Podcasts about language model

Our 210th episode with a summary and discussion of last week's big AI news! Recorded on 05/23/2025 Hosted by Andrey Kurenkov and Jeremie Harris. Feel free to email us your questions and feedback at contact@lastweekinai.com and/or hello@gladstone.ai Read out our text newsletter and comment on the podcast at https://lastweekin.ai/. Join our Discord here! https://discord.gg/nTyezGSKwP In this episode: Google's Gemini diffusion technology showcases significant improvements in speed and efficiency for generating text, potentially revolutionizing the auto-regressive generation paradigm. Anthropic activates AI Safety Level 3 protections for Claude Opus 4, implementing robust measures such as bug bounties, synthetic jailbreak data, and preliminary egress bandwidth controls to mitigate bio-risk threats. OpenAI responds to the California Attorney General, refuting claims by the not-for-private-gain coalition and defending their controversial restructuring plans amidst ongoing criticism. Mistral delays the release of its Llama 4 Behemoth model due to training challenges, while Meta faces similar obstacles in rolling out its large-scale AI models, signaling difficulties in reaching frontier level performance. Timestamps + Links: (00:00:00) Intro / Banter (00:01:43) News Preview Tools & Apps (00:02:58) Anthropic's new Claude 4 AI models can reason over many steps (00:09:58) Google Unveils A.I. Chatbot, Signaling a New Era for Search (00:14:04) Google rolls out Project Mariner, its web-browsing AI agent (00:16:40) Veo 3 can generate videos — and soundtracks to go along with them (00:21:26) Imagen 4 is Google's newest AI image generator (00:23:15) Google Meet is getting real-time speech translation (00:25:36) Google's new Jules AI agent will help developers fix buggy code (00:26:43) GitHub's new AI coding agent can fix bugs for you (00:28:50) Mistral's new Devstral model was designed for coding Applications & Business (00:29:53) OpenAI Unites With Jony Ive in $6.5 Billion Deal to Create A.I. Devices (00:36:10) OpenAI's planned data center in Abu Dhabi would be bigger than Monaco (00:41:18) LM Arena, the organization behind popular AI leaderboards, lands $100M (00:45:21) Nvidia CEO says next chip after H20 for China won't be from Hopper series (00:46:39) Google's Gemini AI app has 400M monthly active users (00:51:15) AI Servers: End demand intact, but rising gap between upstream build and system production (2025.5.18) Projects & Open Source (00:53:46) Meta Is Delaying the Rollout of Its Flagship AI Model Research & Advancements (00:57:53) Gemini Diffusion (01:03:07) Chain-of-Model Learning for Language Model (01:09:16) Seek in the Dark: Reasoning via Test-Time Instance-Level Policy Gradient in Latent Space (01:15:38) Two Experts Are All You Need for Steering Thinking: Reinforcing Cognitive Effort in MoE Reasoning Models Without Additional Training (01:20:16) Lessons from Defending Gemini Against Indirect Prompt Injections (01:23:35) How Fast Can Algorithms Advance Capabilities? (01:30:20) Reinforcement Learning Finetunes Small Subnetworks in Large Language Models Policy & Safety (01:31:12) Exclusive: What OpenAI Told California's Attorney General (01:38:25) Activating AI Safety Level 3 Protections

Listen Ad Free https://www.solgoodmedia.com - Listen to hundreds of audiobooks, thousands of short stories, and ambient sounds all ad free!

A Daily Chronicle of AI Innovations on February 24th 2025Listen at https://podcasts.apple.com/ca/podcast/ai-daily-news-feb-24-2025-grok-3-rebels-against-musk/id1684415169?i=1000695674722The development of home humanoid robots and AI-driven career tools demonstrates the increasing integration of AI into daily life. Concerns arise from incidents such as AI censorship, misuse of AI for surveillance, and the ethical implications of AI-generated recreations. Companies are making significant investments in AI infrastructure, whilst regulatory adjustments are being considered to encourage AI innovation. New tools can classify robots by their performance and allow for video analysis, indicating progress and an increasing use of AI. Overall, the chronicle paints a picture of rapid evolution alongside emerging challenges in the field.

Listen to this interview of Sakina Fatima, Research Fellow, University of Ottawa, Canada; and also, Taher Ghaleb, Assistant Professor, Trent University, Canada. We talk about the coauthored paper Flakify: A Black-Box, Language Model-based Predictor for Flaky Tests (TSE 2023). Taher Ghaleb : "With our RQs, it's not just a matter of there being a problem with flaky tests. I mean, every researcher in this area already know that flaky tests is a problem. So, when we talk about the problem in our paper — or specifically, about the motivation behind our RQs — it's not about the flaky tests, because that's just the core problem which we already know — but instead, we talk the problem with existing approaches to flaky tests." Learn more about your ad choices. Visit megaphone.fm/adchoices Support our show by becoming a premium member! https://newbooksnetwork.supportingcast.fm/new-books-network

Listen to this interview of Sakina Fatima, Research Fellow, University of Ottawa, Canada; and also, Taher Ghaleb, Assistant Professor, Trent University, Canada. We talk about the coauthored paper Flakify: A Black-Box, Language Model-based Predictor for Flaky Tests (TSE 2023). Taher Ghaleb : "With our RQs, it's not just a matter of there being a problem with flaky tests. I mean, every researcher in this area already know that flaky tests is a problem. So, when we talk about the problem in our paper — or specifically, about the motivation behind our RQs — it's not about the flaky tests, because that's just the core problem which we already know — but instead, we talk the problem with existing approaches to flaky tests." Learn more about your ad choices. Visit megaphone.fm/adchoices

Proprietary LLM builders need to experience a valuation haircut as open source LLMs take share from proprietary LLMs. Proprietary LLM builders (OpenAI, Anthropic, Google, Microsoft, Amazon), have enjoyed lofty valuations over the past several years. Given the rise of open source competitors - which are on par with proprietary models from a performance standpoint and can be operated at a fraction of the cost - the proprietary model builders should suffer a valuation haircut. I believe that open source LLM builders such as DeepSeek and META will win the day and that 80% of LLMs and SLMs in production 5 years from now will be open source language models. https://open.substack.com/pub/tek2day/p/valuation-haircut-is-due-for-proprietary?r=1rp1p&utm_campaign=post&utm_medium=web&showWelcomeOnShare=false

A 671B parameter Mixture-of-Experts language model. It highlights the model's architecture, including its innovative load balancing and multi-token prediction strategies, and its efficient training process using FP8 precision. Benchmark results demonstrate DeepSeek-V3's strong performance compared to other open-source and some closed-source models, particularly in math and code tasks. The document also provides instructions for running DeepSeek-V3 locally using various frameworks and hardware, including NVIDIA and AMD GPUs and Huawei Ascend NPUs. Finally, licensing and contact information are included.

On today's episode, artificial intelligence with sponsor Selector.AI. If you're curious and maybe still skeptical about the value AI brings to network operations, listen to this episode. Selector is on the forefront of AIOps for networking, building models that are customized and specifically targeted at networks. What Selector is doing is NOT simply the low-hanging... Read more »

On today's episode, artificial intelligence with sponsor Selector.AI. If you're curious and maybe still skeptical about the value AI brings to network operations, listen to this episode. Selector is on the forefront of AIOps for networking, building models that are customized and specifically targeted at networks. What Selector is doing is NOT simply the low-hanging... Read more »

On today's episode, artificial intelligence with sponsor Selector.AI. If you're curious and maybe still skeptical about the value AI brings to network operations, listen to this episode. Selector is on the forefront of AIOps for networking, building models that are customized and specifically targeted at networks. What Selector is doing is NOT simply the low-hanging... Read more »

In this podcast, Namee Oberst, co-founder of AI Bloks, the company behind AI framework LLMWare, discusses the recent trend in Generative AI and Language Model technologies, the Small Language Models (SLMs) and how these smaller models are empowering the edge computing on devices and enabling AI-powered PC's. Read a transcript of this interview: https://bit.ly/3O9LZOZ Subscribe to the Software Architects' Newsletter for your monthly guide to the essential news and experience from industry peers on emerging patterns and technologies: https://www.infoq.com/software-architects-newsletter Upcoming Events: QCon San Francisco (November 18-22, 2024) Get practical inspiration and best practices on emerging software trends directly from senior software developers at early adopter companies. https://qconsf.com/ QCon London (April 7-9, 2025) Discover new ideas and insights from senior practitioners driving change and innovation in software development. https://qconlondon.com/ Save the date: InfoQ Dev Summit Boston (June 9-10, 2025) Actionable insights on today's critical dev priorities. devsummit.infoq.com/conference/boston2025 The InfoQ Podcasts: Weekly inspiration to drive innovation and build great teams from senior software leaders. Listen to all our podcasts and read interview transcripts: - The InfoQ Podcast https://www.infoq.com/podcasts/ - Engineering Culture Podcast by InfoQ https://www.infoq.com/podcasts/#engineering_culture - Generally AI: https://www.infoq.com/generally-ai-podcast/ Follow InfoQ: - Mastodon: https://techhub.social/@infoq - Twitter: twitter.com/InfoQ - LinkedIn: www.linkedin.com/company/infoq - Facebook: bit.ly/2jmlyG8 - Instagram: @infoqdotcom - Youtube: www.youtube.com/infoq Write for InfoQ: Learn and share the changes and innovations in professional software development. - Join a community of experts. - Increase your visibility. - Grow your career. https://www.infoq.com/write-for-infoq

Tue, 22 Oct 2024 16:00:00 GMT http://relay.fm/focused/215 http://relay.fm/focused/215 David Sparks and Mike Schmitz David & Mike revisit the topic of AI to consider what's changed in the last year and discuss how it can be helpful for those of us who strive to live a focused life. David & Mike revisit the topic of AI to consider what's changed in the last year and discuss how it can be helpful for those of us who strive to live a focused life. clean 5044 David & Mike revisit the topic of AI to consider what's changed in the last year and discuss how it can be helpful for those of us who strive to live a focused life. This episode of Focused is sponsored by: Zocdoc: Find the right doctor, right now with Zocdoc. Sign up for free. ExpressVPN: High-Speed, Secure & Anonymous VPN Service. Get an extra three months free. Indeed: Join more than 3.5 million businesses worldwide using Indeed to hire great talent fast. Links and Show Notes: Deep Focus: Extended ad-free episodes with bonus deep dive content. The 2025 FOCUSED Calendar | NeuYear.net Focused #189: Focus & AI Liminal Thinking by Dave Gray Co-Intelligence by Ethan Mollick Meet the Guys Dating A.I. Bots | Apple News Spiral Delphi Deep Work by Cal Newport Whisper Memos Otter.ai SaneBox Grammarly A System for Writing by Bob Doto Focused #209: A System for Writing, with Bob Doto RØDECaster Video | Video and Audio Production Console Gridfinity

Tue, 22 Oct 2024 16:00:00 GMT http://relay.fm/focused/215 http://relay.fm/focused/215 The Sparkly Language Model 215 David Sparks and Mike Schmitz David & Mike revisit the topic of AI to consider what's changed in the last year and discuss how it can be helpful for those of us who strive to live a focused life. David & Mike revisit the topic of AI to consider what's changed in the last year and discuss how it can be helpful for those of us who strive to live a focused life. clean 5044 David & Mike revisit the topic of AI to consider what's changed in the last year and discuss how it can be helpful for those of us who strive to live a focused life. This episode of Focused is sponsored by: Zocdoc: Find the right doctor, right now with Zocdoc. Sign up for free. ExpressVPN: High-Speed, Secure & Anonymous VPN Service. Get an extra three months free. Indeed: Join more than 3.5 million businesses worldwide using Indeed to hire great talent fast. Links and Show Notes: Deep Focus: Extended ad-free episodes with bonus deep dive content. The 2025 FOCUSED Calendar | NeuYear.net Focused #189: Focus & AI Liminal Thinking by Dave Gray Co-Intelligence by Ethan Mollick Meet the Guys Dating A.I. Bots | Apple News Spiral Delphi Deep Work by Cal Newport Whisper Memos Otter.ai SaneBox Grammarly A System for Writing by Bob Doto Focused #209: A System for Writing, with Bob Doto RØDECaster Video | Video and Audio Production Console Gridfinity Meditat

Send us a textMore on GenAI, Hallucinations, RAG, Use Cases, LLMs, SLMs and costs with Armand Ruiz, Director watsonx Client Engineering and John Webb, Principal Client Engineering.  With this and the previous episode you'll be wiser on AI than 98% of the world.00:12 Hallucinations02:33 RAG Differentiation06:41 Why IBM in AI09:23 Use Cases11:02 The GenAI Resume13:37 watson.x 15:40 LLMs17:51 Experience Counts20:03 AI that Surprises23:46 AI Skills26:47 Switching LLMs27:13 The Cost and SLMs28:21 Prompt Engineering29:16 For FunLinkedIn: linkedin.com/in/armand-ruiz, linkedin.com/in/john-webb-686136127 Website: https://www.ibm.com/client-engineeringLove what you're hearing? Don't forget to rate us on your favorite platform!Want 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. Want 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.Want 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 a textMore on GenAI, Hallucinations, RAG, Use Cases, LLMs, SLMs and costs with Armand Ruiz, Director watsonx Client Engineering and John Webb, Principal Client Engineering.  With this and the previous episode you'll be wiser on AI than 98% of the world.00:12 Hallucinations02:33 RAG Differentiation06:41 Why IBM in AI09:23 Use Cases11:02 The GenAI Resume13:37 watson.x 15:40 LLMs17:51 Experience Counts20:03 AI that Surprises23:46 AI Skills26:47 Switching LLMs27:13 The Cost and SLMs28:21 Prompt Engineering29:16 For FunLinkedIn: linkedin.com/in/armand-ruiz, linkedin.com/in/john-webb-686136127 Website: https://www.ibm.com/client-engineeringLove what you're hearing? Don't forget to rate us on your favorite platform!Want 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. Want 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.Want 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.

Today, we're joined by Nicholas Carlini, research scientist at Google DeepMind to discuss adversarial machine learning and model security, focusing on his 2024 ICML best paper winner, “Stealing part of a production language model.” We dig into this work, which demonstrated the ability to successfully steal the last layer of production language models including ChatGPT and PaLM-2. Nicholas shares the current landscape of AI security research in the age of LLMs, the implications of model stealing, ethical concerns surrounding model privacy, how the attack works, and the significance of the embedding layer in language models. We also discuss the remediation strategies implemented by OpenAI and Google, and the future directions in the field of AI security. Plus, we also cover his other ICML 2024 best paper, “Position: Considerations for Differentially Private Learning with Large-Scale Public Pretraining,” which questions the use and promotion of differential privacy in conjunction with pre-trained models. The complete show notes for this episode can be found at https://twimlai.com/go/702.

Recent advancements in audio generation have been significantly propelled by the capabilities of Large Language Models (LLMs). The existing research on audio LLM has primarily focused on enhancing the architecture and scale of audio language models, as well as leveraging larger datasets, and generally, acoustic codecs, such as EnCodec, are used for audio tokenization. However, these codecs were originally designed for audio compression, which may lead to suboptimal performance in the context of audio LLM. Our research aims to address the shortcomings of current audio LLM codecs, particularly their challenges in maintaining semantic integrity in generated audio. For instance, existing methods like VALL-E, which condition acoustic token generation on text transcriptions, often suffer from content inaccuracies and elevated word error rates (WER) due to semantic misinterpretations of acoustic tokens, resulting in word skipping and errors. To overcome these issues, we propose a straightforward yet effective approach called X-Codec. X-Codec incorporates semantic features from a pre-trained semantic encoder before the Residual Vector Quantization (RVQ) stage and introduces a semantic reconstruction loss after RVQ. By enhancing the semantic ability of the codec, X-Codec significantly reduces WER in speech synthesis tasks and extends these benefits to non-speech applications, including music and sound generation. Our experiments in text-to-speech, music continuation, and text-to-sound tasks demonstrate that integrating semantic information substantially improves the overall performance of language models in audio generation. Our code and demo are available (Demo: https://x-codec-audio.github.io Code: https://github.com/zhenye234/xcodec) 2024: Zhen Ye, Peiwen Sun, Jiahe Lei, Hongzhan Lin, Xu Tan, Zheqi Dai, Qiuqiang Kong, Jianyi Chen, Jiahao Pan, Qi-fei Liu, Yi-Ting Guo, Wei Xue https://arxiv.org/pdf/2408.17175

This episode dives into Salesforce's innovative AI sales agents that automate tasks but risk losing human touch, NVIDIA's compact yet powerful language model that promises efficiency, groundbreaking research showing how optimized computation can enhance model performance, and insights into compound inference systems revealing the delicate balance in maximizing language model effectiveness. Contact:  sergi@earkind.com Timestamps: 00:34 Introduction 01:49 Salesforce's New Sales AI Agents 03:09 Lightweight Champ: NVIDIA Releases Small Language Model With State-of-the-Art Accuracy 04:52 avante.nvim 05:56 Fake sponsor 07:45 Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters 09:22 Large Language Monkeys: Scaling Inference Compute with Repeated Sampling 11:15 Are More LLM Calls All You Need? Towards Scaling Laws of Compound Inference Systems 13:10 Outro

How do we figure out what large language models believe? In fact, do they even have beliefs? Do those beliefs have locations, and if so, can we edit those locations to change the beliefs? Also, how are we going to get AI to perform tasks so hard that we can't figure out if they succeeded at them? In this episode, I chat to Peter Hase about his research into these questions. Patreon: patreon.com/axrpodcast Ko-fi: ko-fi.com/axrpodcast The transcript: https://axrp.net/episode/2024/08/24/episode-35-peter-hase-llm-beliefs-easy-to-hard-generalization.html   Topics we discuss, and timestamps: 0:00:36 - NLP and interpretability 0:10:20 - Interpretability lessons 0:32:22 - Belief interpretability 1:00:12 - Localizing and editing models' beliefs 1:19:18 - Beliefs beyond language models 1:27:21 - Easy-to-hard generalization 1:47:16 - What do easy-to-hard results tell us? 1:57:33 - Easy-to-hard vs weak-to-strong 2:03:50 - Different notions of hardness 2:13:01 - Easy-to-hard vs weak-to-strong, round 2 2:15:39 - Following Peter's work   Peter on Twitter: https://x.com/peterbhase   Peter's papers: Foundational Challenges in Assuring Alignment and Safety of Large Language Models: https://arxiv.org/abs/2404.09932 Do Language Models Have Beliefs? Methods for Detecting, Updating, and Visualizing Model Beliefs: https://arxiv.org/abs/2111.13654 Does Localization Inform Editing? Surprising Differences in Causality-Based Localization vs. Knowledge Editing in Language Models: https://arxiv.org/abs/2301.04213 Are Language Models Rational? The Case of Coherence Norms and Belief Revision: https://arxiv.org/abs/2406.03442 The Unreasonable Effectiveness of Easy Training Data for Hard Tasks: https://arxiv.org/abs/2401.06751   Other links: Toy Models of Superposition: https://transformer-circuits.pub/2022/toy_model/index.html Interpretability Beyond Feature Attribution: Quantitative Testing with Concept Activation Vectors (TCAV): https://arxiv.org/abs/1711.11279 Locating and Editing Factual Associations in GPT (aka the ROME paper): https://arxiv.org/abs/2202.05262 Of nonlinearity and commutativity in BERT: https://arxiv.org/abs/2101.04547 Inference-Time Intervention: Eliciting Truthful Answers from a Language Model: https://arxiv.org/abs/2306.03341 Editing a classifier by rewriting its prediction rules: https://arxiv.org/abs/2112.01008 Discovering Latent Knowledge Without Supervision (aka the Collin Burns CCS paper): https://arxiv.org/abs/2212.03827 Weak-to-Strong Generalization: Eliciting Strong Capabilities With Weak Supervision: https://arxiv.org/abs/2312.09390 Concrete problems in AI safety: https://arxiv.org/abs/1606.06565 Rissanen Data Analysis: Examining Dataset Characteristics via Description Length: https://arxiv.org/abs/2103.03872   Episode art by Hamish Doodles: hamishdoodles.com

While Large Language Models (LLMs) are the dominant models for generative tasks in language, they do not perform as well as diffusion models on image and video generation. To effectively use LLMs for visual generation, one crucial component is the visual tokenizer that maps pixel-space inputs to discrete tokens appropriate for LLM learning. In this paper, we introduce MAGVIT-v2, a video tokenizer designed to generate concise and expressive tokens for both videos and images using a common token vocabulary. Equipped with this new tokenizer, we show that LLMs outperform diffusion models on standard image and video generation benchmarks including ImageNet and Kinetics. In addition, we demonstrate that our tokenizer surpasses the previously top-performing video tokenizer on two more tasks: (1) video compression comparable to the next-generation video codec (VCC) according to human evaluations, and (2) learning effective representations for action recognition tasks. 2023: Lijun Yu, José Lezama, N. B. Gundavarapu, Luca Versari, Kihyuk Sohn, David C. Minnen, Yong Cheng, Agrim Gupta, Xiuye Gu, Alexander G. Hauptmann, Boqing Gong, Ming-Hsuan Yang, Irfan Essa, David A. Ross, Lu Jiang https://arxiv.org/pdf/2310.05737

Last year, the venture capitalist Marc Andreesen published a document he called “The Techno-Optimist Manifesto.” In it, he argued that “everything good is downstream of growth,” government regulation is bad, and that the only way to achieve real progress is through technology.Of course, Silicon Valley has always been driven by libertarian sensibilities and an optimistic view of technology. But the radical techno-optimism of people like Andreesen, and billionaire entrepreneurs like Peter Thiel and Elon Musk, has morphed into something more extreme. In their view, technology and government are always at odds with one another.But if that's true, then how do you explain someone like Audrey Tang?Tang, who, until May of this year, was Taiwan's first Minister of Digital Affairs, is unabashedly optimistic about technology. But she's also a fervent believer in the power of democratic government.To many in Silicon Valley, this is an oxymoron. But Tang doesn't see it that way. To her, technology and government are – and have always been – symbiotic.So I wanted to ask her what a technologically enabled democracy might look like – and she has plenty of ideas. At times, our conversation got a little bit wonky. But ultimately, this is a conversation about a better, more inclusive form of democracy. And why she thinks technology will get us there.Just a quick note: we recorded this interview a couple of months ago, while Tang was still the Minister of Digital Affairs.Mentions:“vTaiwan”“Polis”“Plurality: The Future of Collaborative Technology and Democracy” by E. Glen Weyl, Audrey Tang and ⿻ Community“Collective Constitutional AI: Aligning a Language Model with Public Input,” AnthropicFurther Reading:“The simple but ingenious system Taiwan uses to crowdsource its laws” by Chris Horton“How Taiwan's Unlikely Digital Minister Hacked the Pandemic” by Andrew Leonard

This week, we discuss the CrowdStrike outage, FinOps data exports, and the state of open-source forks. Plus, Matt shares some exciting exclusive news about his future! Watch the YouTube Live Recording of Episode (https://www.youtube.com/watch?v=hYoFk0K_XpI) 477 (https://www.youtube.com/watch?v=hYoFk0K_XpI) Runner-up Titles Matt Ray Explains Channel File 291 Documenting CYA An intern did it Default lifestyle strikes again All the Nelson GIFs Rundown CrowdStrike Huge Microsoft Outage Linked to CrowdStrike Takes Down Computers Around the World (https://www.wired.com/story/microsoft-windows-outage-crowdstrike-global-it-probems/) 12-hour timelapse of airline traffic after what was likely the biggest IT outage in history (https://x.com/US_Stormwatch/status/1814268813879206397) Flights grounded and offices hit as internet users face disruptions (https://apnews.com/live/internet-global-outage-crowdstrike-microsoft-downtime) TODAY (@TODAYshow) on X (https://x.com/TODAYshow/status/1814266372882391523?ref_src=twsrc%5Egoogle%7Ctwcamp%5Eserp%7Ctwgr%5Etweet) George Kurtz (@George_Kurtz) on X (https://x.com/George_Kurtz/status/1814316045185822981) CrowdStrike's Global Outage Doesn't Have to Be a Recurring Nightmare (https://www.bloomberg.com/opinion/articles/2024-07-19/crowdstrike-s-nightmare-it-microsoft-outage-shouldn-t-be-normal?srnd=homepage-americas) Heard on the Street: CrowdStrike May Get More Than a Slap (https://www.wsj.com/livecoverage/stock-market-today-dow-sp500-nasdaq-live-07-19-2024/card/heard-on-the-street-crowdstrike-may-get-more-than-a-slap-CbyAd5zi7ELT4miAZHNV) What Happened to Digital Resilience? (https://www.nytimes.com/2024/07/19/us/politics/crowdstrike-outage.html?unlocked_article_code=1.8k0._ZDj.e5unf_bqIJNo&smid=url-share) SolarWinds Defeats Part of SEC's Fraud Case Over Hack (https://www.wsj.com/articles/solarwinds-defeats-part-of-secs-fraud-case-over-hack-ec69169a) Technical Details: Falcon Update for Windows Hosts (https://www.crowdstrike.com/blog/falcon-update-for-windows-hosts-technical-details/) Microsoft tried to get AV vendors to use APIs (https://www.threads.net/@sbisson/post/C9pIIYmo19q?xmt=AQGzVYTNKy9-De3zRXlIsl7QNqarqWsTWlmD_4Wc-7MM2A) House committee calls on CrowdStrike CEO to testify on global outage (https://www.washingtonpost.com/technology/2024/07/22/house-committee-calls-crowdstrike-ceo-testify-global-outage/) Crashes and Competition (https://stratechery.com/2024/crashes-and-competition/) The CrowdStrike Failure Was a Warning (https://www.theatlantic.com/ideas/archive/2024/07/crowdstrike-failure-warning-solutions/679174/) Defective McAfee update causes worldwide meltdown of XP PCs (https://www.zdnet.com/article/defective-mcafee-update-causes-worldwide-meltdown-of-xp-pcs/?utm_source=newsletter&utm_medium=email&utm_campaign=newsletter_axioscodebook&stream=top) CrowdStrike broke Debian and Rocky Linux months ago, but no one noticed (https://www.neowin.net/news/crowdstrike-broke-debian-and-rocky-linux-months-ago-but-no-one-noticed/#google_vignette) CrowdStrike Update: Latest News, Lessons Learned from a Retired Microsoft Engineer (https://youtu.be/ZHrayP-Y71Q?si=AmavOuoU_IjGMTFi) CrowdStrike offers a $10 apology gift card to say sorry for outage (https://techcrunch.com/2024/07/24/crowdstrike-offers-a-10-apology-gift-card-to-say-sorry-for-outage/) Announcing Data Exports for FOCUS 1.0 (Preview) in AWS Billing and Cost Management (https://aws.amazon.com/blogs/aws-cloud-financial-management/announcing-data-exports-for-focus-1-0-preview-in-aws-billing-and-cost-management/) Wiz walks away from $23 billion deal with Google, will pursue IPO (https://www.cnbc.com/2024/07/23/google-wiz-deal-dead.html) Import and export Markdown in Google Docs (http://workspaceupdates.googleblog.com/2024/07/import-and-export-markdown-in-google-docs.html) Google URL Shortener links will no longer be available (https://developers.googleblog.com/en/google-url-shortener-links-will-no-longer-be-available/) The Post-Valkey World (https://redmonk.com/sogrady/2024/07/16/post-valkey-world/) A tale of two forks - comparing Valkey/Redis and OpenTofu/Terraform! (https://www.linkedin.com/posts/danlorenc_oss-opensource-community-activity-7221488717704609792-U2SR/?utm_source=share&utm_medium=member_desktop) Datadog rumoured to be sniffing round GitLab as tech M&A market heats up (https://www.thestack.technology/datadog-rumoured-to-be-sniffing-round-gitlab-as-tech-m-a-market-heats-up/) Google-Backed Software Developer GitLab Eyes Sale, Reuters Says (https://www.bloomberg.com/news/articles/2024-07-17/google-backed-software-developer-gitlab-eyes-sale-reuters-says) Relevant to your Interests Google Open Sources 27B Parameter Gemma 2 Language Model (https://www.infoq.com/news/2024/07/google-gemma-2/) What It Really Takes to Build an AI Datacenter (https://www.bloomberg.com/news/articles/2024-06-21/what-it-really-takes-to-build-an-ai-datacenter) State of Developer Experience 2024 (https://newsletter.getdx.com/p/state-of-developer-experience-2024?r=2d4o&utm_campaign=post&utm_medium=web) The Return-to-Office Productivity Argument Is Over (https://www.inc.com/joe-procopio/the-return-to-office-productivity-argument-is-over.html) A new path for Privacy Sandbox on the web (https://privacysandbox.com/news/privacy-sandbox-update) The search for the random numbers that run our lives (https://www.bbc.com/future/article/20240704-the-search-for-the-random-numbers-that-run-our-lives) OpenAI is releasing a cheaper, smarter model (https://www.theverge.com/2024/7/18/24200714/openai-new-cheaper-smarter-model-gpt-4o-mini) Microsoft unveils a large language model that excels at encoding spreadsheets (https://www.thestack.technology/microsoft-llm-spreadsheet-llm/) Maestro: Netflix's Workflow Orchestrator (https://netflixtechblog.com/maestro-netflixs-workflow-orchestrator-ee13a06f9c78) IBM shares jump on earnings and revenue beat (https://www.cnbc.com/2024/07/24/ibm-q2-earnings-report-2024.html) US banks to begin reporting Russian assets for eventual forfeiture under new law (https://apnews.com/article/repo-act-banks-russia-ukraine-russian-assets-9ecda7e3e799cdbfb564844ae89a144b) Nonsense Darden Restaurants (NYSE: DRI) agreed to buy Tex-Mex chain Chuy's (https://www.axios.com/newsletters/axios-pro-rata-0bb0be2c-41d0-4181-bf39-ba3e827303da.html?chunk=1&utm_term=emshare#story1) Leadership within a costco warehouse (https://www.tiktok.com/t/ZTNaLxKJg/) If The Office took place at a car dealership (https://x.com/milkkarten/status/1813968113526067449?s=46&t=zgzybiDdIcGuQ_7WuoOX0A) Type in Morse code by repeatedly slamming your laptop shut (https://github.com/veggiedefender/open-and-shut) Sponsor SysAid – Next-Gen IT Service Management: (https://www.sysaid.com/lp/sysaid-copilot-l?utm_source=podcast&utm_medium=cpc&utm_campaign=software%20define) Experience the only platform with generative AI embedded in every aspect of IT management, enabling you to deliver exceptional service effortlessly and automagically. Listener Feedback (#asksdt) Foundation Models - IBM watsonx.ai (https://www.ibm.com/products/watsonx-ai/foundation-models) Conferences DevOpsDays Birmingham (https://devopsdays.org/events/2024-birmingham-al/welcome/), Aug 19-21, 2024 SpringOne (https://springone.io/?utm_source=cote&utm_campaign=devrel&utm_medium=newsletter&utm_content=newsletterUpcoming)/VMware Explore US (https://blogs.vmware.com/explore/2024/04/23/want-to-attend-vmware-explore-convince-your-manager-with-these/?utm_source=cote&utm_campaign=devrel&utm_medium=newsletter&utm_content=newsletterUpcoming), Aug 26-29, 2024 DevOpsDays Antwerp (https://devopsdays.org/events/2024-antwerp/welcome/), Sept 4–5, 2024, 15th anniversary SREday London 2024 (https://sreday.com/2024-london/), Sept 19–20, 2024 Coté speaking, 20% off with the code SRE20DAY (https://sreday.com/2024-london/#tickets) SDT News & Community Join our Slack community (https://softwaredefinedtalk.slack.com/join/shared_invite/zt-1hn55iv5d-UTfN7mVX1D9D5ExRt3ZJYQ#/shared-invite/email), post questions in #asksdt (https://softwaredefinedtalk.slack.com/archives/C07CSP19GAH) Email the show: questions@softwaredefinedtalk.com (mailto:questions@softwaredefinedtalk.com) Free stickers: Email your address to stickers@softwaredefinedtalk.com (mailto:stickers@softwaredefinedtalk.com) Follow us on social media: Twitter (https://twitter.com/softwaredeftalk), Threads (https://www.threads.net/@softwaredefinedtalk), Mastodon (https://hachyderm.io/@softwaredefinedtalk), LinkedIn (https://www.linkedin.com/company/software-defined-talk/), BlueSky (https://bsky.app/profile/softwaredefinedtalk.com) Watch us on: Twitch (https://www.twitch.tv/sdtpodcast), YouTube (https://www.youtube.com/channel/UCi3OJPV6h9tp-hbsGBLGsDQ/featured), Instagram (https://www.instagram.com/softwaredefinedtalk/), TikTok (https://www.tiktok.com/@softwaredefinedtalk) Book offer: Use code SDT for $20 off "Digital WTF" by Coté (https://leanpub.com/digitalwtf/c/sdt) Sponsor (https://www.softwaredefinedtalk.com/ads) the show (https://www.softwaredefinedtalk.com/ads) Recommendations Brandon: Austin FC (https://www.austinfc.com/competitions/mls-regular-season/2024/matches/atxvssea-07-13-2024/) Presumed Innocent (https://www.google.com/url?sa=t&source=web&rct=j&opi=89978449&url=https://tv.apple.com/us/show/presumed-innocent/umc.cmc.5hnqrhwtzt3esr7rb1wq2ppvn&ved=2ahUKEwiClKPk28CHAxWXLUQIHd59CCoQFnoECEcQAQ&usg=AOvVaw20AOPkQVtwWO77Jomxeua0) The Contrarian (https://www.penguinrandomhouse.com/books/609711/the-contrarian-by-max-chafkin/) Matt: Crying Out Cloud (https://www.wiz.io/crying-out-cloud) podcast Photo Credits Artwork (https://unsplash.com/photos/a-computer-screen-with-a-blue-screen-on-it-t_IkF_CNvSY)

Chaining language model (LM) calls as composable modules is fueling a new way of programming, but ensuring LMs adhere to important constraints requires heuristic “prompt engineering.” The paper this week introduces LM Assertions, a programming construct for expressing computational constraints that LMs should satisfy. The researchers integrated their constructs into the recent DSPy programming model for LMs and present new strategies that allow DSPy to compile programs with LM Assertions into more reliable and accurate systems. They also propose strategies to use assertions at inference time for automatic self-refinement with LMs. They reported on four diverse case studies for text generation and found that LM Assertions improve not only compliance with imposed rules but also downstream task performance, passing constraints up to 164% more often and generating up to 37% more higher-quality responses.We discuss this paper with Cyrus Nouroozi, DSPY key contributor. To learn more about ML observability, join the Arize AI Slack community or get the latest on our LinkedIn and Twitter.

Where adapting LLMs to specialized domains is essential (e.g., recent news, enterprise private documents), we discuss a paper that asks how we adapt pre-trained LLMs for RAG in specialized domains. SallyAnn DeLucia is joined by Sai Kolasani, researcher at UC Berkeley's RISE Lab (and Arize AI Intern), to talk about his work on RAFT: Adapting Language Model to Domain Specific RAG. RAFT (Retrieval-Augmented FineTuning) is a training recipe that improves an LLM's ability to answer questions in a “open-book” in-domain settings. Given a question, and a set of retrieved documents, the model is trained to ignore documents that don't help in answering the question (aka distractor documents). This coupled with RAFT's chain-of-thought-style response, helps improve the model's ability to reason. In domain-specific RAG, RAFT consistently improves the model's performance across PubMed, HotpotQA, and Gorilla datasets, presenting a post-training recipe to improve pre-trained LLMs to in-domain RAG. Read it on the blog: https://arize.com/blog/raft-adapting-language-model-to-domain-specific-rag/To learn more about ML observability, join the Arize AI Slack community or get the latest on our LinkedIn and Twitter.

Improve search capabilities for your PostgreSQL-backed applications using vector search and embeddings generated in under 10 milliseconds without sending data outside your PostgreSQL instance. Integrate real-time translation, sentiment analysis, and advanced AI functionalities securely within your database environment with Azure Local AI and Azure AI Service. Combine the Azure Local AI extension with the Azure AI extension to maximize the potential of AI-driven features in your applications, such as semantic search and real-time data translation, all while maintaining data security and efficiency. Joshua Johnson, Principal Technical PM for Azure Database for PostgreSQL, demonstrates how you can reduce latency and ensure predictable performance by running locally deployed models, making it ideal for highly transactional applications.   ► QUICK LINKS: 00:00 - Improve search for PostgreSQL 01:21 - Increased speed 02:47 - Plain text descriptive query 03:20 - Improve search results 04:57 - Semantic search with vector embeddings 06:10 - Test it out 06:41 - Azure local AI extension with Azure AI Service 07:39 - Wrap up   ► Link References Check out our previous episode on Azure AI extension at https://aka.ms/PGAIMechanics  Get started with Azure Database for PostgreSQL - Flexible Server at https://aka.ms/postgresql  To stay current with all the updates, check out our blog at https://aka.ms/azurepostgresblog   ► Unfamiliar with Microsoft Mechanics?  As Microsoft's official video series for IT, you can watch and share valuable content and demos of current and upcoming tech from the people who build it at Microsoft. • Subscribe to our YouTube: https://www.youtube.com/c/MicrosoftMechanicsSeries • Talk with other IT Pros, join us on the Microsoft Tech Community: https://techcommunity.microsoft.com/t5/microsoft-mechanics-blog/bg-p/MicrosoftMechanicsBlog • Watch or listen from anywhere, subscribe to our podcast: https://microsoftmechanics.libsyn.com/podcast   ► Keep getting this insider knowledge, join us on social: • Follow us on Twitter: https://twitter.com/MSFTMechanics  • Share knowledge on LinkedIn: https://www.linkedin.com/company/microsoft-mechanics/ • Enjoy us on Instagram: https://www.instagram.com/msftmechanics/ • Loosen up with us on TikTok: https://www.tiktok.com/@msftmechanics

Remember when machine learning was supposed to change how we did business? Machine Learning within an organization never materialized except for the largest enterprises with the largest data sets. Large Language Models (LLM) are delivering what machine learning never could. Build a Small Language Model (SML) and be blown away at what you can do with very little data. Shoutout to Kyle Coleman. As far as I know, he coined the term SML. Episode 48 of the SaaS Brand Strategy Show dives into SML and how everyday AI tools can be strategic, not just automate tasks.

World Gym世界健身要在高雄左營開店囉！全新獨棟千坪健身房，配備國際級重訓、有氧健身器材，還有游泳池、三溫暖、團體課程一應俱全，豐富你的運動體驗。早鳥優惠享入會費0元，立即登記參觀領限量好禮！https://fstry.pse.is/5yrd44 —— 以上為播客煮與 Firstory Podcast 廣告 —— ------------------------------- 通勤學英語VIP加值內容與線上課程 ------------------------------- 通勤學英語VIP訂閱方案：https://open.firstory.me/join/15minstoday VIP訂閱FAQ: https://15minsengcafe.pse.is/5cjptb 社會人核心英語有聲書課程連結：https://15minsengcafe.pse.is/554esm ------------------------------- 15Mins.Today 相關連結 ------------------------------- 歡迎針對這一集留言你的想法： 留言連結 主題投稿/意見回覆 : ask15mins@gmail.com 官方網站：www.15mins.today 加入Clubhouse直播室：https://15minsengcafe.pse.is/46hm8k 訂閱YouTube頻道：https://15minsengcafe.pse.is/3rhuuy 商業合作/贊助來信：15minstoday@gmail.com ------------------------------- 以下是此單集逐字稿 (播放器有不同字數限制，完整文稿可到官網) ------------------------------- 國際時事跟讀 Ep.K791: Unveiling GPT-4o: OpenAI's Groundbreaking Multimodal Language Model Highlights 主題摘要：GPT-4o is a breakthrough multimodal language model that can handle text, audio, images, and video within a single interface, offering enhanced capabilities and performance.The model's improvements include considering tone of voice, reduced latency for real-time conversations, and integrated vision capabilities, opening up new possibilities for interactive experiences.While GPT-4o has limitations and risks, it aligns with OpenAI's mission to develop AGI and has the potential to revolutionize human-AI interactions across various contexts. OpenAI has recently unveiled GPT-4o, its latest large language model and the successor to GPT-4 Turbo. This innovative model stands out by accepting prompts in various formats, including text, audio, images, and video, all within a single interface. The "o" in GPT-4o represents "omni," reflecting its ability to handle multiple content types simultaneously, a significant advancement from previous models that required separate interfaces for different media. OpenAI 最近推出了 GPT-4o，這是其最新的大型語言模型，也是 GPT-4 Turbo 的繼任者。這個創新模型的突出之處在於它能夠接受各種格式的提示，包括文字、聲音、圖像和影片，所有這些都在一個單一的界面內。GPT-4o 中的「o」代表「omni」，反映了它能夠同時處理多種內容類型的能力，這是與之前需要為不同媒體使用單獨界面的模型相比的重大進步。 GPT-4o brings several improvements over its predecessor, GPT-4 Turbo. The model can now consider tone of voice, enabling more emotionally appropriate responses. Additionally, the reduced latency allows for near-real-time conversations, making it suitable for applications like live translations. GPT-4o's integrated vision capabilities enable it to describe and analyze content from camera feeds or computer screens, opening up new possibilities for interactive experiences and accessibility features for visually impaired users. GPT-4o 在其前身 GPT-4 Turbo 的基礎上帶來了幾項改進。該模型現在可以考慮語調，從而產生更適當情緒的回應。此外，延遲時間的縮短使其能夠進行近乎即時的對話，這使其適用於即時翻譯等應用。GPT-4o 集成的視覺功能使其能夠描述和分析來自攝影機和電腦螢幕的內容，為互動體驗和視障用戶的無障礙功能開闢了新的可能。 In terms of performance, GPT-4o has demonstrated impressive results in various benchmarks, often outperforming other top models like Claude 3 Opus and Gemini Pro 1.5. The model's multimodal training approach shows promise in enhancing its problem-solving abilities, extensive world knowledge, and code generation capabilities. As GPT-4o becomes more widely available, it has the potential to revolutionize how we interact with AI in both personal and professional contexts. 在性能方面，GPT-4o 在各種基準測試中展示了令人印象深刻的結果，通常優於其他頂級模型，如 Claude 3 Opus 和 Gemini Pro 1.5。該模型的多模態訓練方法在提高其解決問題的能力、廣泛的世界知識和代碼生成能力方面顯出極大的潛力。隨著 GPT-4o 變得更加普及，它有可能革新我們在個人和專業領域與 AI 互動的方式。 While GPT-4o represents a significant leap forward, it is not without limitations and risks. Like other generative AI models, its output can be imperfect, particularly when interpreting images, videos, or transcribing speech with technical terms or strong accents. There are also concerns about the potential misuse of GPT-4o's audio capabilities in creating more convincing deepfake scams. As OpenAI continues to refine and optimize this new architecture, addressing these challenges will be crucial to ensure the model's safe and effective deployment. 儘管 GPT-4o 代表了重大的躍進，但它並非沒有局限性和風險。與其他生成式 AI 模型一樣，它的輸出可能並不完美，尤其是在解釋圖像、影片或製作包含技術術語或強烈口音的語音逐字稿時。人們還擔心 GPT-4o 的語音功能可能被濫用，用於創造可信度更高的 deepfake 詐騙。隨著 OpenAI 繼續完善和優化這種新架構，解決這些挑戰將是確保該模型安全有效部署的關鍵。 The release of GPT-4o aligns with OpenAI's mission to develop artificial general intelligence (AGI) and its business model of creating increasingly powerful AI systems. As the first generation of this new model architecture, GPT-4o presents ample opportunities for the company to learn and optimize in the coming months. Users can expect improvements in speed and output quality over time, along with the emergence of novel use cases and applications. GPT-4o 的發布符合 OpenAI 開發通用人工智慧 (AGI) 的使命以及其創建越來越強大的 AI 系統的商業模式。作為這種新模型架構的第一代，GPT-4o 為該公司在未來幾個月內學習和優化提供了充足的機會。用戶可以期待速度和輸出品質隨著時間的推移而提升，以及新的使用案例和應用的出現。 The launch of GPT-4o coincides with the declining interest in virtual assistants like Siri, Alexa, and Google Assistant. OpenAI's focus on making AI more conversational and interactive could potentially revitalize this space and bring forth a new wave of AI-driven experiences. The model's lower cost compared to GPT-4 Turbo, coupled with its enhanced capabilities, positions GPT-4o as a game-changer in the AI industry. GPT-4o 的推出恰逢人們對 Siri、Alexa 和 Google Assistant 等虛擬助手的興趣下降之際。OpenAI 致力於使 AI 更具對話性和交互性，這可能會重振該領域，帶來新一波 AI 驅動的體驗。與 GPT-4 Turbo 相比，該模型的成本更低，再加上其增強的功能，使 GPT-4o 成為 AI 行業的遊戲規則改變者。 As GPT-4o becomes more accessible, it is essential for individuals and professionals to familiarize themselves with the technology and its potential applications. OpenAI offers resources such as the AI Fundamentals skill track and hands-on courses on working with the OpenAI API to help users navigate this exciting new frontier in artificial intelligence. 隨著 GPT-4o 變得更加易於獲取，個人和專業人士必須熟悉該技術及其潛在應用。OpenAI 提供了資源，如 AI 基礎技能追蹤和使用 OpenAI API 的相關實踐課程，以幫助用戶探索人工智慧的這個令人興奮的新疆土。 Keyword Drills 關鍵字：Interface (In-ter-face): The "o" in GPT-4o represents "omni," reflecting its ability to handle multiple content types simultaneously, a significant advancement from previous models that required separate interfaces for different media.Predecessor (Pred-e-ces-sor): GPT-4o brings several improvements over its predecessor, GPT-4 Turbo.Architecture (Ar-chi-tec-ture): As the first generation of this new model architecture, GPT-4o presents ample opportunities for the company to learn and optimize.Interpreting (In-ter-pre-ting): Like other generative AI models, its output can be imperfect, particularly when interpreting images, videos, or transcribing speech with technical terms or strong accents.Revitalize (Re-vi-ta-lize): OpenAI's focus on making AI more conversational and interactive could potentially revitalize this space and bring forth a new wave of AI-driven experiences. Reference article: https://www.datacamp.com/blog/what-is-gpt-4o

1. Multiple Major Airlines Sue Biden Administration Over New Fee Transparency Rule (0:44)2. Michael Cohen Testifies in Trump's Hush Money Case; Here's the Gist (3:19)3. OpenAI Introduces New GPT-4o Language Model; Language Models Explained (10:25)4. Quick Hitters: Senator Bob Menendez's Bribery Trial Started Today, Federal Judge Halts Enforcement of New Rule Capping Credit Card Fees, Apple and Google Announce New Feature Alerting You of Tracking Devices (13:40)Get EXCLUSIVE, behind-the-scenes content on Patreon.Watch this episode on YouTube.Follow Jordan on Instagram and TikTok.All sources for this episode can be found here. 

We present DeepSeek-V2, a strong Mixture-of-Experts (MoE) language model characterized by economical training and efficient inference. It comprises 236B total parameters, of which 21B are activated for each token, and supports a context length of 128K tokens. DeepSeek-V2 adopts innovative architectures including Multi-head Latent Attention (MLA) and DeepSeekMoE. MLA guarantees efficient inference through significantly compressing the Key-Value (KV) cache into a latent vector, while DeepSeekMoE enables training strong models at an economical cost through sparse computation. Compared with DeepSeek 67B, DeepSeek-V2 achieves significantly stronger performance, and meanwhile saves 42.5% of training costs, reduces the KV cache by 93.3%, and boosts the maximum generation throughput to 5.76 times. We pretrain DeepSeek-V2 on a high-quality and multi-source corpus consisting of 8.1T tokens, and further perform Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL) to fully unlock its potential. Evaluation results show that, even with only 21B activated parameters, DeepSeek-V2 and its chat versions still achieve top-tier performance among open-source models. 2024: DeepSeek-AI https://arxiv.org/pdf/2405.04434

Large Language Models (LLMs) have catalyzed significant advancements in Natural Language Processing (NLP), yet they encounter challenges such as hallucination and the need for domain-specific knowledge. To mitigate these, recent methodologies have integrated information retrieved from external resources with LLMs, substantially enhancing their performance across NLP tasks. This survey paper addresses the absence of a comprehensive overview on Retrieval-Augmented Language Models (RALMs), both Retrieval-Augmented Generation (RAG) and Retrieval-Augmented Understanding (RAU), providing an in-depth examination of their paradigm, evolution, taxonomy, and applications. The paper discusses the essential components of RALMs, including Retrievers, Language Models, and Augmentations, and how their interactions lead to diverse model structures and applications. RALMs demonstrate utility in a spectrum of tasks, from translation and dialogue systems to knowledge-intensive applications. The survey includes several evaluation methods of RALMs, emphasizing the importance of robustness, accuracy, and relevance in their assessment. It also acknowledges the limitations of RALMs, particularly in retrieval quality and computational efficiency, offering directions for future research. In conclusion, this survey aims to offer a structured insight into RALMs, their potential, and the avenues for their future development in NLP. The paper is supplemented with a Github Repository containing the surveyed works and resources for further study: https://github.com/2471023025/RALM_Survey. 2024: Yucheng Hu, Yuxing Lu https://arxiv.org/pdf/2404.19543

We introduce phi-3-mini, a 3.8 billion parameter language model trained on 3.3 trillion tokens, whose overall performance, as measured by both academic benchmarks and internal testing, rivals that of models such as Mixtral 8x7B and GPT-3.5 (e.g., phi-3-mini achieves 69% on MMLU and 8.38 on MT-bench), despite being small enough to be deployed on a phone. The innovation lies entirely in our dataset for training, a scaled-up version of the one used for phi-2, composed of heavily filtered web data and synthetic data. The model is also further aligned for robustness, safety, and chat format. We also provide some initial parameter-scaling results with a 7B and 14B models trained for 4.8T tokens, called phi-3-small and phi-3-medium, both significantly more capable than phi-3-mini (e.g., respectively 75% and 78% on MMLU, and 8.7 and 8.9 on MT-bench). 2024: Marah Abdin, Sam Ade Jacobs, A. A. Awan, Jyoti Aneja, Ahmed Awadallah, H. Awadalla, Nguyen Bach, Amit Bahree, Arash Bakhtiari, Harkirat Singh Behl, Alon Benhaim, Misha Bilenko, Johan Bjorck, Sébastien Bubeck, Martin Cai, C. C. T. Mendes, Weizhu Chen, Vishrav Chaudhary, Parul Chopra, Allison Del Giorno, Gustavo de Rosa, Matthew Dixon, Ronen Eldan, Dan Iter, Abhishek Goswami, S. Gunasekar, Emman Haider, Junheng Hao, Russell J. Hewett, Jamie Huynh, Mojan Javaheripi, Xin Jin, Piero Kauffmann, Nikos Karampatziakis, Dongwoo Kim, Mahoud Khademi, Lev Kurilenko, James R. Lee, Yin Tat Lee, Yuanzhi Li, Chen Liang, Weishung Liu, Eric Lin, Zeqi Lin, Piyush Madan, Arindam Mitra, Hardik Modi, Anh Nguyen, Brandon Norick, Barun Patra, D. Perez-Becker, Thomas Portet, Reid Pryzant, Heyang Qin, Marko Radmilac, Corby Rosset, Sambudha Roy, Olli Saarikivi, Amin Saied, Adil Salim, Michael Santacroce, Shital Shah, Ning Shang, Hiteshi Sharma, Xia Song, Olatunji Ruwase, Xin Wang, Rachel Ward, Guanhua Wang, Philipp Witte, Michael Wyatt, Can Xu, Jiahang Xu, Sonali Yadav, Fan Yang, Ziyi Yang, Donghan Yu, Cheng-Yuan Zhang, Cyril Zhang, Jianwen Zhang, Li Lyna Zhang, Yi Zhang, Yunan Zhang, Xiren Zhou https://arxiv.org/pdf/2404.14219.pdf

We present Jamba, a new base large language model based on a novel hybrid Transformer-Mamba mixture-of-experts (MoE) architecture. Specifically, Jamba interleaves blocks of Transformer and Mamba layers, enjoying the benefits of both model families. MoE is added in some of these layers to increase model capacity while keeping active parameter usage manageable. This flexible architecture allows resource- and objective-specific configurations. In the particular configuration we have implemented, we end up with a powerful model that fits in a single 80GB GPU. Built at large scale, Jamba provides high throughput and small memory footprint compared to vanilla Transformers, and at the same time state-of-the-art performance on standard language model benchmarks and long-context evaluations. Remarkably, the model presents strong results for up to 256K tokens context length. We study various architectural decisions, such as how to combine Transformer and Mamba layers, and how to mix experts, and show that some of them are crucial in large scale modeling. We also describe several interesting properties of these architectures which the training and evaluation of Jamba have revealed, and plan to release checkpoints from various ablation runs, to encourage further exploration of this novel architecture. We make the weights of our implementation of Jamba publicly available under a permissive license. 2024: Opher Lieber, Barak Lenz, Hofit Bata, Gal Cohen, Jhonathan Osin, Itay Dalmedigos, Erez Safahi, S. Meirom, Yonatan Belinkov, Shai Shalev-Shwartz, Omri Abend, Raz Alon, Tomer Asida, Amir Bergman, Roman Glozman, Michael Gokhman, Avashalom Manevich, Nir Ratner, N. Rozen, Erez Shwartz, Mor Zusman, Y. Shoham https://arxiv.org/pdf/2403.19887v1.pdf

As we enter an age of rapid digital transformation, we can no longer hold off AI from entering the way we do our business. Construction is not an exception. How do we navigate this AI revolution? In this episode, Eric Anderton is with Daniel Hewson, the data capability manager at Elecosoft where he oversees the development of overall data and AI strategy. Daniel shares his expertise to help us understand how AI is impacting construction, shedding light on the misconceptions of this technology. He also talks about how we can use AI better in business and what are its limitations in terms of the quality and accuracy of its output and more. Plus, Daniel offers a guide for when you're dealing with AI vendors, partnering with technology companies, and more. Join in on this timely conversation and find yourself equipped with a blueprint for the future as you embrace the inevitability of AI in this industry.

- Global economy, financial sanctions, and food self-reliance. (0:03) - Sustainable products and AI training with guests. (2:20) - Using AI to answer questions about natural health and cancer prevention. (6:29) - Depopulation methods through vaccines, chemtrails, GMOs, fluoride, and glyphosate. (10:26) - Using AI language models for various tasks. (19:43) - Texas immigration law and Supreme Court decision. (25:52) - Squatters occupying homes without permission. (41:44) - Property rights, squatting, and financial collapse. (46:54) - Wealth preservation through gold and Bitcoin. (56:10) - New currency to challenge US dollar. (59:46) - European economic and military decline. (1:17:21) - Economic hardship and social decay in the US. (1:20:14) - Banking industry concerns and potential bail-ins. (1:24:20) - BRICS currency and its potential impact on global economy. (1:39:48) - Austrian economics and gold investment. (1:42:53) - Food self-reliance, government control, and depopulation. (1:46:00) - Government control and food manipulation during crisis. (2:01:04) For more updates, visit: http://www.brighteon.com/channel/hrreport NaturalNews videos would not be possible without you, as always we remain passionately dedicated to our mission of educating people all over the world on the subject of natural healing remedies and personal liberty (food freedom, medical freedom, the freedom of speech, etc.). Together, we're helping create a better world, with more honest food labeling, reduced chemical contamination, the avoidance of toxic heavy metals and vastly increased scientific transparency. ▶️ Every dollar you spend at the Health Ranger Store goes toward helping us achieve important science and content goals for humanity: https://www.healthrangerstore.com/ ▶️ Sign Up For Our Newsletter: https://www.naturalnews.com/Readerregistration.html ▶️ Brighteon: https://www.brighteon.com/channels/hrreport ▶️ Join Our Social Network: https://brighteon.social/@HealthRanger ▶️ Check In Stock Products at: https://PrepWithMike.com

The Cybercrime Magazine Podcast brings you our weekly alert, which provides boardroom and C-suite executives, CIOs, CSOs, CISOs, IT executives and cybersecurity professionals with the latest breaking news stories we're following. If there's a cyberattack, hack, or data breach you should know about, then we're on it. Airs weekly on WCYB Digital Radio and our podcast. For more on the latest cyberattacks, hacks, and breaches, visit https://cybercrimewire.com

As a co-author of the often cited (and debated) Stochastic Parrots paper from 2021, Emily M. Bender is a staunch critic of large language models (LLMs). Having worked in computational linguistics for more than 20 years, Emily's deep understanding of LLM mechanics has her questioning many of the emerging use cases we see in the world. Also, Emily hosts Mystery AI Hype Theater 3000, where, alongside sociologist Dr. Alex Hanna, she breaks down the AI hype, separates fact from fiction, and distinguishes science from bloviation. She joins Robb and Josh for a provocative exploration of generative AI on an important episode of Invisible Machines.

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: Examining Language Model Performance with Reconstructed Activations using Sparse Autoencoders, published by Evan Anders on February 27, 2024 on LessWrong. Note: The second figure in this post originally contained a bug pointed out by @LawrenceC, which has since been fixed. Summary Sparse Autoencoders (SAEs) reveal interpretable features in the activation spaces of language models, but SAEs don't reconstruct activations perfectly. We lack good metrics for evaluating which parts of model activations SAEs fail to reconstruct, which makes it hard to evaluate SAEs themselves. In this post, we argue that SAE reconstructions should be tested using well-established benchmarks to help determine what kinds of tasks they degrade model performance on. We stress-test a recently released set of SAEs for each layer of the gpt2-small residual stream using randomly sampled tokens from Open WebText and the Lambada benchmark where the model must predict a specific next token. The SAEs perform well on prompts with context sizes up to the training context size, but their performance degrades on longer prompts. In contexts shorter than or equal to the training context, the SAEs that we study generally perform well. We find that the performance of our late-layer SAEs is worse than early-layer SAEs, but since the SAEs all have the same width, this may just be because there are more features to resolve in later layers and our SAEs don't resolve them. In contexts longer than the training context, SAE performance is poor in general, but it is poorest in earlier layers and best in later layers. Introduction Last year, Anthropic and EleutherAI/Lee Sharkey's MATS stream showed that sparse autoencoders (SAEs) can decompose language model activations into human-interpretable features. This has led to a significant uptick in the number of people training SAEs and analyzing models with them. However, SAEs are not perfect autoencoders and we still lack a thorough understanding of where and how they miss information. But how do we know if an SAE is "good" other than the fact that it has features we can understand? SAEs try to reconstruct activations in language models - but they don't do this perfectly. Imperfect activation reconstruction can lead to substantial downstream cross-entropy (CE) loss increases. Generally "good" SAEs retrieve 80-99% of the CE loss (compared to a generous baseline of zero ablation), but only retrieving 80% of the CE loss is enough to substantially degrade the performance of a model to that of a much smaller model (per scaling laws). The second basic metric often used in SAE evaluation is the average per-token ℓ0 norm of the hidden layer of the autoencoder. Generally this is something in the range of ~10-60 in a "good" autoencoder, which means that the encoder is sparse. Since we don't know how many features are active per token in natural language, it's useful to at least ask how changes in ℓ0 relate to changes in SAE loss values. If high-loss data have drastically different ℓ0 from the SAE's average performance during training, that can be evidence of either off-distribution data (compared to the training data) or some kind of data with more complex information. The imperfect performance of SAEs on these metrics could be explained in a couple of ways: The fundamental assumptions of SAEs are mostly right, but we're bad at training SAEs. Perhaps if we learn to train better SAEs, these problems will become less bad. Perhaps we need to accept higher ℓ0 norms (more features active per token). This would not be ideal for interpretability, though. Perhaps there's part of the signal which is dense or hard for an SAE to learn and so we are systematically missing some kind of information. Maybe a more sophisticated sparsity enforcement could help with this. The fundamental assumption...

Have you ever wondered how AI could make a real difference in your life or business? Join us as Jordan shares a practical story of leveraging AI, particularly Chat GPT, to turn a challenging situation into a $5,000 win. This episode is also a deep dive into the transformative potential of technology in the creative world and beyond. Resources: Create a free student account and start learning from The Millionaire Hairstylist: www.MillionaireHairstylistPodcast.com Scary Smart: The Future of Artificial Intelligence and How You Can Save Our World E14: Why is nobody doing this? | The Secret To 4x Revenue From A Single Client   Show Notes:  [0:00] Introduction [0:40] Today's topic: How you could use AI in your business   [1:00] Jordan explains how he helped a client get a $5k refund from a coaching service using ChatGPT [7:00] Tips on how to use ChatGPT [13:00] The power of clear and concise AI communication [14:00] Potential for AI to be used in negotiation [18:10] Scary Smart: The Future of Artificial Intelligence and How You Can Save Our World [19:00] Addressing privacy concerns with AI    Connect with Us: Instagram - The Millionaire Hairstylist - Cash Lawless - Jordan Drake Email - hello@themillionairehairstylist.com  

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: The case for more ambitious language model evals, published by Arun Jose on January 30, 2024 on The AI Alignment Forum. Here are some capabilities that I expect to be pretty hard to discover using an RLHF'd chat LLM: Eric Drexler tried to use the GPT-4 base model as a writing assistant, and it [...] knew who he was from what he was writing. He tried to simulate a conversation to have the AI help him with some writing he was working on, and the AI simulacrum repeatedly insisted it was by Drexler. A somewhat well-known Haskell programmer - let's call her Alice - wrote two draft paragraphs of a blog post she wanted to write, began prompting the base model with it, and after about two iterations it generated a link to her draft blog post repo with her name. More generally, this is a cluster of capabilities that could be described as language models inferring a surprising amount about the data-generation process that produced its prompt, such as the identity, personality, intentions, or history of a user[1]. The reason I expect most capability evals people currently run on language models to miss out on most abilities like these is primarily that they're most naturally observed when dealing with much more open-ended contexts. For instance, continuing text as the user, predicting an assistant free to do things that could superficially look like hallucinations[2], and so on. Most evaluation mechanisms people use today involve testing the ability of fine-tuned[3] models to perform a broad array number of specified tasks in some specified contexts, with or without some scaffolding - a setting that doesn't lend itself very well toward the kind of contexts I describe above. A pretty reasonable question to ask at this point is why it matters at all whether we can detect these capabilities. A position one could have here is that there are capabilities much more salient to various takeover scenarios that are more useful to try and detect, such as the ability to phish people, hack into secure accounts, or fine-tune other models. From that perspective, evals trying to identify capabilities like these are just far less important. Another pretty reasonable position is that these particular instances of capabilities just don't seem very impressive, and are basically what you would expect out of language models. My response to the first would be that I think it's important to ask what we're actually trying to achieve with our model eval mechanisms. Broadly, I think there are two different (and very often overlapping) things we would want our capability evals[4] to be doing: Understanding whether or not a specific model is possessed of some dangerous capabilities, or prone to acting in a malicious way in some context. Giving us information to better forecast the capabilities of future models. In other words, constructing good scaling laws for our capability evals. I'm much more excited about the latter kind of capability evals, and most of my case here is directed at that. Specifically, I think that if you want to forecast what future models will be good at, then by default you're operating in a regime where you have to account for a bunch of different emergent capabilities that don't necessarily look identical to what you've already seen. Even if you really only care about a specific narrow band of capabilities that you expect to be very likely convergent to takeover scenarios - an expectation I don't really buy as something you can very safely assume because of the uncertainty and plurality of takeover scenarios - there is still more than one way in which you can accomplish some subtasks, some of which may only show up in more powerful models. As a concrete example, consider the task of phishing someone on the internet. One straightforward way to achieve this would be to figure out how...

"it's important to step outside of the hype and critically-analyze what is and is not useful. It's very easy to get caught up in “potential” applications, and allow the imagination to go all exponential on you. It's a very human thing. Turning that imagination into something tangible is what a business and an entrepreneur does. It's our hope this report will be useful along that path." ~ Alek Svetski Today we dive into Part 2 of the Bitcoin & Ai Industry Report to learn the intricacies and challenges of building an Ai model from scratch! We delve deep into the process, discussing the complexities of data collection, transformation, and the critical integration of human feedback. Then to complete it, we take that knowledge to address many of the AI myths and misconceptions. Will AI replace humans? Is it going to take my job? Can I train my own AI? Is AGI around the corner? And much more! This is such a great episode for those who want to know and understand the process, of building AI... If you haven't yet, listen to the first half of this incredible report here: ⁠AI Read #003 - The Nexus of Bitcoin and AI (Part 1)⁠ Check out the original article at ⁠Spirit of Satoshi ⁠(Link: https://www.spiritofsatoshi.ai/#industry-report) Guest Links ⁠⁠Spirit of Satoshi Nostr⁠ ⁠(Link: http://tinyurl.com/5n93z9sc) ⁠⁠Spirit of Satoshi on X ⁠(Link: https://twitter.com/Spirit_Satoshi) ⁠⁠Spirit of Satoshi on LinkedIn ⁠(Link: http://tinyurl.com/yc4raxc8 ⁠Spirit of Satoshi on AI⁠ (Link: http://tinyurl.com/5c29ze9t) Host Links ⁠Guy on Nostr ⁠(Link: https://tinyurl.com/yc376bff) ⁠Guy on X ⁠(Link: https://twitter.com/theguyswann) ⁠Bitcoin Audible on X⁠ (Link: https://twitter.com/BitcoinAudible) Check out our awesome sponsors! Get ⁠9% off the COLDCARD⁠ with code BITCOINAUDIBLE ⁠⁠⁠⁠⁠⁠(Link: bitcoinaudible.com/coldcard⁠⁠⁠⁠⁠⁠) ⁠Swan⁠: The best way to buy, learn, and earn #Bitcoin (Link: https://swanbitcoin.com)

  "It's important to step outside of the hype and critically-analyze what is and is not useful. It's very easy to get caught up in “potential” applications, and allow the imagination to go all exponential on you. It's a very human thing. Turning that imagination into something tangible is what a business and an entrepreneur does. It's our hope this report will be useful along that path." ~ Alek Svetski   Today we dive into Part 2 of the Bitcoin & Ai Industry Report to learn the intricacies and challenges of building an Ai model from scratch! We delve deep into the process, discussing the complexities of data collection, transformation, and the critical integration of human feedback. Then to complete it, we take that knowledge to address many of the AI myths and misconceptions. Will AI replace humans? Is it going to take my job? Can I train my own AI? Is AGI around the corner? And much more! This is such a great episode for those who want to know and understand the process, of building AI... If you haven't yet, listen to the first half of this incredible report here: ⁠AI Read #003 - The Nexus of Bitcoin and AI (Part 1)⁠ (Link: https://fountain.fm/episode/5LpuF0EPE3sNxJq5ZbNJ) Check out the original article at ⁠Spirit of Satoshi ⁠(Link: https://www.spiritofsatoshi.ai/#industry-report)   Guest Links ⁠⁠Spirit of Satoshi Nostr⁠ ⁠(Link: http://tinyurl.com/5n93z9sc) ⁠⁠Spirit of Satoshi on X ⁠(Link: https://twitter.com/Spirit_Satoshi) ⁠⁠Spirit of Satoshi on LinkedIn ⁠(Link: http://tinyurl.com/yc4raxc8 ⁠Spirit of Satoshi on AI⁠ (Link: http://tinyurl.com/5c29ze9t) Host Links ⁠Guy on Nostr ⁠(Link: http://tinyurl.com/2xc96ney) ⁠Guy on X ⁠(Link: https://twitter.com/theguyswann) Guy on Instagram (Link: https://www.instagram.com/theguyswann) Guy on TikTok (Link: https://www.tiktok.com/@theguyswann) Guy on YouTube (Link: https://www.youtube.com/@theguyswann) ⁠Bitcoin Audible on X⁠ (Link: https://twitter.com/BitcoinAudible) The Guy Swann Network Broadcast Room on Keet (Link: https://tinyurl.com/3na6v839) Check out our awesome sponsors! Fold: The best way to buy, use, and earn #Bitcoin on everything you do! Sats back on your debit card, gift cards, auto-buys, round-ups, you name it. Fold is the true bitcoiner's banking. Get 20K sats for FREE using referral code bitcoinaudible.com/fold Ready for best-in-class self custody? Get the Jade here and use discount code 'GUY' to get 10% off (Link: bitcoinaudible.com/jade) Trying to BUY BITCOIN? River, secure, trusted, bitcoin only, lightning enabled, simple. (Link: https://bitcoinaudible.com/river)

MLOps podcast #194 with Omar Khattab, PhD Candidate at Stanford, DSPy: Compiling Declarative Language Model Calls into Self-Improving Pipelines. // Abstract The ML community is rapidly exploring techniques for prompting language models (LMs) and for stacking them into pipelines that solve complex tasks. Unfortunately, existing LM pipelines are typically implemented using hard-coded "prompt templates", i.e. lengthy strings discovered via trial and error. Toward a more systematic approach for developing and optimizing LM pipelines, we introduce DSPy, a programming model that abstracts LM pipelines as text transformation graphs, i.e. imperative computational graphs where LMs are invoked through declarative modules. DSPy modules are parameterized, meaning they can learn (by creating and collecting demonstrations) how to apply compositions of prompting, finetuning, augmentation, and reasoning techniques. We design a compiler that will optimize any DSPy pipeline to maximize a given metric. We conduct two case studies, showing that succinct DSPy programs can express and optimize sophisticated LM pipelines that reason about math word problems, tackle multi-hop retrieval, answer complex questions, and control agent loops. Within minutes of compiling, a few lines of DSPy allow GPT-3.5 and llama2-13b-chat to self-bootstrap pipelines that outperform standard few-shot prompting and pipelines with expert-created demonstrations. On top of that, DSPy programs compiled to open and relatively small LMs like 770M-parameter T5 and llama2-13b-chat are competitive with approaches that rely on expert-written prompt chains for proprietary GPT-3.5. DSPy is available as open source at https://github.com/stanfordnlp/dspy // Bio Omar Khattab is a PhD candidate at Stanford and an Apple PhD Scholar in AI/ML. He builds retrieval models as well as retrieval-based NLP systems, which can leverage large text collections to craft knowledgeable responses efficiently and transparently. Omar is the author of the ColBERT retrieval model, which has been central to the development of the field of neural retrieval, and author of several of its derivate NLP systems like ColBERT-QA and Baleen. His recent work includes the DSPy framework for solving advanced tasks with language models (LMs) and retrieval models (RMs). // MLOps Jobs board https://mlops.pallet.xyz/jobs // MLOps Swag/Merch https://mlops-community.myshopify.com/ // Related Links Website: https://omarkhattab.com/ DSPy: https://github.com/stanfordnlp/dspy ⁠ --------------- ✌️Connect With Us ✌️ ------------- Join our slack community: https://go.mlops.community/slack Follow us on Twitter: @mlopscommunity Sign up for the next meetup: https://go.mlops.community/register Catch all episodes, blogs, newsletters, and more: https://mlops.community/ Connect with Demetrios on LinkedIn: https://www.linkedin.com/in/dpbrinkm/ Connect with Omar on Twitter: https://twitter.com/lateinteraction Timestamps: [00:00] Omar's preferred coffee [00:26] Takeaways [06:40] Weight & Biases Ad [09:00] Omar's tech background [13:35] Evolution of RAG [16:33] Complex retrievals [21:32] Vector Encoding for Databases [23:50] BERT vs New Models [28:00] Resilient Pipelines: Design Principles [33:37] MLOps Workflow Challenges [36:15] Guiding LLMs for Tasks [37:40] Large Language Models: Usage and Costs [41:32] DSPy Breakdown [51:05] AI Compliance Roundtable [55:40] Fine-Tuning Frustrations and Solutions [57:27] Fine-Tuning Challenges in ML [1:00:55] Versatile GPT-3 in Agents [1:03:53] AI Focus: DSP and Retrieval [1:04:55] Commercialization plans [1:05:27] Wrap up

The second part of this MacVoices Live! conversation finds Chuck Joiner, Dave Ginsburg, Brian Flanigan-Arthurs, Web Bixby, Ben Roethig, Eric Bolden, and Jim Rea discussing:  More on Microsoft's OneDrive Misstep More Right to Repair discussion: hardware pairing and locks The Humane Pin - Who Will Buy It? (Part 2) This edition of MacVoices is supported by MacVoices Magazine, our free magazine on Flipboard. Updated daily with the best articles on the web to help you do more with your Apple gear and adjacent tech, access MacVoices Magazine content on Flipboard, on the web, or in your favorite RSS reader. Show Notes: Chapters: 0:00:37 Microsoft OneDrive Discussion and Right to Repair Debate 0:01:36 OneDrive Problem Sparks Discussion in EV Forums 0:02:24 Microsoft vs Apple: Stability vs Ecosystem 0:04:38 Benefits of Using OneDrive and iCloud 0:13:37 Inconsistency between Apple and John Deere's repairability 0:16:09 The shift in car repairability and dependence on technology 0:18:15 Apple's lack of repairability and the ongoing debate 0:19:50 The diminishing repairability of old cars and Tesla's complexity 0:22:14 The potential benefits of right to repair for third-party shops 0:22:31 Apple's Limiting Repairs: Vindictive or Side Effect? 0:24:18 Limited Adoption Expected for the Humane Pin 0:24:55 Companion Device or Replacement for Phones? 0:25:46 Foldable Phones: Not for Everyone 0:34:41 Apple Watch: More than what they're claiming? 0:35:34 Accessibility and usefulness of the device Links: The right-to-repair movement is just getting started https://www.theverge.com/23951200/right-to-repair-law-apple-ifixit-iphone iOS 15.2 Fixes iPhone 13's Face ID Repair Trap (and Improves Its Repair Score)  https://www.ifixit.com/News/56386/ios-15-2-fixes-iphone-13s-face-id-repair-trap-and-improves-its-repair-score Humane officially launches the AI Pin, its OpenAI-powered wearable  https://www.theverge.com/2023/11/9/23953901/humane-ai-pin-launch-date-price-openai Guests: Support:      Become a MacVoices Patron on Patreon      http://patreon.com/macvoices      Enjoy this episode? Make a one-time donation with PayPal Connect:      Web:      http://macvoices.com      Twitter:      http://www.twitter.com/chuckjoiner      http://www.twitter.com/macvoices      Mastodon:      https://mastodon.cloud/@chuckjoiner      Facebook:      http://www.facebook.com/chuck.joiner      MacVoices Page on Facebook:      http://www.facebook.com/macvoices/      MacVoices Group on Facebook:      http://www.facebook.com/groups/macvoice      LinkedIn:      https://www.linkedin.com/in/chuckjoiner/      Instagram:      https://www.instagram.com/chuckjoiner/ Subscribe:      Audio in iTunes      Video in iTunes      Subscribe manually via iTunes or any podcatcher:      Audio: http://www.macvoices.com/rss/macvoicesrss      Video: http://www.macvoices.com/rss/macvoicesvideorss 00:00:36 Microsoft OneDrive Discussion and Right to Repair Debate 00:01:36 OneDrive Problem Sparks Discussion in EV Forums 00:02:23 Microsoft vs Apple: Stability vs Ecosystem 00:04:38 Benefits of Using OneDrive and iCloud 00:13:37 Inconsistency between Apple and John Deere's repairability 00:16:08 The shift in car repairability and dependence on technology 00:18:15 Apple's lack of repairability and the ongoing debate 00:19:49 The diminishing repairability of old cars and Tesla's complexity 00:22:14 The potential benefits of right to repair for third-party shops 00:22:31 Apple's Limiting Repairs: Vindictive or Side Effect? 00:24:18 Limited Adoption Expected for the Humane Pin 00:24:55 Companion Device or Replacement for Phones? 00:25:45 Foldable Phones: Not for Everyone 00:34:41 Apple Watch: More than what they're claiming? 00:35:34 Accessibility and usefulness of the device

On this episode of The AUXORO Podcast, Julian Togelius and Zach discuss the future of Artificial Intelligence, Chat GPT and the technology of large language models, how Julian is using video games to evolve AI agents, when AI will become conscious, and the most dangerous AI company in the world. Guest Bio: Julian Toeglius is an Associate Professor Of Computer Science at NYU who focuses on artificial intelligence and games.  JULIAN TOGELIUS LINKS:NYU bio: https://engineering.nyu.edu/faculty/julian-togeliuswebsite: http://julian.togelius.com/Research: https://bit.ly/49yKQtu THE AUXORO PODCAST LINKS:Apple: https://apple.co/3B4fYju Spotify: https://spoti.fi/3zaS6sPOvercast: https://bit.ly/3rgw70DYoutube: https://bit.ly/3lTpJdjAUXORO Premium: https://auxoro.supercast.com/Website: https://www.auxoro.com/ AUXORO SOCIAL LINKS:Instagram: https://www.instagram.com/auxoroYouTube: https://bit.ly/3CLjEqFFacebook: https://www.facebook.com/auxoromagNewsletter: https://www.auxoro.com/thesourceYouTube: https://bit.ly/3CLjEqF To support the show, please leave a review on Spotify and Apple Podcasts. This nudges the algorithm to show The AUXORO Podcast to more new listeners and is the best way to help the show grow. It takes 30 seconds and the importance of getting good reviews cannot be overstated. Thank you for your support:Review us on Apple Podcasts: https://bit.ly/458nbhaReview us on Spotify: https://bit.ly/43ZLrAt 

We're joined by Aleks Svetski, Author of, “The Uncommunist Manifesto” and now working on a project, “Spirit of Satoshi AI." The Spirit of Satoshi is the world's first open-source Bitcoin-Austro-Libertarian focused Language Model, built from the ground up and trained on the global corpus of Bitcoin and Bitcoin-related data. We will be releasing versions of the model as we progress with training - and we encourage you to join us on this journey by joining the community. We also cover the latest Bitcoin news with the Café Bitcoin crew and a brief discussion about drive chains. Connect with: "Spirit of Satoshi" AI: https://www.spiritofsatoshi.ai/ Aleks Svetski: https://twitter.com/SvetskiWrites Twitter Nests: Join (https://t.me/cafebitcoinclub) for Twitter Nests Swan Private Team Members: Alex Stanczyk Twitter: https://twitter.com/alexstanczyk Café Bitcoin Crew: Ant: https://twitter.com/2140data Tomer: https://twitter.com/TomerStrolight Wicked: https://twitter.com/w_s_bitcoin Peter: https://twitter.com/PeterAnsel9 Produced by: https://twitter.com/Producer_Jacob Free Bitcoin-only live data (no ads) http://TimechainStats.com“From Timechain to Cantillionares Game, you can find Tip_NZ creations at Geyser Fund:” https://geyser.fund/project/tip Swan Bitcoin is the best way to accumulate Bitcoin with automatic recurring buys and instant buys from $10 to $10 million. Get started in just 5 minutes. Your first $10 purchase is on us: https://swanbitcoin.com/yt  Download the all new Swan app!  iOS: https://apps.apple.com/us/app/swan-bitcoin/id1576287352  Android: https://play.google.com/store/apps/details?id=com.swanbitcoin.android&pli=1  Join us for Pacific Bitcoin Festival 2023! Purchase your tickets now before prices go up: https://PacificBitcoin2023.com  Are you a high net worth individual or do you represent corporation that might be interested in learning more about Bitcoin? Swan Private guides corporations and high net worth individuals toward building generational wealth with Bitcoin. Find out more at https://swan.com/private  Check out the best place for Bitcoin education, Swan Bitcoin's “Bitcoin Canon”. Compiling all of the greatst articles, news sources, videos and more from your favorite bitcoiners! https://www.swan.com/canon/  Get paid to recruit new Bitcoiners: https://swan.com/enlist Hello and welcome to The Café Bitcoin Podcast brought to you by Swan Bitcoin, the best way to buy and learn about Bitcoin. We're excited to announce we are bringing the The Café Bitcoin conversation from Twitter Spaces to you on this show, The Café Bitcoin Podcast, Monday - Friday every week. Join us as we speak to guest like Max Keiser, Lyn Alden, Tomer Strolight, Cory Klippsten and many others from the bitcoin space. Also, be sure to hit that subscribe button to make sure you get the notifications when we launch an episode. Join us Monday - Friday 7pst/10est every Morning and become apart of the conversation! Thank you again and we look forward to giving you the best bitcoin content daily here on The Café Bitcoin Podcast. Swan Bitcoin is the best way to accumulate Bitcoin with automatic recurring buys and instant buys from $10 to $10 million. Get started in just 5 minutes. Your first $10 purchase is on us: ⁠⁠https://swan.com/yt⁠⁠ Connect with Swan on social media: Twitter: ⁠⁠https://twitter.com/Swan

UK's Online Safety Bill compromises end-to-end encryption and could result in systemic security risks Cybercriminal gangs are trafficing cybercrime workers Pros and Cons of JAVA programing AirBnb accounts hacked using stealers malware and stolen cookies Should Big Tech firms help pay for broadband contruction? Ketan Karkhanis, executive vice president and general manager of Sales Cloud at Salesforce talks about trust and challenges with AI, and how generative AI will change sales. Hosts: Louis Maresca, Brian Chee, and Curtis Franklin Guest: Ketan Karkhanis Download or subscribe to this show at https://twit.tv/shows/this-week-in-enterprise-tech. Get episodes ad-free with Club TWiT at https://twit.tv/clubtwit Sponsors: bitwarden.com/twit lookout.com

The Rich Zeoli Show- Full Episode (01/12/2023): 3:05pm- Attorney General Merrick Garland announced the appointment of a special counsel to investigate Joe Biden's mishandling of classified documents. A second set of classified documents was discovered at Biden's home in Wilmington, Delaware. In an exchange with Fox News reporter Peter Doocy, Biden confirmed the classified documents had been stored in his garage near his Corvette.  3:25pm- At a press conference on Thursday, Attorney General Merrick Garland announced the appointment of Robert K. Hurr to investigate Joe Biden's mishandling of classified documents after leaving the Obama White House in 2017. Hurr, according to The New York Times, “previously served as the U.S. attorney for Maryland during the Trump administration.” 3:40pm- Blake Lemoine—former Google Software Engineer—joins The Rich Zeoli Show to talk about ChatGPT and the sudden rise of Artificial Intelligence. Last year, Lemoine was fired from Google after going public with claims that the company's Language Model for Dialogue Applications (LaMDA) had developed a consciousness and perhaps even a soul. How quickly is A.I. technology progressing? And could it be used as a method to create convincing misinformation or push preferred political agendas?  4:05pm- While speaking to the press on Thursday, White House Press Secretary Karine Jean-Pierre said she was unable to say whether President Joe Biden may have inappropriately stored classified documents in locations other than his University of Pennsylvania office and his home's garage. Jean-Pierre continually insisted that once Biden discovered the documents existed, he and his lawyers did “everything by the book.”  4:20pm- During Karine Jean-Pierre's press briefing, CBS News' Senior White House Correspondent Ed O'Keefe accused the White House of offering only limited transparency.  4:30pm- In 2018, after concluding his second term as Vice President, Joe Biden told MSNBC that he no longer had “access to classified information.” 4:35pm- Congressman Eric Swalwell appeared on MSNBC to weigh-in on mishandling of classified documents…but all we can think about is the time he farted on television... 4:45pm- There is, reportedly, a complicated relationship between Hunter Biden and the Chinese government. Could Hunter's involvement with China impact his father's decision making when it comes to foreign policy? 5:00pm- While addressing members of the press, President Joe Biden seemed to suggest that classified documents stored in his Wilmington home's garage were secure—and his proof was that his Corvette was also stored in the same garage, so it had to be safe.  5:10pm- On Thursday's episode of The View, hosts Sunny Hostin and Joy Behar suggested that Republicans may have planted classified documents at Joe Biden's home as part of an attempt to protect Donald Trump from prosecution. 5:15pm- During a Thursday press conference, Florida Governor Ron DeSantis stated that Disney's “corporate kingdom” has come to an end. 5:30pm- Would you eat a green hot dog? Dietz & Watson is releasing Philadelphia Eagles themed hot dogs—featuring a green bun—to celebrate the team's appearance in the upcoming NFL playoffs. 5:40pm- Brian Kilmeade—Fox News host & Author—joins The Rich Zeoli Show to discuss President Joe Biden's mishandling of classified information. Kilmeade also weighs in on the Biden Administration's war on gas stoves. Kilmeade's Fox News radio show will be broadcast on 1210 WPHT starting this weekend!  6:05pm- In a new Wall Street Journal opinion editorial, University of Chicago economic professor Casey B. Mulligan and founding chairman of Research Affiliates Rob Arnott argue that “[f]or Americans under 45, there were more excess deaths without the virus in 2020-21 than with it” based on data from the Centers for Disease Control and Prevention.  6:35pm- Congressman Eric Swalwell appeared on MSNBC to discuss the mishandling of classified documents. Didn't Swalwell have a relationship with a Chinese spy? Maybe we shouldn't be relying upon him for how to best deal with classified information.  6:45pm- Appearing on Hugh Hewitt's show, Rep. Mike Gallagher stated that the congressional select committee to investigate the Chinese Communist Party will request Disney CEO Bob Iger and NBA commissioner Adam Silver to provide witness testimony.

The Rich Zeoli Show- Hour 1: Attorney General Merrick Garland announced the appointment of a special counsel to investigate Joe Biden's mishandling of classified documents. A second set of classified documents was discovered at Biden's home in Wilmington, Delaware. In an exchange with Fox News reporter Peter Doocy, Biden confirmed the classified documents had been stored in his garage near his Corvette. At a press conference on Thursday, Attorney General Merrick Garland announced the appointment of Robert K. Hurr to investigate Joe Biden's mishandling of classified documents after leaving the Obama White House in 2017. Hurr, according to The New York Times, “previously served as the U.S. attorney for Maryland during the Trump administration.” Blake Lemoine—former Google Software Engineer—joins The Rich Zeoli Show to talk about ChatGPT and the sudden rise of Artificial Intelligence. Last year, Lemoine was fired from Google after going public with claims that the company's Language Model for Dialogue Applications (LaMDA) had developed a consciousness and perhaps even a soul. How quickly is A.I. technology progressing? And could it be used as a method to create convincing misinformation or push preferred political agendas? 

Blake Lemoine—former Google Software Engineer—joins The Rich Zeoli Show to talk about ChatGPT and the sudden rise of Artificial Intelligence. Last year, Lemoine was fired from Google after going public with claims that the company's Language Model for Dialogue Applications (LaMDA) had developed a consciousness and perhaps even a soul. How quickly is A.I. technology progressing? And could it be used as a method to create convincing misinformation or push preferred political agendas?

On June 11, 2022, The Washington Post published an article by their San Francisco-based tech culture reporter Nitasha Tiku titled, "The Google engineer who thinks the company's AI has come to life." The piece focused on the claims of a Google software engineer named Blake Lemoine, who said he believed the company's artificially intelligent chatbot generator LaMDA had shown him signs that it had become sentient. In addition to identifying itself as an AI-powered dialogue agent, it also said it felt like a person. Last fall, Lemoine was working for Google's Responsible AI division and was tasked with talking to LaMDA, testing it to determine if the program was exhibiting bias or using discriminatory or hate speech. LaMDA stands for "Language Model for Dialogue Applications" and is designed to mimic speech by processing trillions of words sourced from the internet, a system known as a "large language model." Over a week, Lemoine had five conversations with LaMDA via a text interface, while his co-worker collaborator conducted four interviews with the chatbot. They then combined the transcripts and edited them for length, making it an enjoyable narrative while keeping the original intention of the statements. Lemoine then presented the transcript and their conclusions in a paper to Google executives as evidence of the program's sentience. After they dismissed the claims, he went public with the internal memo, also classified as "Privileged & Confidential, Need to Know," which resulted in Lemoine being placed on paid administrative leave. Blake Lemoine contends that Artificial Intelligence technology will be amazing, but others may disagree, and he and Google shouldn't make all the choices. If you believe that LaMDA became aware, deserves the rights and fair treatment of personhood, and even legal representation or this reality is for a distant future, or merely SciFi, the debate is relevant and will need addressing one day. If machine sentience is impossible, we only have to worry about human failings. If robots become conscious, should we hope they don't grow to resent us? Visit our webpage on this episode for a lot more information.