]: I, it makes sense to me because I,Swyx [00:30:24]: This is the cloud, right? YeahAkshat [00:30:25]: Like, the way you move around your KV cache and how efficiently you can do it, how efficiently you move, your weights from your training GPUs to your inference GPUs in RL is there's a lot of degrees of freedom, and it is a systems problemSwyx [00:30:41]: YeahAkshat [00:30:41]: Moving memory aroundSwyx [00:30:42]: YeahAkshat [00:30:43]: Scheduling.Swyx [00:30:44]: This shows you how primitive my understanding of networking stuff is.Swyx [00:30:46]: Is this like the domain of WireGuard as well?Akshat [00:30:50]: Not quite.Swyx [00:30:51]: It's adjacent?Swyx [00:30:53]: Explain everything.Akshat [00:30:54]: Sure.Swyx [00:30:56]: How do we move memory around GPUs?Akshat [00:30:58]: Well, so sorry. Yeah, that is memory. Sorry, I was talking more, and maybe I was talking like five minutes back, about the private IPv6, addressing that you've set up.Swyx [00:31:09]: Yeah.Akshat [00:31:09]: Is it like it's a VPN?Swyx [00:31:10]: Yeah, it is like a VPN, and yeah, WireGuard is, yeah, you're right. It is,Akshat [00:31:16]: Right. Yeah, you already moved on to new topicsSwyx [00:31:17]: A similarAkshat [00:31:18]: OkaySwyx [00:31:19]: In the same space, WireGuard is, encrypted and this is,Akshat [00:31:23]: And you don't need encryption.Swyx [00:31:23]: Yeah.Akshat [00:31:24]: Yeah.Swyx [00:31:24]: This is not encrypted. that's the main difference. This is TCP and we have eBPF programs that will reject or allow the TCP connection based on whether you're allowed to do it.Akshat [00:31:35]: Used to involve a full sidecar, but now you have eBPF in the Linux kernel.Swyx [00:31:39]: Yeah.Akshat [00:31:40]: Yeah. I don't know if this is a natural follow-on to the topic of like my skepticism on distributed training is that while, like, people spend a lot of money on, like, cables to hook up GPUs, and even that is not, like, fast enough, and that's the bottleneck, is your networking fast enough?Swyx [00:31:59]: Yeah. So I guess you're talking about fully distributed training like, Dialog or something which is like cross data centerAkshat [00:32:06]: That would be, yes.Swyx [00:32:07]: That's the extreme.Akshat [00:32:08]: Yeah.Swyx [00:32:08]: You're in the middle, and then other people would have like the Mellanox cables up in, like, their actual data center.Akshat [00:32:14]: When you run multi-node training on Modal, RDMA, I think Mellanox, is, or InfiniBand is like a, is all seen as RDMA. but it's a way to bypass the TCP networking stack and, transfer, stuff much faster, between one node, to the other. And we have I think like 3 terabit per second, internal networkingSwyx [00:32:40]: OkayAkshat [00:32:40]: Which is the standard that's needed.Swyx [00:32:42]: Okay. So I misunderstood whatAkshat [00:32:43]: 50Swyx [00:32:43]: What part of the stack you wereAkshat [00:32:44]: 50 gigs overSwyx [00:32:45]: YeahAkshat [00:32:45]: If you wentSwyx [00:32:45]: YeahAkshat [00:32:46]: RDMA.Swyx [00:32:46]: Okay.Swyx [00:32:48]: Yeah. I, very impressive work.Multi-Node Training, Post-Training, and Auto ResearchSwyx [00:32:52]: So effectively you're extending like the model philosophy to the training cluster, like, yeah.Akshat [00:32:59]: Yeah. And we're, we're not going for like large scale training runs. the thing that we've built multi-node training for is, we see a lot of, smaller scale post-training. like, people are post-training like medium sized fund models, so they can, get higher quality on inference. this is a perfect fit, for something like that.Swyx [00:33:21]: Yeah. That is my impression of how a lot of these labs explore branches in post-training and then eventually merge whatever they find in.Akshat [00:33:31]: Yeah. The other use case we've seen for multi-node training is even if you have a big cluster, your researchers are still doing small runsSwyx [00:33:38]: YesAkshat [00:33:39]: Having elasticity thereSwyx [00:33:40]: Right, sureAkshat [00:33:40]: Matters a lot more.Swyx [00:33:41]: Yeah. the, like, this is like the current limiting factor for auto research, which is like you need to give your model some GPUs in order for it to completely run.Akshat [00:33:51]: We have a blog post on auto resource and model is,Swyx [00:33:55]: YeahAkshat [00:33:56]: Yeah, like, turns out to be pretty good substrate for that.Swyx [00:33:59]: So my impression is auto research means many things, likeAkshat [00:34:01]: YeahSwyx [00:34:01]: Anything that Andrej coins. Right now it's still science fair, right? Like not like, I don't know how many people are doing this.Akshat [00:34:08]: We're having a golf.Swyx [00:34:08]: Yeah.Akshat [00:34:09]: I thought the same thing.Swyx [00:34:11]: Yeah, you would know.Akshat [00:34:12]: We, like, our internal both training and inference teams use this the general shape of this quite a bit. like we have this one internal repo called auto inference, which essentially we've automated our own forward-deployed engineering efforts using, this harness, which is, the agent will just spin up a sweep of different things. It'll even run like, NVIDIA inside profiler and it'll like tweak configs and it'll arrive the right thing. it'll change your GPUs both from H200 to B200, and works really well.Swyx [00:34:47]: Nice.Akshat [00:34:47]: So yeah.Swyx [00:34:48]: By the way, I enjoy that your forward-deployed engineering is so technical that you have to do these things.Swyx [00:34:52]: It's very different from forward-deployed engineering from other people.Akshat [00:34:54]: Yeah. For our forward-deployed engineering team is, essentially they're like applied inference researchers or applied training researchers.Swyx [00:35:02]: Someone told me like they have to be able to build, but they also have to be able to sell. do they have to sell or are they like they're good, they're just like post-sale type of thing?Akshat [00:35:09]: It does, being able to talk to a customer and engage effectively with themSwyx [00:35:13]: YeahAkshat [00:35:13]: Matters a lot.Swyx [00:35:14]: They want the same thing.Akshat [00:35:15]: Yeah.Swyx [00:35:15]: ?Akshat [00:35:15]: But it's it's not really a sales, thing. We pair them with-- We have solution architects as well that are more on the sales side.Swyx [00:35:23]: Okay. Let's spend a bit more time on auto research. This is a big focus for for this year. Where does this go? like, have people explored enough? Like, there's all these beautiful charts of like improve and then level off a bit and then you find the next thing. Is this one abstraction up from normal training? Is that how we think about it, or do you think about it differently? Like model level training versus high, like driven hyperparameter search.Auto Inference and Modal BenchAkshat [00:35:51]: Yeah, like,Swyx [00:35:51]: Someone, some people call it like neural architecture search or whatever, right? Like.Akshat [00:35:54]: Yeah, - So the stuff I've seen people do with it is nowhere on the architecture level. It's pretty much tweaking parameters, but it's it's a hyperparameter sweep that's guided by some model intuition, so it's like much more efficient than, whatever other, sweep you would have.Swyx [00:36:12]: Yeah, it's just, it's just a question of where you want to spend your compute?Akshat [00:36:16]: Right.Swyx [00:36:16]: ‘Cause yeah, you can just throw infinite amounts of money on this and somehow you'll bang out Shakespeare?Akshat [00:36:22]: Yeah, infinite monkey.Swyx [00:36:24]: Yeah, so like the very good for model. and I think it's also very important that agents can spin up other agents, can spin up their infrastructure. Like very good for you. how good is our LLMs at generating model code? Like the benefit of existing LLMs is that you are in the data.Akshat [00:36:42]: Yeah. They're, they're surprisingly good. I think like pre Cloud 4 they were not, and then now they're able to shot, stuff out of the box. But we're playing around with releasing like a Modal Bench for like the harderSwyx [00:36:55]: YeahAkshat [00:36:55]: Things, that the LLMs cannot do yet and maybeSwyx [00:36:59]: What's an example of that?Akshat [00:37:01]: I think the things that- Sometimes agents struggle with, without right guidance and a skill is, how to, use the rest of our observability. Like how to. Something is failing, like how do you look at the logs and then update the right thing? It's reasoning about that. But they're able to shot, likeSwyx [00:37:23]: Yeah. You can just add a skill to it?Compute Strategy and Capacity PlanningAkshat [00:37:26]: Yeah. So we have a Modal skill now that. Which is why we built this Modal Bench. It's to find things like that, so we can address them in our tool.Swyx [00:37:35]: Tune a skill. Yeah.Akshat [00:37:36]: Yeah.Swyx [00:37:36]: No. it's it's good. are you facing any shortages? like we talk a lot about GPU shortages, but also CPU, also memory.Swyx [00:37:44]: Yeah.Akshat [00:37:45]: We have had a lot of growth, which means that, there's - we've had to be much better aboutSwyx [00:37:53]: PlanningAkshat [00:37:54]: Proactive capacity planning.Swyx [00:37:55]: Yeah.Akshat [00:37:55]: So we have,Swyx [00:37:57]: Which by the way, like it's like a MBA's like dreamAkshat [00:38:00]: YesSwyx [00:38:00]: Is like just planning this stuff. I think last time you and I talked about something maybe about this.Akshat [00:38:03]: Yeah. we have a really competent team of people that we call, The role is called compute strategy. so yeah, if anyone listening here or wants to work on thatSwyx [00:38:13]: Compute strategy?Akshat [00:38:13]: Yeah.Swyx [00:38:14]: I think,Akshat [00:38:14]: I feel like,Swyx [00:38:15]: I think the normies call it FP&A or something.Akshat [00:38:18]: Well, it's more It's it's not FP&A. It's it's There's a lot of interesting financial questions of like what is the blend between one year and three-year reservations? how do we forecast our own capacity? how do we. especially since our capacity is very fungible across different GPU types and different regions, like you have to model a lot of it. and you also have to have an opinion on how the supply chain is gonna evolve, and then you have to like, take bets,Swyx [00:38:49]: YeahAkshat [00:38:49]: Based on that.Swyx [00:38:50]: Tokenomics.Akshat [00:38:50]: Yeah.Swyx [00:38:51]: This is like probably a not a real point, but, I was trying to think about like what other industries. I was trying to think about like, we cannot be first to like these kinds of problems.Akshat [00:38:59]: Yeah.Swyx [00:39:00]: And what other industries have had this? And I was like, airlines with fuel and like they have to hedge their fuel and like, I think for a long time Southwest because they made like a hero fuel bet, they like were like super low cost becauseAkshat [00:39:12]: OhSwyx [00:39:12]: Compared to everyone else.Akshat [00:39:14]: Yeah. I hadn't thought about that.Vibhu [00:39:16]: We're at a fun time too?Akshat [00:39:18]: Yeah. It's. A lot of the compute business in general, for us is also about being very good about capacity management. That is how you have great unit, economics. but also over time it's how you can unlock more value for customers. Like, one of the things we're building now is like a way for customers to get, If they don't care about latency, like get much cheaper pricing and they'll get results back in like next 24 hours or something, like a batch tier essentially.Batch Tiers and Latency-Insensitive WorkloadsSwyx [00:39:47]: Yeah.Akshat [00:39:47]: And those are levers we have because we control the whole stack and scheduling and whatnot to give people a sufficientSwyx [00:39:53]: Yeah. I feel like they're not as popular. Like those, like the Frontier Labs have all those APIs. They're not as popular as they should be.Akshat [00:40:00]: The demand that we see for something like that is not for LLMs. although sometimes people wanna run evals andSwyx [00:40:08]: OkayAkshat [00:40:08]: Synthetic data prep and there it makes sense.Swyx [00:40:10]: Okay.Akshat [00:40:11]: But it's from a lot of LLM companies, like people who are doing computational bio, like they have to run really big batch jobs and they don't care about when they get it back.Swyx [00:40:22]: Yeah. And like they have a reasonable. It's it's also like a cousin to the stopping problem of like, will this finish in time?Akshat [00:40:30]: Yeah. You can bound it.Swyx [00:40:33]: Yeah.Akshat [00:40:33]: Like you can give peopleSwyx [00:40:34]: YeahAkshat [00:40:34]: SLAs on it.Swyx [00:40:35]: Yeah. I think what's, what's interesting is like the next phase of model.Swyx [00:40:38]: Like what, do people expect from you, now that you're established and you're like well-known compute player among all these leading companies. You had an inference launch week, and we talked a little bit about the launches. like what else? Like what else should people know?What Modal Builds NextAkshat [00:40:55]: We are building primitives that make our users' lives much easier. So, I think for example, with LLM inference, thousands more companies are gonna post-train their own models and, deploy open source models for inference. so we're thinking a lot about what is the best product shape for that. And, that involves everything from our training gym to, then, endpoints that get frontier-level performance. again, but I haven't talked to anyone. It looks somewhat different on other verticals. Like, we're also seeing a lot of real-time, audio-video stuff in there, which is why like, we're working on things like regional routing, with fallbacks. So you can get GPUs that are as close to users as possible. so you get like low latency for video streaming and whatnot. And then on the agent side, it's,Akshat [00:41:52]: We're still working very closely with our customers because stuff is changing so fast in terms of what they need. And, I think beyond sandboxes and persistent file systems, there's a lot of other things people will need from this agent stack as they build production agents. So yeah, we're thinking about those other things that fit in there.Swyx [00:42:13]: I want to ask what the other things are.Akshat [00:42:15]: Yeah. I probably should share right now.Swyx [00:42:17]: I think-- I think, okay, so, I do think a lot about the principal components of cloud, and you do talk about compute storage networking.Akshat [00:42:25]: Yeah.Swyx [00:42:25]: Because so far for me, it's fine. so far for the. the first couple generations of cloud, it's fine. What's different, qualitatively different about agents that you need some new permission level? Like a lot of people, okay, and I'll just kinda spew tokens at you until it like hopefully sparks something.Akshat [00:42:43]: Yeah.Swyx [00:42:44]: Like the new level now is whatever Claude Code does, which is dangerously scope permissions or like allow list by command or like whatever, right? And sometimes they're like, “Well, okay, we have like this adaptive thinking mode where like, just trust me, bro. I will make the calls for you.” Is that it? like mediated permissions.Hard Guardrails vs. LLM-Mediated PermissionsVibhu [00:43:03]: Now you're looping it with a goal and letting it roll.Akshat [00:43:06]: Yeah, I'm, I'm skeptical of LLM media permission for stuff that is at the sandbox level because you do want hard boundaries.Swyx [00:43:16]: Yeah.Akshat [00:43:16]: Otherwise, someone can exfiltrate stuff.Swyx [00:43:20]: But likeAkshat [00:43:20]: YeahSwyx [00:43:20]: Maybe that's old school thinking. Maybe we're the dinosaurs.Swyx [00:43:23]: Maybe the AI OS or the LLM OS is really the kernel is a goddamn LLM.Swyx [00:43:30]: Like it makes you feel uncomfortable.Akshat [00:43:31]: Yeah, I'm, I'm toldSwyx [00:43:32]: But that's what trusting the LLM is. Like imagine a spherical cow perfect LLM.Akshat [00:43:36]: Right.Swyx [00:43:37]: That it.Akshat [00:43:39]: Maybe.Swyx [00:43:41]: I wanna test the boundaries, right?Akshat [00:43:42]: Yeah.Swyx [00:43:42]: Like, and I don't believe that, but I wanna see where I'm wrong ‘cause that's, that's the consensus.Akshat [00:43:49]: Yeah. I think you always need hard guardrails when you want, And you can pair those with softer guardrails, right? And that's gonna be a lot of mediated.Managed Agents and Specialized SandboxesSwyx [00:44:00]: There. I'll also get you a end with a couple of your commentary on like the ecosystem outside of Modal. Manage agents. Everyone has one. Gemini, OpenAI, Claude, very useful for you, but also like it is their way of starting to edge into your space.Akshat [00:44:17]: Yeah.Swyx [00:44:17]: What's going on?Akshat [00:44:19]: Yeah, we're, very excited to partner with Anthropic and some of the other foundation labs, will not name who we're also working with. the way we see it is the manage agent thing is a great place to start if you're starting out building an agent and, But then when you get to, building something more production grade, like you're a company that's like Ramp that's building their own, Ramp also runs their accounting agent on us, so their external-facing agent. You need a lot more control over, your compute primitive on things like, what sort - how do you persist different files that the agent has access to, and how do you snapshot and restore? How do you control the networking? maybe you want GPUs. When you get to that point, you kinda want, a specialized sandbox provider, that gives you those things, and that's the role that we are trying to play.Swyx [00:45:15]: YeahAkshat [00:45:16]: We don't really have an opinion on the harness, whether it runs - it's a cloud-managed agent, and you hook it up to Model Sandbox, or you run the harness in Model Sandbox. We'll see where people converge with that.Swyx [00:45:26]: Yeah. Do you any opinions on like the meta harnesses, or just another layer on top of these things?Akshat [00:45:31]: You mean like the OpenPipeSwyx [00:45:33]: OpenPipe is one. I think Vercel had one, which I can't remember the name of right now. Fredshot had one. and then, to me, most recently was Data Databricks that had Omnigen. All these are meta harness. Like it's kinda pseudo agent cloud type things.Akshat [00:45:50]: I personally have not played around with them.Swyx [00:45:53]: Yeah.Akshat [00:45:53]: Build agents with them.Swyx [00:45:54]: Everything's bullish Modal, as long as it consumes more infra.Akshat [00:45:57]: That's why we're focusing on the infra layer. It's somewhere where our, relative competence is and, also it's a hard problem to solve.Swyx [00:46:06]: Yeah. I will say like just generally reflecting on that, I don't know if - if there's other topics on Modal, but like just generally reflecting as an infra person, not as intense as you, but in that field, this has like been the most exciting time in infra. Like it was boring for a while, and you couldn't really get people excited about data infrastructure. Like Eric would get on Data Console, everyone just watched the video and like say, “Look at how many sandboxes I can spin up,” and no one gave a crap.Why Infrastructure Became Exciting AgainAkshat [00:46:39]: Yeah.Swyx [00:46:40]: And like now everyone gives a crap.Akshat [00:46:42]: That's true. It is a very exciting time, and I think a lot of that's driven by just the amount of scale all of this stuff needs.Swyx [00:46:50]: I think the, like a lot of your initiatives or a lot of your like product directions make sense in retrospect, which is like the best kind, but I wouldn't necessarily have thought about it myself, which.Akshat [00:47:00]: We need the predictions.Swyx [00:47:02]: I think there's a lot that you just don't even see, right? Like you have the batch, you have the voice, you have the multimodal, but what else?Akshat [00:47:10]: What else is coming up for usSwyx [00:47:11]: Yeah. Where do you see things going?Akshat [00:47:13]: Yeah. I, in generalBiotech, Robotics, and Non-LLM AI WorkloadsAkshat [00:47:15]: It's it's clear that there's there's a huge shift happening. I think one thing that's not as obvious to people because LLM inference gets talked about so much and is also we work a lot of companies that are, doing things like drug discovery and computational bio, like the Chai Discoveries of the world. Big things are probably gonna happen there. we work a lot of robotics companies that are putting robots in like active deployments and getting good results out of them.Swyx [00:47:45]: Is there Air Gap Modal? Is there a version that is like prem air gapped whatever?Akshat [00:47:50]: No. We,Swyx [00:47:51]: You should cloud only.Akshat [00:47:51]: Yeah.Swyx [00:47:52]: Yeah. Okay. But yeah, so what you're saying is like because you're focused on primitives and they're good primitives, you find use cases in all these kinds of things.Akshat [00:48:01]: Yeah.Swyx [00:48:01]: Probably diversifies you a little bit away from LMS all the time.Akshat [00:48:05]: Yeah, absolutely. We're, we'- our goal isn't to only serve the LLM inference market.Swyx [00:48:10]: There are a lot just on the website, the audio,Akshat [00:48:12]: Yeah. We said both onSwyx [00:48:14]: Computational bio images. Yeah, there's a lot here. There's QTA TTS, customizing. Oh, Chatterbox. there was customizing Whisper.Akshat [00:48:24]: Okay. Yeah.Swyx [00:48:25]: This screen reminds me of a fallen competitor, which Replicate.Model APIs vs. Differentiated AI ProductsSwyx [00:48:31]: What's your postmortem on what happened?Akshat [00:48:34]: This is one thing we've stayed away from is providing an API for models because I think providing model APIs is some of it ends up serving like a really hobbyist market, which is much less sticky.Swyx [00:48:50]: Yeah.Akshat [00:48:50]: And we've always wanted to build for companies that are building products and need more flexibility that's not just an API.Swyx [00:48:57]: Which you can build an API for a model and this is clearly what it is. But you - but what you're saying, you can wrap it into a more fully functioning back end that you run.Akshat [00:49:06]: Yeah. So all of our examples, it's not that spin up this model, here's an API token, use it. They're all code.Swyx [00:49:13]: Okay.Akshat [00:49:13]: And so the point is that this is just an example.Swyx [00:49:16]: Starter code.Akshat [00:49:17]: Yeah. But you can tweak it however you want.Swyx [00:49:20]: Yeah.Akshat [00:49:21]: And if you're like a company building a product, like, computational bio whatnot, yeah.Swyx [00:49:26]: I guess I'm trying to tease out for listenersAkshat [00:49:28]: YeahSwyx [00:49:28]: When does it stop becoming, oh, you're just an API call and you're just a wrapper on API to becoming what you call a product, right?Swyx [00:49:36]: Like, what is that layer? Like what-- Like, more lines of code, but like beyond that, what is the substance that people add that qualifies it to be something more?Akshat [00:49:46]: I think there's a little bit of like a selection effect of like a lot of the companies who do wanna get deeper into that level are probably building something that's more differentiated. And, I think, an example is like - with LLM inference, originally we, worked with companies that were building their own post-training frameworks or they were, - Ramp early in the day was training their own tokenizer and like swapping out the tokenizer in Llama and whatnot. I'm not saying that's, that successful, in that case. But a better example is like, let's say Suno. because Suno, does not use Modal for training.Swyx [00:50:26]: Mikey on the pod. Yeah.Akshat [00:50:27]: But they use Modal for all their inference and that's because they have like a custom-- They have completely custom model architecture and that means that they have to be at the code level and tweak things that are not, just an API.Swyx [00:50:41]: It's interesting as well, like we had, Ethan, most recently on the xAI Groq team make a prediction that like the next tier in video gen is not a better video model, it's a better model or agent that orchestrates video models.Video Agents and Production WorkflowsAkshat [00:50:56]: Oh, interesting.Vibhu [00:50:56]: Language model backbone that can use toolsAkshat [00:50:58]: RightVibhu [00:50:59]: And write code.Akshat [00:51:00]: Like, yes, I can make my second video or my second video from Groq, but I want my minute video.Akshat [00:51:06]: And I'm not going there through normal video gen.Swyx [00:51:10]: Yeah, that's interesting. I - So we have GPU sandboxes and recently have seen a few companies doing agents that do video manipulation or,Akshat [00:51:22]: Yeah. Give it FFmpeg and just do it.Swyx [00:51:23]: Run FFmpeg. But likeAkshat [00:51:25]: That's not enough.Swyx [00:51:25]: Yeah.Akshat [00:51:26]: You need to give it Adobe.Swyx [00:51:27]: Yeah, I hadn't put it together with like it would be a video production thing. in my mind these things were going more towards editingAkshat [00:51:36]: Yeah.Vibhu [00:51:36]: Well, shout out Mantis.Akshat [00:51:37]: I think about this a lot.Swyx [00:51:38]: .Akshat [00:51:41]: Yeah. Sorry.Vibhu [00:51:41]: Luma. Luma Agent is a version of this for video production, but it's a off.Swyx [00:51:46]: I was gonna get your quick takes, on some other stuff that happensGitpod/Ona, CI, and Runtime SandboxesSwyx [00:51:50]: In recent news and just-just see if you have anything interesting. Gitpod, very li

Have you ever built something for yourself that you were almost embarrassed to admit you needed? David Gonzalez did. And it changed how he sees himself entirely. In this episode of The Happy Hustle Podcast, I sit down, well, I hand the mic over, to David Gonzalez during our AI Masterclass Workshop, and he delivers one of the most real, unfiltered breakdowns of agentic AI I have heard all year. David is the founder of Internet Marketing Party, a monthly founder room he has run since 2008 without missing a single month. Almost eighteen years in, he has brought over 20,000 entrepreneurs through his events and facilitated half a billion dollars in introductions and revenue. Names like Alex Hormozi, Jay Abraham, and Craig Clemens have all shared his stage. David built his entire career and net worth on one skill, putting the right people in the right room. He is based out of Austin, Texas, and he is the first to tell you he resisted vibe coding for months before it cracked his whole business wide open. This episode matters because David is not a developer. He is not an engineer. He is, in his words, an offer guy, an operator, someone who knows the nuts and bolts of business but never touched code. And that is exactly why his story lands so hard. If you have ever thought AI was not for you because you are not technical, David is proof that thinking is wrong. He opens by admitting he quit vibe coding months earlier because he was doing it all wrong. His computer ran slow, he had a dozen unfinished projects, and he did not even know what GitHub was. Then a mentor rebuilt his setup in three hours, and his usage went from 61 sessions in April to over 2,285 sessions by June. That is not a typo. That is what happens when the right system finally clicks. David makes a distinction that is worth sitting with. Chat AI, the kind most people use, is basically an expensive browser. You ask it a question, it gives you an answer, and you are still stuck doing the work yourself. Agentic AI, what he calls vibe coding, is a completely different animal. He compares it to a Blackhawk helicopter that flies itself versus a tricycle. You do not need to know how to pilot it. You just need to know where you want to go. To prove it, David builds an app live in front of the room. He shows the difference between a lazy prompt and a detailed one, and the gap in quality is not subtle. A good prompt, he explains, gives the AI the context of a real thinking partner, not just a task list. He calls it inviting a swarm of experts to the table, the same way you might know that one guy at a family cookout who can fix the smoker, close a deal, and give you parenting advice all in one conversation. Then he gets personal. David built himself a tool he calls iEngine, standing for Intelligence, Inference, and Insights. It reads fifteen years of his text messages, over 1.4 million of them, along with his entire call history, and sorts every contact in his life by relationship depth. He started with 7,368 contacts and one keystroke later had them organized down to what actually mattered. It tells him who to reach out to, who has gone cold, and what he owes people. He puts it simply. If you are using a normal CRM, you are the CRM's bitch. He flipped that. But the most powerful moment in this episode has nothing to do with software. David shares a story about realizing he had been wearing an expensive watch as a stand in for his own worth, a way to prove to the world he had made it after growing up near poverty. One day he left the house without it and drove off anyway, and something shifted. He realized AI had not made him enough. It just gave him the tools to access what was already there. He used to let people call him a super connector. Now he knows he is an architect. He builds things. He always could. If you run a business and have ever felt boxed in by what you think AI can or cannot do for you, this one will reframe everything. Head over to https://caryjack.com/podcastin/ and listen to the full episode. David's energy alone is worth the time, but the mindset shift underneath it is what will actually stick with you. Connect with Davidhttps://www.facebook.com/InternetMarketingPartyhttps://www.instagram.com/internetmarketingparty/https://www.youtube.com/@ImarketingPartyhttps://www.linkedin.com/in/davegonzalez/ Find David on this website: https://simplythecoolest.com/ Connect with Cary!https://www.instagram.com/caryjack/https://www.facebook.com/SirCaryJackhttps://www.linkedin.com/in/cary-jack-kendzior/https://twitter.com/thehappyhustlehttps://www.youtube.com/channel/UCFDNsD59tLxv2JfEuSsNMOQ/featured Get a copy of his new book, https://www.thehappyhustle.com/book Sign up for The Journey: 10 Days To Become a Happy Hustler Online Course @ https://thehappyhustle.com/thejourney/ Apply to the Montana Mastermind Epic Camping Adventure @ https://thehappyhustle.com/mastermind/ “It's time to Happy Hustle, a blissfully balanced life you love, full of passion, purpose, and positive impact!” Episode Sponsors: If you're feeling stressed, not sleeping great, or your energy's been kinda meh lately—let me put you on to something that's been a total game-changer for me: Magnesium Breakthrough by BiOptimizers. This ain't your average magnesium—it's got all 7 essential forms that your body needs to chill out, sleep deeper, and feel more balanced. I take it every night and legit notice the difference the next day. No more waking up groggy or tossing and turning all night If you're ready to sleep like a baby, calm your nervous system, and optimize your recovery, go grab yours now at https://www.bioptimizers.com/happy and use code HAPPY10 for 10% OFF. =================================================================== My Green Mattress If you've been waking up with back pain, feeling stiff, or just not getting that deep, quality sleep. This might be what you're missing: My Green Mattress. It's made with clean, non-toxic, and eco-friendly materials, so you're not just sleeping better, you're sleeping healthier too. The comfort and support are on another level, and you can really feel the difference night after night. If you're ready to invest in better sleep and better recovery, check it out at https://thehappyhustle.com/mygreenmattress =================================================================== Ozlo Sleep If you've been struggling to fall asleep, stay asleep, or just wake up feeling actually rested, let me put you on to something that's been a total game-changer: Ozlo Sleep. These aren't your typical sleep buds. They're designed to block out noise and help your brain fully relax, so you can drift off faster and stay in deep, uninterrupted sleep. Perfect if you're a light sleeper or just want that next-level rest. If you're ready to upgrade your sleep and wake up feeling recharged, check out https://ozlosleep.com and save $80 OFF using code HAPPY.

Sean Juroviesky is a senior security engineer at SoundCloud. In this episode, he joins host Paul John Spaulding to discuss shift happening in AI around where inference actually runs, a recent situation with AWS Bedrock and Anthropic routing inference outside the EU, and more. • For more on cybersecurity, visit us at https://cybersecurityventures.com

Amos joins us to break down why inference is the new oil and companies will hedge inference costs the same way airlines hedge jet fuel, why AntSeed is open router without open router in the middle, and why all network revenue goes to buy and burn with no company capturing value. 14 years in crypto. This is what he's building next.Amos Meiri is Co-Founder of AntSeed, an open-source P2P marketplace for AI inference, and a 14-year crypto veteran who helped build colored coins alongside Vitalik before Ethereum existed.The Rollup is where the leaders of digital assets and finance converge. Live from the financial capital of the world.Timestamps00:00 Intro03:17 Inference Is New Oil06:27 AI Financialization Just Starting09:27 BitTorrent Model For Inference11:59 1,000 Users Three Months14:45 Open Market Creates Competition17:31 Foundation Not A Company20:01 Liquid Inference Token Explained22:38 Venice DM Token Parallel25:36 Permissionless Inference Is FutureGuest Socials:Amos Meiri X: https://x.com/AmosMeiriAntSeed X: https://x.com/AntSeedAIAntSeed Website: https://antseed.com/Partners: Better than Banks. Transparent capital efficiency earning the highest yields in DeFi. Learn more here: https://infinifi.xyz/---Dinari - Over 230 1:1 backed tokenized stocks, ETFs & more with dividends. US-based SEC transfer agent. Available on 5+ chains & via API. https://dinari.com/---Relay is the fastest and most reliable way to swap any token on any chain. Learn more here: https://relay.link/bridge---Zama is an open source cryptography company that builds state-of-the-art Fully Homomorphic Encryption (FHE) solutions for blockchain.Learn more here: https://www.zama.org/---Trezor is the creator of the first-ever hardware wallet. Securing crypto for 2M+ users worldwide. 100% open source. Learn more here: https://affil.trezor.io/aff_c?offer_i...---

My guests today are Gavin Uberti and Rob Wachen, the founders of Etched.  A few years ago, when they set out to build a better AI chip than the largest companies in the world, almost everyone I called told me it could not be done. They have since done it, taping out a working chip on their first attempt and becoming the first hardware company founded after ChatGPT to do so. They already have more than a billion dollars of customer demand for their first product, and have raised eight hundred million dollars to build it.  Etched builds chips and systems designed to run AI models faster and at lower cost. They started the company in 2023, and that product is a complete rack for inference, the chip along with the boards, the power delivery, the interconnects, and the manufacturing to produce it all. We talk about the technical bets behind their architecture, how they hired industry legends and paired them with elite 22 year-olds, and why they believe inference will become one of the largest markets in the world. I think you will find the story of what they have built hard to forget. Please enjoy my conversation with Gavin and Rob. For the full show notes, transcript, and links to mentioned content, check out the episode page ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠here⁠⁠⁠⁠⁠.  ----- Become a Colossus member to get our quarterly print magazine and private audio experience, including exclusive profiles and early access to select episodes. Subscribe at ⁠colossus.com/subscribe⁠. ----- ⁠Ramp's⁠ mission is to help companies manage their spend in a way that reduces expenses and frees up time for teams to work on more valuable projects. Go to⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ ⁠ramp.com/invest⁠⁠ to sign up for free and get a $250 welcome bonus. ----- Trusted by thousands of businesses, ⁠Vanta⁠ continuously monitors your security posture and streamlines audits so you can win enterprise deals and build customer trust without the traditional overhead. Invest Like the Best listeners get a special offer of $1,000 off Vanta when you go to ⁠vanta.com/invest⁠.  ----- WorkOS⁠ is the infrastructure B2B and AI-native companies use to sell to enterprise. It covers everything enterprise security requires: SSO, SCIM, RBAC, Audit Logs, AI governance, and more. Trusted by 2,000+ fast-growing companies, including OpenAI, Anthropic, Cursor, and Vercel. ----- Rogo is the AI platform for finance. They're building agents for Wall Street that are trained to understand how bankers and investors actually do work: from diligence and modeling, to turning analysis into deliverables. To learn more, visit rogo.ai/invest. ----- ⁠Ridgeline⁠ has built a complete, real-time, modern operating system for investment managers. It handles trading, portfolio management, compliance, customer reporting, and much more through an all-in-one real-time cloud platform. Visit⁠ ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ridgelineapps.com⁠. ----- Editing and post-production work for this episode was provided by The Podcast Consultant. Timestamps: (00:00:00) Welcome to Invest Like The Best (00:02:07) Gavin Uberti and Rob Wachen (00:03:54) Two 21-Year-Olds Taking on NVIDIA (00:07:52) The Two Technical Bets Behind Their Architecture (00:14:15) Why Inference Becomes the Biggest Market (00:20:23) Rob and Gavin's Origins Stories (00:28:38) How They Recruit Industry Legends (00:36:30) Moving a Dozen Engineers to Bangalore for Six Months (00:38:01) Speed Wins (00:43:58) Getting More Concurrency Out of Every Megawatt (00:52:44) Vertical Integration (00:57:43) Hardest Obstacles to Overcome (01:01:09) Raising The Largest AI Chip Series A Ever (01:06:29) TSMC (01:13:20) Designing Gen 2 for Gigawatt-Scale Production (01:16:42) Why Machines Don't Think Like People (01:20:03) A Year of Compute Compressed Into a Month (01:23:44) The Trillion-Dollar Data Center (01:26:19) The Kindest Thing

When AI is at its best, the conversations can feel uncanny—almost magical in their accuracy, relevance, and speed. For that you can thank the AI agents that work together behind the scenes to search, reason, and sift through all your content to get you what you need to do your job. We talk with Jongmin Baek and Marta Mendez, two Dropbox machine learning engineers, about building conversational AI that's helpful, useful, and grounded in your team's shared context, so you can spend more time on the work that really matters. ~ ~ ~  Working Smarter is brought to you by Dropbox. Find, organize, and share your work—all in one place—with context-aware AI from Dropbox. You can listen to more episodes of Working Smarter on Apple Podcasts, Spotify, YouTube, Amazon Music, or wherever you get your podcasts. To read more stories and past interviews, visit workingsmarter.ai This show would not be possible without the talented team at Cosmic Standard: producer Ben Montoya, sound engineer Aja Simpson, technical director Jacob Winik, and executive producer Eliza Smith. Special thanks to our illustrator Fanny Luor, marketing consultant Meggan Ellingboe, and editorial support from Catie Keck.  Our theme song was composed by Doug Stuart.  Working Smarter is hosted by Matthew Braga. Thanks for listening!

Dakin Campbell, AI Finance Reporter, talks with TITV Host Akash Pasricha about the former AWS chief taking the helm of Helix Infrastructure Partners, a $10 billion data center company aiming to compete in the AI infrastructure race. We also talk with The Information's Catherine Perloff about Anthropic shifting its pricing model with Amazon to tokens, which will increase costs for the tech giant. We then chat with Laura Bratton about how Anthropic's new Claude integration for Slack is sparking cannibalization concerns among Salesforce employees. Finally, we get into OpenAI's secret internal optimizations that cut model inference costs by more than half with our AI reporter Stephanie Palazzolo.Articles discussed on this episode: https://www.theinformation.com/articles/amazon-pay-anthropic-technology-new-dealhttps://www.theinformation.com/articles/new-kkr-venture-hunts-deals-clear-data-center-logjamhttps://www.theinformation.com/newsletters/ai-agenda/openai-discovers-new-way-cut-inference-costs-halfhttps://www.theinformation.com/articles/salesforce-employees-worry-anthropics-invasion-slackSubscribe: YouTube: https://www.youtube.com/@theinformation The Information: https://www.theinformation.com/subscribe_hSign up for the AI Agenda newsletter: https://www.theinformation.com/features/ai-agendaTITV airs weekdays on YouTube, X and LinkedIn at 10AM PT / 1PM ET. Or check us out wherever you get your podcasts.Follow us:X: https://x.com/theinformationIG: https://www.instagram.com/theinformation/TikTok: https://www.tiktok.com/@titv.theinformationLinkedIn: https://www.linkedin.com/company/theinformation/Chapters:00:00 - Introduction01:13 - Adam Selipsky Leads $10B AI Infrastructure Play12:56 - Anthropic Hits Amazon With Token Pricing Shift20:46 - Salesforce Staff Clashing Over Claude in Slack31:11 - OpenAI Sneaks Out 50% Inference Cost Drop

Plus: Qualcomm to acquire AI software firm Modular in $3.9 billion stock deal. And Zoox debuts redesigned robotaxi for large-scale production. Julie Chang hosts. Learn more about your ad choices. Visit megaphone.fm/adchoices

Explore how the latest advancements in AI are shifting from traditional training to inference-focused efficiencies, and how companies like Adaptation Labs are pioneering adaptive, full-stack AI solutions that democratize control across industries.Key topics:The evolution from compute-heavy training models to efficient inference layersHow inference costs are changing despite increasing AI demandThe role of adaptive, gradient-free learning in democratizing AI customizationChallenges with the last 5% reliability gap and continuous learningThe importance of full-stack optimization—from data to interfaces in AI systemsFuture trends: decentralized AI, edge computing, and ongoing innovationTimestamps:00:00 - Introduction to AI trends: scaling vs inference efficiencies01:01 - Sudip's background: Google Brain, DeepMind, and inference infrastructure01:34 - The rapid growth of foundation and large language models02:36 - Comparing traditional ML project timelines to large foundation models04:20 - The transformative potential of foundation models in enterprise and underserved communities05:33 - The shift from task-specific models to general-purpose foundation models07:07 - How inference costs have evolved: the rising demand vs falling per-token costs08:37 - The challenge of inference in trillion-parameter models and the move towards smaller, verticalized models10:14 - Factors driving high inference costs: model size, reasoning, agentic workloads12:13 - The probabilistic nature of inference and API pricing complexities13:07 - Variability in inference costs and demand in real-world scenarios14:14 - The autoregressive, sequential nature of LLM inference and system challenges16:45 - Cost implications of autoregressive inference and the move to more efficient, localized models18:18 - The motivation behind Adaptation Labs: democratizing AI control and customization19:47 - Adaptive, gradient-free continual learning and environment interaction21:26 - Co-optimizing full-stack AI: systems, interfaces, and models22:34 - How interface design impacts AI adoption and continuous learning23:55 - The evolution of techniques: from foundational training to open-source innovations26:18 - Handling the ‘last 5%' reliability challenge in enterprise AI deployments28:02 - The importance of system feedback and adaptive learning in coding and decision-making31:12 - Adaptive Data and AutoScientist: seamless data transformation and model co-optimization32:55 - Use cases: finance, low-resource languages, long context data34:13 - The role of inference techniques and creating high-quality data for customization36:10 - Future of adaptive, task-specific interfaces and continuous, real-time learning38:49 - Full-stack AI: data, models, interfaces, and their iterative feedback loops41:18 - The competition between fine-tuning and adaptive inference techniques43:29 - The origin of new inference techniques: industry labs, open source, and innovation hubs45:27 - The “last 5%” reliability gap: why it's critical and how dynamic learning can help48:27 - Hardware vs software optimization in AI systems and the future of systemic efficiency51:25 - Growing AI demand, hardware constraints, and the opportunity for systemic innovation52:48 - The shift from training to inference and decentralized AI models at the edge54:12 - Final thoughts: the evolving landscape and long-term AI innovationConnect with Sudip:LinkedInConnect with Nataraj:⁠LinkedIn⁠

How do you build AI that actually understands you and the work you do? It all starts with having the right context.  We talk with Dropbox staff product manager Noorain Noorani and principal engineer Sean-Michael Lewis about the art of context engineering and how Dropbox connects to all the tools your team needs for work—so you get AI that works wherever you do.  ~ ~ ~  Working Smarter is brought to you by Dropbox. Find, organize, and share your work—all in one place—with context-aware AI from Dropbox. You can listen to more episodes of Working Smarter on Apple Podcasts, Spotify, YouTube, Amazon Music, or wherever you get your podcasts. To read more stories and past interviews, visit workingsmarter.ai This show would not be possible without the talented team at Cosmic Standard: producer Ben Montoya, sound engineer Aja Simpson, technical director Jacob Winik, and executive producer Eliza Smith. Special thanks to our illustrator Fanny Luor, marketing consultant Meggan Ellingboe, and editorial support from Catie Keck. Our theme song was composed by Doug Stuart.  Working Smarter is hosted by Matthew Braga. Thanks for listening!

Semiconductors have moved from the background of the technology stack to the center of the AI economy. What used to be a specialized industry discussed mostly by engineers and investors is now shaping the speed, cost, and strategic direction of modern computing.In this episode of TechSurge, host Michael Marks speaks with Stacy Rasgon, Managing Director and Senior Analyst covering U.S. semiconductors and semiconductor capital equipment at Bernstein Research. Stacy has spent years analyzing the chip industry across cycles, but argues that the current moment feels different in scale: AI demand has created an unprecedented scramble for compute, memory pricing has surged, and companies across the stack are being forced to rethink capacity, architecture, and capital allocation.The conversation explains the 4 different kinds of semiconductor cycles—supply, inventory, product, and demand — and why Stacy believes the industry is currently in a demand cycle of unusual magnitude. The discussion also unpacks the distinction between DRAM and NAND, why high-bandwidth memory is becoming strategically central to AI systems, and how the physical realities of wafer capacity and silicon area are constraining supply in ways the broader market often misses.Stacy and Michael also discuss the hardware economics behind the current boom, with Michael pressing Stacy on why compute remains so scarce and how companies are improving performance through packaging and system design. Michael then moves the conversation beyond market headlines to the core business questions: who is actually paying for this compute, which use cases are generating real revenue, and whether AI spending is creating durable economic value or simply shifting costs elsewhere. Together, these questions highlight two of the episode's clearest insights: coding may be one of the earliest AI applications with meaningful willingness to pay, and inference, not training, is the real test of whether the current buildout becomes a lasting business or just another expensive wave of infrastructure.Stacy explains the concentration of power among the major wafer fabrication equipment players, the rise of ASICs as a meaningful share of AI silicon, Broadcom's rapidly expanding AI opportunity, and the growing role of Chinese companies as new entrants, especially in memory and semiconductor equipment. Along the way, the conversation asks the defining question facing the sector: is this just another semiconductor upswing, or the first true supercycle the industry has seen? Stacy believes that this might be the biggest supercycle he has seen in his career.Sign up for our newsletter at techsurgepodcast.com for updates on upcoming TechSurge Live Summits and future episodes.Links:Stacy Rasgon on LinkedIn: https://www.linkedin.com/in/stacy-rasgon-6924963Bernstein: https://www.alliancebernstein.com/corporate/en/home.htmlReferences Mentioned During the DiscussionNVIDIA Blackwell Platform: https://www.nvidia.com/en-us/data-center/blackwell-platform/High Bandwidth Memory (HBM) overview from Micron: https://www.micron.com/products/memory/hbmDRAM overview from IBM: https://www.ibm.com/think/topics/dramNAND flash overview from IBM: https://www.ibm.com/think/topics/nand-flash-memoryFurther ReadingMcKinsey on the semiconductor industry outlook: https://www.mckinsey.com/industries/semiconductors/our-insights/the-semiconductor-industry-in-2025Semiconductor Industry Association: 2025 State of the U.S. Semiconductor Industry: https://www.semiconductors.orgNVIDIA on the Blackwell architecture and AI infrastructure roadmap: https://www.nvidia.com/en-us/data-center/blackwell-platform/Broadcom AI investor materials and infrastructure commentary: https://investors.broadcom.comASML on lithography and advanced chip manufacturing: https://www.asml.com/en/technologyMicron on HBM and AI memory demand: https://www.micron.com/products/memory/hbmChapters[00:00:00] — Highlights[00:00:26] — Welcome to  the Episode[00:01:29] — Meet Stacy Rasgon[00:02:01] — Is This the First Real Semiconductor Supercycle?[00:05:33] — Inside the Strongest Memory Cycle in History [00:09:14] — Can Innovation Keep Up With AI Demand?[00:11:33] — Chiplets, Blackwell, and the New Economics of Compute [00:12:37] — What Could Signal the Cycle Is Slowing[00:14:26] — Vertical Integration at the Hyperscales [00:16:36] — The Difference between Apple and Meta[00:17:15] — What is Vertical Integration Being Done For?[00:18:15] — Will other bottlenecks develop as This Progresses? [00:21:13] — Oligopoly Pricing in the Market[00:22:22] — Any New Entrants into Memory?[00:23:46] — Why the Industry Must Pivot From Training to Inference[00:25:10] — Agentic Coding and the First Real AI Revenues[00:26:57] — Groq, Low-Latency Inference, and What GPUs Cannot Do Alone[00:29:28] —-Could The Smaller Companies All be Bought Up ?[00:30:19] — Why Semiconductor Equipment Matters More Than Ever [00:31:00] — How Semiconductor Equipment is Affected by the Cycle[00:32:55] — A Long Upcycle for Semiconductor Equipment Guys?[00:33:13] — The Big Five and the Rise of Chinese Equipment Players[00:34:24] — The Effects of Geopolitics[00:35:02] — Broadcom's Quiet AI Breakout[00:40:46] — ASICs vs GPUs and the Next Wave of Custom Chips[00:41:06] — Intel, Foundry Strategy, and the Long Turnaround[00:46:46] —-The Risks the Market May Still Be Underestimating[00:49:32] — Where Startups Still Have Room to Win[00:50:39] — What the Semiconductor Industry Could Look Like Next Year

The Information's Asia Bureau Chief Jing Yang breaks down the unique five-year lockup and partnership structure behind DeepSeek's massive $7.4 billion capital raise. AI Finance Reporter Dakin Campbell explains the financial underpinnings and balance sheet risks of Broadcom's $35 billion hardware financing backstop for Anthropic. Then, Nvidia Reporter Phoebe Liu shares exclusive data showing how the chip giant grew its AI inference market share to 74% over the past year. Finally, MNTN CEO Mark Douglas analyzes the consolidation wave driving Fox's $22 billion acquisition of Roku.Articles discussed on this episode: https://www.theinformation.com/articles/polymarket-kalshi-take-steps-block-fraud-ringshttps://www.theinformation.com/newsletters/the-briefing/rokus-timely-exit-salesforces-dealmaking-revvinghttps://www.theinformation.com/newsletters/ai-infrastructure/nvidias-share-ai-inference-chip-market-appears-risinghttps://www.theinformation.com/articles/deepseek-closes-record-7-billion-plus-funding-unusual-deal-structureSubscribe: YouTube: https://www.youtube.com/@theinformation The Information: https://www.theinformation.com/subscribe_hSign up for the AI Agenda newsletter: https://www.theinformation.com/features/ai-agendaTITV airs weekdays on YouTube, X and LinkedIn at 10AM PT / 1PM ET. Or check us out wherever you get your podcasts.Follow us:X: https://x.com/theinformationIG: https://www.instagram.com/theinformation/TikTok: https://www.tiktok.com/@titv.theinformationLinkedIn: https://www.linkedin.com/company/theinformation/Chapters:00:00 - Introduction 01:13 - Inside DeepSeek's Uncommon $7.4B Funding Round 13:52 - Polymarket and Kalshi Tech Steps to Curb Fraud Rings 22:28 - Broadcom's Risky $35B Move to Finance Anthropic Chips 30:53 - Nvidia Gathers Speed with Rising AI Inference Market Share 38:43 - Fox to Acquire Roku for $22B in Streaming Consolidation

In this episode of the Crazy Wisdom Podcast, host Stewart Alsop sits down with Larry Swanson, creator of the Knowledge Graph Insights Podcast, for their second conversation together. The two cover a wide range of interconnected topics, starting with a correction Larry makes about the true origin of the term "artificial intelligence," tracing it back to the 1956 Dartmouth Conference and its distinction from Norbert Wiener's cybernetics. From there, the conversation moves through the history and structure of knowledge graphs, ontologies, RDF (Resource Description Framework), and the W3C standards process, touching on concepts like the T-box, A-box, and C-box, as well as the 25th anniversary of the Semantic Web paper. Stewart and Larry also dig into the limitations of large language models — particularly around reasoning, confabulation, and what Larry describes as "cognitive surrender" — and why symbolic AI and knowledge engineering may hold answers that the neural network world hasn't fully embraced. The episode also ventures into consciousness, panpsychism, Michael Pollan's ideas, and Stewart's own hands-on experience vibe coding a personal chatbot to replace functionality he feels he's lost with recent changes to Claude. Larry's podcast can be found at kgi.fm.Timestamps00:00 - Stewart introduces Larry Swanson; Larry corrects the record on AI's origin, distinguishing it from Norbert Wiener's cybernetics at the 1956 Dartmouth conference.05:00 - Larry discusses interviewing semantic web paper coauthors on its 25th anniversary; RDF's hidden ubiquity compared to SIM cards powering everything invisibly.10:00 - Knowledge graphs explained through t-box terms, a-box assertions, and Dave McComb's c-box; IKEA's three-layer knowledge graph as a practical example.15:00 - Stewart connects metadata complexity to AI needs; faceted search explained as c-box attributes driving product filtering experiences.20:00 - RDF 1.2 reification standards discussed; W3C's rigorous recommendation process powering governments and enterprises worldwide through collaborative standards.25:00 - Cyc project examined as influential "successful failure"; Pat Hayes bringing description logic into semantic web; LLMs lacking true reasoning capability.30:00 - Epistemological fault lines between human and computer intelligence; cognitive surrender paper reveals no intelligence threshold protects against AI manipulation.35:00 - Stewart's Claude regression problem drives chatbot vibe coding quest; small language models and domain-specific approaches explored as alternatives.40:00 - Consciousness discussion through Michael Pollan's panpsychism lens; language versus cognition disconnect revealing LLMs as pure token-stitching without genuine thought.45:00 - Context graphs as purpose-built knowledge graphs for AI; Stewart's planning agents versus coding agents architecture and ground truth verification problem.50:00 - Docs-as-code versus code-as-docs paradigm shift; knowledge graphs as universal verifiers against validated facts; RDF 1.2 enabling provenance and degrees of certainty.55:00 - Jessica Talisman's Knowledge Graph Academy recommended for onboarding; kgi.fm podcast shared; knowledge representation community needs better abstraction for wider adoption.Key Insights1. The term "artificial intelligence" was not a marketing gimmick but was coined deliberately at the 1956 Dartmouth Conference to distinguish the work of John McCarthy from Norbert Wiener's cybernetics. The two camps represented genuinely different approaches, and the AI label was a form of intentional intellectual branding rather than empty promotion.2. The semantic web, often called the most successful failure in technology history, has quietly embedded itself everywhere despite never achieving its original vision. Technologies like RDF power metadata standards inside every Adobe product and form the invisible backbone of government systems, enterprise data infrastructure, and cultural heritage organizations worldwide.3. Knowledge graphs are best understood as an ontology combined with all the instances that populate it. The distinction between things and strings, popularized by Google in 2012, captures the core idea that knowledge representation is about concepts as distinct from the labels we give them.4. The t-box, a-box, and c-box framework offers a practical model for understanding knowledge architecture. The t-box holds terminology and concepts, the a-box holds assertions about specific instances, and the c-box manages the attributes, taxonomies, and controlled vocabularies that sit between them and enable things like faceted search.5. Large language models produce fluent, convincing output but lack genuine reasoning, epistemological grounding, or judgment. Research on cognitive surrender shows that even people who understand how LLMs work are still susceptible to being misled by their fluency, meaning intelligence and awareness offer no reliable protection against being deceived.6. The gap between language and cognition matters deeply when evaluating AI. Evidence from people with aphasia shows that thinking can occur without language, which suggests LLMs, being purely language-based systems, are missing a fundamental layer of cognition that cannot be recovered through more tokens or better training.7. Knowledge graphs and RDF-based representation are well suited to the problem of verification and grounding in AI systems. Rather than relying on vectorized embeddings of language, a knowledge graph can store validated, provenance-tracked facts with degrees of certainty, making it a natural foundation for building trustworthy AI applications.

Brendan Burke says now is the time for tech, and Intel (INTC) has a growing role in the AI buildout. He believes current CEO Lip-Bu Tan will turn Intel into a core collaborator with AI hyperscalers that serves as a compelling foundry alternative to TSMC (TSM). Brendan also expects Intel to serve as a strong inferencing and CPU manufacturer as spending for AI accelerates. ======== Schwab Network ========Empowering every investor and trader, every market day.Subscribe to the Market Minute newsletter - https://schwabnetwork.com/subscribeDownload the iOS app - https://apps.apple.com/us/app/schwab-network/id1460719185Download the Amazon Fire Tv App - https://www.amazon.com/TD-Ameritrade-Network/dp/B07KRD76C7Watch on Sling - https://watch.sling.com/1/asset/191928615bd8d47686f94682aefaa007/watchWatch on Vizio - https://www.vizio.com/en/watchfreeplus-exploreWatch on DistroTV - https://www.distro.tv/live/schwab-network/Follow us on X – https://twitter.com/schwabnetworkFollow us on Facebook – https://www.facebook.com/schwabnetworkFollow us on LinkedIn - https://www.linkedin.com/company/schwab-network/About Schwab Network - https://schwabnetwork.com/about

Artificial intelligence is moving faster than ever, but as AI models continue to grow in size and complexity, the challenges surrounding inference performance are becoming impossible to ignore. In this week's podcast, ElastixAI CEO Dr. Mohammad Rastegari and I chat about how we can overcome those challenges and why a different approach to AI infrastructure is necessary for the next generation of AI innovation. We also explore the key bottlenecks limiting inference performance, how ElastixAI is tackling these issues, and why FPGAs are emerging as a compelling platform for accelerating large language model inference.

Stanislas Polu is Co-Founder & CTO of Dust — the enterprise AI agent platform used by 51,000 workers at 3,000+ companies. Before Dust, he spent three years on OpenAI's research team under Ilya Sutskever, working on mathematical reasoning in language models, and prior to that was an engineer at Stripe. He brings a rare combination of frontier AI research and product-building experience to the enterprise agent space.MCP, Agents & the $40M Bet on Multiplayer AI // MLOps Podcast #384 with Stanislas Polu, Co-Founder & CTO of Dust

Vamshi Ambati has spent more than two decades in AI, through the symbolic era, statistical era, and the neural wave we're experiencing today. A CMU PhD, founder of LatentStructure and Predera (which was acquired), now an investor at Virama Ventures, he's one of the sharper voices on what's actually happening under the hood of the AI boom.We discuss a simple question: Who wins when models become cheaper and more abundant? And try to answer this by looking at how inference spend v/s compute spend is shifting, and why inference may become the biggest infrastructure opportunity of the next decade.Vamshi explains what actually goes into the cost of a token, why AI is simultaneously getting cheaper and more expensive, and why the inference market alone could reach $1.3 trillion by 2030. If you're building in AI or someone who wants a clear mental model of where this industry is headed, this conversation is for you. 00:00 - Trailer0:45 - How an AI researcher thinks after 20 years05:53 - Where enterprise AI adoption is headed08:35 - Drawing parallels between cloud and AI11:20 - If building is cheap, what's valuable?13:37 - Can computing get cheaper?16:41 - What is inference, really?22:22 - Why coding and customer support got eaten first?26:48 - Which technologies are overvalued and undervalued?29:56 - An accidental entrepreneur's journey33:15 - Why is healthcare slow to adopt technology?38:59 - Landing Walmart as a customer42:36 - Should founders build in services if product isn't visible?43:47 - Is Palantir a product company or a services company?44:15 - How to win as a forward-deployed company46:23 - What it takes to land large enterprise customers49:20 - Building sales muscles as a technical founder-------------India's talent has built the world's tech—now it's time to lead it.This mission goes beyond startups. It's about shifting the center of gravity in global tech to include the brilliance rising from India.What is Neon Fund?We invest in seed and early-stage founders from India and the diaspora building world-class Enterprise AI companies. We bring capital, conviction, and a community that's done it before.Subscribe for real founder stories, investor perspectives, economist breakdowns, and a behind-the-scenes look at how we're doing it all at Neon.-------------Check us out on:Website: https://neon.fund/Instagram: https://www.instagram.com/theneonshoww/LinkedIn: https://www.linkedin.com/company/beneon/Twitter: https://x.com/TheNeonShowwConnect with Siddhartha on:LinkedIn: https://www.linkedin.com/in/siddharthaahluwalia/Twitter: https://x.com/siddharthaa7-------------This video is for informational purposes only. The views expressed are those of the individuals quoted and do not constitute professional advice.Send us Fan Mail

Roman Chernin is Co-Founder and Chief Business Officer of Nebius, one of the fastest-growing AI infrastructure companies in the world. Today, Nebius operates some of the largest AI compute clusters globally and serves leading AI labs, enterprises, and developers. Today, Nebius has a market cap of $57BN.  AGENDA:  00:00 — Why AI Infrastructure Is Not a Bubble 05:00 — The Real Impact of Open Source on OpenAI & Anthropic 11:00 — Jevons Paradox: Why Cheaper AI Creates More Demand 13:00 — The Four Layers of AI Infrastructure Explained 19:00 — If Nebius Had 10x More Capacity Tomorrow 26:00 — The Shift from Training to Inference and Agents 31:00 — How Token Factory Cuts AI Costs by 70% 44:00 — Sovereign AI, Europe, and the Future of Model Building 49:00 — Competing Against Hyperscalers with 10x More Capital 59:00 — The Biggest Threat to Nebius Isn't Competition—It's Consolidation    

✅ New autonomous agents. ✅ Canva designs made for you. ✅ Codex upgrades to make your business move. If you had your head down in spreadsheets this week, you missed some MAJOR AI upgrades that are available now. We track what's hot and what's not and break it all down on Fridays with our Friday Features. Autonomous Copilot agents, new Codex tools, Github CoPilot app and 7 more AI updates you should be using — An Everyday AI Chat with Jordan WilsonNewsletter: Sign up for our free daily newsletterMore on this Episode: Episode PageToday's Episode on LinkedIn: Thoughts on this? Join the convo on LinkedIn and connect with other AI leaders.Upcoming Episodes: Check out the upcoming Everyday AI Livestream lineupWebsite: YourEverydayAI.comEmail The Show: info@youreverydayai.comConnect with Jordan on LinkedInTopics Covered in This Episode:OpenAI Codex Role-Specific Plugins LaunchMicrosoft Build Conference AI Feature ReleasesChatGPT Memory and Business Account UpgradesMicrosoft Flash Image Model for PowerPointCanva Integrated with ChatGPT and CodexGitHub Copilot Standalone Desktop App PreviewMicrosoft Autopilot Always-On Work AgentsOpenAI Models Now Available on AWS BedrockCodex Sites: AI-Built Internal Web AppsTimestamps:00:00 OpenAI's big money moves03:47 Explaining role-specific plugins09:02 Microsoft's new image model release11:09 Microsoft's AI strategy and Canva update14:23 Canva integration with ChatGPT16:56 GitHub Copilot's new canvas feature20:46 AI token subscription changes24:42 AWS adds OpenAI models to Bedrock28:25 Introducing OpenAI's CodeX Sites Feature32:07 Launch of OpenAI's New Plug-in34:16 Overview of podcast structureKeywords: Autonomous copilot agents, Codex tools, GitHub Copilot app, OpenAI Codex, ChatGPT business accounts, OpenAI enterprise, Microsoft Build conference, Microsoft always-on agents, AWS AI updates, Canva plugin, ChatGPT memory upgrade, Windows Codex integration, Microsoft Flash model, Enterprise apps integration, Role-specific plugins, Sales data analytics, Product design AI, Creative production AI, Investment banking plugin, Public equity investing, Data analytics plugin, Workspace admins, App permissions, Role-aware work agent, Financial research automation, Microsoft image generation model, PowerPoint AI integration, OneDrive AI features, Visual design creation, Canva app for ChatGPT, Canva MCP server, Agentic context carry, Full screen design preview, GitHub Copilot desktop app, GitHub Copilot Canvas, Agent-native command center, Parallel agent work tree, Code app interface, Model options in GitHub, Token usage limits, Subscription token subsidizing, Anthropic token efficiency, Amazon Bedrock, GPT-4, GPT-4.5, Small language models, Token reckoning, Security governance, Inference engine, Code app sidebar, Codex Sites, Internal dashboards, Project trackers, Interactive web apps, Shareable AI apps, Enterprise data connectors, ChatGPT Canvas, Automated workflow, Workplace authentication, Creative briefs repository.Send Everyday AI and Jordan a text message. (We can't reply back unless you leave contact info) Start Here ▶️Not sure where to start when it comes to AI? Start with our Start Here Series. You can listen to the first drop -- Episode 691 -- or get free access to our Inner Cricle community and all episodes: StartHereSeries.com Also, here's a link to the entire series on a Spotify playlist. 

This lecture discusses the William Clifford's 1877 essay "The Ethics Of Belief", in which he makes and argued for the central claim "it is wrong always, everywhere, and for any one, to believe anything upon insufficient evidence." It focuses on the third section of his essay, titled "The Limits Of Inference" in which Clifford discusses conditions for having well-founded beliefs of matters we don't have direct experience of, for example matters of everyday life, science, or history. We inevitably rely upon the assumption that the future or present will resemble what we have experienced in the past To support my ongoing work, go to my Patreon site - www.patreon.com/sadler If you'd like to make a direct contribution, you can do so here - www.paypal.me/ReasonIO You can find over 4,000 philosophy videos in my main YouTube channel - www.youtube.com/user/gbisadler Get Clifford's The Ethics of Belief - https://amzn.to/41WkkYA

SUMMARY: After the first successful AI IPO of 2026, we dig into what makes the Cerebras WSE architecture unique in the market for fast inference. GUEST: Andy Hock, at Chief Strategy Officer at Cerebras AISHOW: 1033SHOW TRANSCRIPT: The Enterprise AI Show #1033 TranscriptSHOW VIDEO: https://youtu.be/ed2nVbOtZiASHOW SPONSORS:OutShift - “Scaling Out Superintelligence”  The Internet of Cognition architectureShareGate - ShareGate Protect. Microsoft 365 Governance, we got this!Nasuni - Activate your data for AI and request a demoSHOW NOTES:OpenAI announces 750MW partnership with CerebrasCerebras and AWS partnershipCerebras announces IPOTopic 1 - Welcome to the show. Tell us about your background, and what you focus on today. Topic 2 - For anyone that's not familiar with Cerebras, give us an overview of the company, and especially an overview on the Cerebras technologies (e.g. Wafer-Scale Engine).Topic 3 - Cerebras' WSE architecture is different from many of the GPU or GPU-like architectures in the market today. Centralized vs. distributed architectures always have their tradeoffs. Walk us through the technical and economic value of the Cerebras architecture.Topic 4 - Congratulations on the recent IPO (raised $5.55B). Let's use that as a point in time vs the previous planned IPO. How has the market changed in that timeframe, and how has the Cerebras position changed? Topic 5 - Cerebras (today) offer both WSE hardware, and Cerebras Cloud (API) - very different GTM paths. Can we expect both of those to stay top priorities, or have the market dynamics shifted such that the priorities shift more towards the WSE business - as we're seeing OpenAI, AWS and other engagements announced?Topic 6 - Is Cerebras a training and inference company, or are the economics of inference significantly different enough that it needs to be the sole focus of the company (for now)? Topic 7 - How much effort is it for any company to add support for the Cerebras chips if they have previously been using other architectures?Topic 8 - An IPO is a major milestone for any company, but the markets will now look for your future story. How do you see the AI market evolving over the next 2-5 years, and what are some things that people aren't understanding yet about how it will evolve?FEEDBACK?Email: show @ the enterprise ai show dot comeBluesky: @TheEntAIShow.bsky.socialTwitter/X: @TheEntAIShowInstagram: @TheEntAIShow

In this episode of Alexa's Input (AI), I sat down with Rob Shaw from Red Hat to talk about how AI inference evolved from a simple model serving problem into a large-scale distributed systems problem.We explored the infrastructure shifts behind modern LLM serving, including how vLLM and PagedAttention changed the economics and efficiency of inference, why KV cache management became one of the most important bottlenecks in production AI systems, and how orchestration layers like llm-d are emerging to coordinate distributed inference.We also discuss:how LLM inference differs from traditional model serving runtimesKV cache, prefix caching, and cache-aware routingwhy throughput and latency became major infrastructure challengeslong-context agents and repeated inference callsdistributed inference on Kubernetesintelligent routing, flow control, and load balancingprefill/decode disaggregationenterprise AI deployment realitiesvLLM has become one of the most important open-source projects in AI infrastructure, and llm-d represents a newer shift toward treating inference as a coordinated distributed system rather than just a single runtime problem.If you want to better understand the systems layer beneath modern AI applications, this episode is a deep dive into where inference infrastructure is heading next.General Podcast LinksWatch: ⁠⁠⁠⁠⁠⁠https://www.youtube.com/@alexa_griffith⁠⁠⁠⁠⁠⁠Read: ⁠⁠⁠⁠⁠⁠⁠⁠https://alexasinput.substack.com/⁠⁠⁠⁠⁠⁠⁠⁠Listen:⁠⁠ ⁠⁠https://creators.spotify.com/pod/profile/alexagriffith/⁠⁠⁠⁠More: ⁠⁠⁠⁠⁠⁠https://linktr.ee/alexagriffith⁠⁠⁠⁠⁠⁠Learn more about the host atWebsite: ⁠⁠⁠⁠⁠⁠https://alexagriffith.com/⁠⁠⁠⁠⁠⁠LinkedIn: ⁠⁠⁠⁠⁠⁠https://www.linkedin.com/in/alexa-griffith/⁠⁠⁠⁠⁠⁠Find out more about the guest at:LinkedIn: https://www.linkedin.com/in/robert-shaw-1a01399a/ Red Hat Articles: https://developers.redhat.com/author/robert-shawGithub: https://github.com/robertgshaw2-redhat ResourcesvLLM Website: https://vllm.ai/vLLM GitHub Repository: https://github.com/vllm-project/vllmllm-d Website: https://llm-d.ai/llm-d GitHub Repository - https://github.com/llm-d/llm-d KeywordsAI inference, VLLM, LMD, distributed inference, GPU optimization, open source AI, Kubernetes, multi-cluster deployment, AI infrastructure, enterprise AI AI infrastructure, Kubernetes, model optimization, speculative decoding, mixture of experts, AI deployment, performance tuning, AI systems, neural network scaling Key TopicsEvolution of vLLM and llm-dDistributed inference and routingGPU utilization and performance optimizationOpen source AI infrastructureEnterprise deployment challenges and solutions Standardization in Kubernetes for NIC exposurePerformance optimizations: quantization and speculative decodingMixture of experts architecture and parallelism strategiesFlow control and request scheduling in AI systemsEmerging hardware for AI inference, Cerebras processorReinforcement learning and AI system supportModular architecture of vLLM and ecosystem projects

Modern work can be frustrating and chaotic—if you don't have the right tools. From context engineering to multimodal search, go behind the scenes and hear how Dropbox engineers are building AI that actually understands you, so you can focus on the work that matters most. If you're new to Working Smarter, we've travelled from the F1 track to the bottom of a lake, and heard real stories from chefs, doctors, lawyers, and founders about how AI is helping them do more of what they love about their jobs. But in our third season, we're talking to the people behind the tools—the engineers and product leaders building helpful, time-saving AI features into the Dropbox experience you already know and trust. You'll hear all about their work on agents, inference, security, and, of course, how the people building AI use AI themselves. ~ ~ ~  Working Smarter is brought to you by Dropbox. Find, organize, and share your work—all in one place—with context-aware AI from Dropbox. You can listen to more episodes of Working Smarter on Apple Podcasts, Spotify, YouTube, Amazon Music, or wherever you get your podcasts. To read more stories and past interviews, visit workingsmarter.ai This show would not be possible without the talented team at Cosmic Standard: producer Ben Montoya, sound engineer Aja Simpson, technical director Jacob Winik, and executive producer Eliza Smith. Special thanks to our illustrator Fanny Luor, marketing consultant Meggan Ellingboe, and editorial support from Catie Keck.  Our theme song was composed by Doug Stuart.  Working Smarter is hosted by Matthew Braga. Thanks for listening!

Stewart Alsop sat down with Michael Shackelford to discuss their experiences building applications through vibe coding—the practice of using AI to create software without traditional programming expertise. Stewart, who runs the AI Whispers community in Buenos Aires and hosts the Crazy Wisdom podcast (with over 660 interviews), shared how he went from teaching people prompt engineering to building his own video conferencing software as a Riverside.fm replacement, while Michael opened up about his year-long journey creating Genrupt Inc, an AI-powered content generation tool for e-commerce sellers. The conversation covered everything from the decline in quality of Claude's reasoning capabilities and how Chinese companies used distillation attacks to copy Anthropic's models, to the importance of spaced repetition systems for managing knowledge in the age of LLMs, with both sharing battle-tested prompting strategies like asking AI to "explain it to me in genius terms" and using deep research queries to reverse engineer how competitors build their products.Show Notes:- Dan Martell's book "Buy Back Your Time" was mentioned as one of the best business books for thinking about life and business- Check out John Vervaeke's "Awakening from the Meaning Crisis" for understanding relevance realization and why AI fundamentally cannot determine what's relevant to humans without being toldTimestamps00:00 Michael discusses being exhausted from getting his app ready for launch, working nonstop with AI to prepare landing page for podcast traffic driving beta signups05:00 Stewart explains starting AI Whispers in Buenos Aires after leaving OpenAI vendor company, meeting early adopters like Torin who was building mind-reading EEG technology10:00 Discussion of how corporations resist AI adoption due to political games and job security fears while some companies use AI as excuse for pandemic-era layoffs15:00 Stewart describes teaching workshops on using LLMs as linguistic tools rather than coding tools, noting technical people often lack humanities background needed for prompting20:00 Explaining chatbot wrappers, API calls, and how Anthropic's reasoning quality declined after Chinese distillation attacks copied their secret sauce developed with philosophers25:00 Technical discussion of model training, fine-tuning versus RAG for new information, and different approaches to updating AI knowledge beyond initial training30:00 Stewart describes building podcast recording software to replace expensive Riverside, struggling with syncing audio and video files across different computer clocks35:00 Discussion of critical factors in vibe coding, discovering unknown technical requirements, and how AIs don't automatically reveal missing information40:00 Stewart's reverse engineering process using deep research function to study competitors' hiring and technology stacks, separating planning agents from coding agents45:00 Prompting techniques including "explain like I know everything" and using spaced repetition systems to capture valuable prompts and technical knowledge50:00 Michael explains his Generux app for generating ecommerce content using Amazon review data analysis to inform high-converting listing images and videos55:00 Discussion of founder mentality involving self-delusion about project timelines, Michael working nine-plus hours daily for nine months on app development60:00 Comparing Amazon's expert software to prosumer software approach, discussing distribution challenges and future robotics applications for customized products65:00 Stewart demonstrates spaced repetition app for memory improvement and knowledge retention, explaining relevance realization problem that AI agents cannot solve without embodimentKey Insights1. Stewart Alsop started AI Whisperers in Buenos Aires after leaving his role at Invisible Technologies, which was OpenAI's largest vendor for RLHF work. He noticed that machine learning engineers at tech companies lacked the humanities background needed to properly interact with large language models, which are fundamentally linguistic tools. This led him to create weekly workshops teaching non-technical people how to use AI effectively, running events every Thursday for two years straight. The group attracted intense geeks from the start and eventually led to Stewart speaking right after Vitalik Buterin at DevConnect, marking a significant milestone for the community.2. Large corporations are resistant to AI adoption due to multiple factors including political dynamics within organizations and employees fearing job loss. Many companies that grew during the pandemic are now using AI as an excuse to downsize when the real issue is inefficiency from rapid expansion. Stewart observed that even technical people in machine learning often don't understand how to properly use AI tools because they lack linguistic and humanities training. The fundamental problem is educational, requiring companies to train people how to use these new tools while those same people resist learning them.3. Vibe coding has evolved significantly with Claude Code being a game changer that reduced the technical barrier to entry. Before Claude Code, developers needed substantial technical knowledge to work through constant doom loops and debugging cycles. The success of coding AI tools stems from thirty years of testing infrastructure that provides clear yes or no feedback on whether code works. This infrastructure doesn't exist in the same way for manufacturing, science, and other fields, which is why software became the dominant area for AI assistance initially.4. Claude's quality degradation over recent months resulted from multiple factors including distillation attacks by Chinese companies who reverse engineered Anthropic's reasoning capabilities. Anthropic had hired philosophers, sociologists, and psychologists to develop exceptional reasoning in Claude 4.5, but this was expensive to run. When Chinese models like Kimi copied these capabilities at one tenth the cost, and when mainstream users flooded the platform before Anthropic's planned IPO, the company had to reduce quality to manage computational costs. This represents a significant loss for power users who relied on Claude's superior reasoning abilities.5. Stewart built a podcast recording application to replace Riverside because he needed API access to automate workflows, which Riverside wanted one thousand dollars monthly to provide. The technical challenge involves syncing audio and video from local recordings on multiple computers with different clocks through a server, then merging them so voices match lip movements. This problem requires understanding complex timing issues across different network conditions and file formats. Stewart has been working through AI psychosis for months on this FFMPEG pipeline problem, illustrating how vibe coding still requires building intuition about technical problems even without traditional coding knowledge.6. The transition from expert software to prosumer software represents a major opportunity for AI-enabled tools. Expert software like Photoshop, Blender, and terminal interfaces have extreme complexity that intimidates beginners, but AI is making these capabilities accessible through natural language. The reign of specialists is ending as generalists with broad knowledge and curiosity can now build complete applications by leveraging AI to fill technical gaps. This shift particularly benefits entrepreneurs and founders who specialize in getting into difficult situations and figuring them out, even when they originally thought tasks would be easier than they turned out to be.7. Building applications with AI requires accepting massive time investments beyond initial estimates and developing strategies for overcoming knowledge gaps. Michael estimated his ecommerce content generation app would take months but spent nearly a year working over nine hours daily, while Stewart spent months solving audio-video sync issues. Success requires using tools like deep research to understand how competitors solve problems, maintaining separate planning and coding agents, and learning to ask the right questions. The key insight is that vibe coders can achieve ninety percent of functionality independently, but the final ten percent often requires understanding specific technical concepts that AI cannot intuit without proper context and domain knowledge.

Amidst the increasing urgency of powering data centers, a new solution has entered the mix: send them out to sea. In this episode, Shayle speaks to Garth Sheldon-Coulson, co-founder and CEO of Panthalassa. The company is building 85-meter steel "nodes" – taller than Big Ben – that it deploys into the deep ocean. These untethered, self-propelled nodes harness wave energy to power AI clusters, then beam their data back to land via satellite. The technology isn't without its fair share of logistic complications, but it nonetheless offers a pathway to powering the AI boom that's largely independent from grid or fuel constraints. Shayle and Garth cover topics including: - The physics and mechanics that power Panthalassa's nodes - The significance of building an autonomous fleet - The energy generation waiting to be tapped in the open ocean - The logistics and unit economics behind scaling Panthalassa's technology - Why deep-sea compute is well-suited for long-running workloads like inference and reinforcement learning - Catalyst: AI scaling pathways: On grid, on edge, off grid, off planet - Catalyst: How to build more hydropower - Latitude Media: Are Thiel-funded floating data centers enough to make wave energy pencil?  - Open Circuit: Grid utilization vs expansion: The 100 GW debate - Latitude Media: What geothermal can learn from offshore wind's demise Credits: Hosted by Shayle Kann. Produced and edited by Max Savage Levenson. Original music and engineering by Sean Marquand. Stephen Lacey is our executive editor. Catalyst is brought to you by EnergyHub. EnergyHub helps utilities build next-generation virtual power plants that unlock reliable flexibility at every level of the grid. See how EnergyHub helps unlock the power of flexibility at scale, and deliver more value through cross-DER dispatch with their leading Edge DERMS platform, by visiting energyhub.com. Tune into Critical Capital, a brand new podcast from Crux and Latitude Studios. Hosted by Crux CEO Alfred Johnson, Critical Capital explores the interlocking forces powering clean and critical infrastructure. Join us every other Tuesday for in-depth conversations at the intersection of energy, government, finance, and global markets. Listen here, or wherever you get podcasts. Catalyst is brought to you by FischTank PR, an award-winning climate and energy tech, renewables, and sustainability-focused PR firm dedicated to elevating the work of both early-stage and established companies. Learn more about their PR approach and how they can support your company's messaging by visiting ⁠fischtankpr.com⁠.

“OpenAI has only two AI accelerator compute vendors in production today, Cerebras and Nvidia,” Cerebras CEO Andrew Feldman says. Four days after Cerebras went public, Feldman joined Bloomberg Intelligence's Kunjan Sobhani to discuss the company's next chapter and the rapidly shifting AI infrastructure landscape. Feldman breaks down the OpenAI deal, the strategic AWS partnership around disaggregated inference and why Cerebras believes fast inference is becoming the industry's defining battleground. He explains how Cerebras evolved from building the world's largest chip to operating one of the fastest inference platforms, why disaggregated inference could reshape hyperscale AI deployments and how the company is navigating power, memory and data-center constraints. The episode also explores the competitive landscape beyond GPUs and Feldman's broader perspective on the next phase of AI compute.

Ajit Ghuman is the co-founder of Monetizely, former VP of Product at Segment, and author of Price to Scale. In this episode, Ajit breaks down one of the biggest pricing challenges AI companies are about to face: what happens when software no longer supports employees — but starts replacing them entirely? If your company is building, pricing, or monetizing AI products, this episode will change how you think about per-seat pricing, buyer psychology, and the future of SaaS monetization.   Why you have to check out today's podcast: Understand why per-user pricing may stop working as AI agents increasingly replace human workflows inside software products. Learn Ajit Ghuman's 3-part "Agentic Pricing Spectrum" for evaluating AI products based on autonomy, operational scope, and output-to-cost dynamics. Discover why buyers are suddenly comfortable with tokens, credits, and bundled AI pricing — even when they don't fully understand what those units actually mean.   "Unless you understand what your market is, who your buyers are, what do they want... it's the only thing that I start with when I do any project." – Ajit Ghuman   Topics Covered: 02:02 – Why Pricing Became the Most Direct Link to Customer Value. How pricing became the clearest connection between products, value, and business strategy.  06:29 – The AI Pricing Problem Nobody Has Fully Solved Yet. Why AI is forcing SaaS companies to rethink seats, tokens, outcomes, and margins.  07:38 – "Zero Human Companies" and the End of Per-User Pricing. Ajit explores a future where AI agents replace entire job functions — and asks the terrifying question: what happens when there's no user left to charge for? 12:30 – Why Cursor Still Charges Per User (For Now). A fascinating breakdown of AI coding tools, human "anchors," and why most AI products still can't fully move to outcome-based pricing. 16:51 – The Coming AI Commoditization Wave. Why Ajit believes agentic AI companies could rise — and collapse — dramatically faster than traditional SaaS businesses. 23:07 – Why Buyers Suddenly Accept Tokens, Credits, and Weird AI Pricing. Ajit explains how ChatGPT normalized token-based pricing — even though most buyers still don't fully understand what they're paying for. 26:00 – The Real Reason AI Pricing Feels So Chaotic Right Now. Inference costs are dropping, users are disappearing, and pricing anchors keep shifting faster than companies can adapt. 29:35 – The One Pricing Principle That Still Matters in the AI Era.  Despite all the chaos around AI monetization, Ajit says successful pricing still starts with deeply understanding your buyers and their problems.   Key Takeaways: "The anchor is still the human… but the moment the human disappears, per-user pricing starts breaking." – Ajit Ghuman  "Agentic AI may compress 20 years of SaaS evolution into just a few years." – Ajit Ghuman    People / Resources Mentioned: Cursor — Used as a real-world example of current AI pricing models Harvey AI — Referenced as an example of high-value AI transformation inside the legal industry Anthropic — Mentioned in relation to inference models powering AI tools OpenAI — Referenced throughout the discussion on tokens and AI pricing behavior Salesforce — Discussed in relation to potential future shifts away from per-seat pricing Zoom — Used as an example of changing pricing priorities during growth stages   Connect with Ajit Ghuman: Website: https://www.getmonetizely.com/  LinkedIn: https://www.linkedin.com/in/ajitpalghuman/   Email: ajit@getmonetizely.com   Connect with Mark Stiving: LinkedIn: https://www.linkedin.com/in/stiving/ Email: mark@impactpricing.com  

On April 29th, the US Senate hosted a panel on the "existential threat" of AI and two of the four panelists worked for the Chinese government. One month earlier, Bernie Sanders and AOC introduced legislation imposing a federal moratorium on American AI data centers. On Bitcoin Policy Hour EP 38, Zack Cohen, Ken Egan, and Zack Shapiro unpack a new Bitcoin Policy Institute report by Sam Lyman exposing the CCP influence operation steering US AI policy. They also cover the Clarity Act vote in Senate Banking, the BRCA fight, and the Digital Asset Parity Act. Sam Lyman's BPI Report: https://www.btcpolicy.org/articles/foreign-influence-in-the-campaign-against-american-ai