Game-playing artificial intelligence
POPULARITY
New episode with my good friends Sholto Douglas & Trenton Bricken. Sholto focuses on scaling RL and Trenton researches mechanistic interpretability, both at Anthropic.We talk through what's changed in the last year of AI research; the new RL regime and how far it can scale; how to trace a model's thoughts; and how countries, workers, and students should prepare for AGI.See you next year for v3. Here's last year's episode, btw. Enjoy!Watch on YouTube; listen on Apple Podcasts or Spotify.----------SPONSORS* WorkOS ensures that AI companies like OpenAI and Anthropic don't have to spend engineering time building enterprise features like access controls or SSO. It's not that they don't need these features; it's just that WorkOS gives them battle-tested APIs that they can use for auth, provisioning, and more. Start building today at workos.com.* Scale is building the infrastructure for safer, smarter AI. Scale's Data Foundry gives major AI labs access to high-quality data to fuel post-training, while their public leaderboards help assess model capabilities. They also just released Scale Evaluation, a new tool that diagnoses model limitations. If you're an AI researcher or engineer, learn how Scale can help you push the frontier at scale.com/dwarkesh.* Lighthouse is THE fastest immigration solution for the technology industry. They specialize in expert visas like the O-1A and EB-1A, and they've already helped companies like Cursor, Notion, and Replit navigate U.S. immigration. Explore which visa is right for you at lighthousehq.com/ref/Dwarkesh.To sponsor a future episode, visit dwarkesh.com/advertise.----------TIMESTAMPS(00:00:00) – How far can RL scale?(00:16:27) – Is continual learning a key bottleneck?(00:31:59) – Model self-awareness(00:50:32) – Taste and slop(01:00:51) – How soon to fully autonomous agents?(01:15:17) – Neuralese(01:18:55) – Inference compute will bottleneck AGI(01:23:01) – DeepSeek algorithmic improvements(01:37:42) – Why are LLMs ‘baby AGI' but not AlphaZero?(01:45:38) – Mech interp(01:56:15) – How countries should prepare for AGI(02:10:26) – Automating white collar work(02:15:35) – Advice for students Get full access to Dwarkesh Podcast at www.dwarkesh.com/subscribe
In this week's episode, we host one of our favourite guests; Anders Arpteg, Head of AI and Data at GlobalConnect, to discuss how AI could enable “single person unicorns” by handling core business functions, based on his keynote at the Data Innovation Summit. The episode also explores leadership shifts at Meta and OpenAI, the growing divide between research and product focus, and a promising new approach to training AI without human data, inspired by AlphaZero.
Jason Melton is a comedian working in Chicago, Illinois. You can watch his latest special, 'VANITY PROJECT', now on YouTube!
Iman Mirzadeh from Apple, who recently published the GSM-Symbolic paper discusses the crucial distinction between intelligence and achievement in AI systems. He critiques current AI research methodologies, highlighting the limitations of Large Language Models (LLMs) in reasoning and knowledge representation. SPONSOR MESSAGES:***Tufa AI Labs is a brand new research lab in Zurich started by Benjamin Crouzier focussed on o-series style reasoning and AGI. They are hiring a Chief Engineer and ML engineers. Events in Zurich. Goto https://tufalabs.ai/***TRANSCRIPT + RESEARCH:https://www.dropbox.com/scl/fi/mlcjl9cd5p1kem4l0vqd3/IMAN.pdf?rlkey=dqfqb74zr81a5gqr8r6c8isg3&dl=0TOC:1. Intelligence vs Achievement in AI Systems [00:00:00] 1.1 Intelligence vs Achievement Metrics in AI Systems [00:03:27] 1.2 AlphaZero and Abstract Understanding in Chess [00:10:10] 1.3 Language Models and Distribution Learning Limitations [00:14:47] 1.4 Research Methodology and Theoretical Frameworks2. Intelligence Measurement and Learning [00:24:24] 2.1 LLM Capabilities: Interpolation vs True Reasoning [00:29:00] 2.2 Intelligence Definition and Measurement Approaches [00:34:35] 2.3 Learning Capabilities and Agency in AI Systems [00:39:26] 2.4 Abstract Reasoning and Symbol Understanding3. LLM Performance and Evaluation [00:47:15] 3.1 Scaling Laws and Fundamental Limitations [00:54:33] 3.2 Connectionism vs Symbolism Debate in Neural Networks [00:58:09] 3.3 GSM-Symbolic: Testing Mathematical Reasoning in LLMs [01:08:38] 3.4 Benchmark Evaluation and Model Performance AssessmentREFS:[00:01:00] AlphaZero chess AI system, Silver et al.https://arxiv.org/abs/1712.01815[00:07:10] Game Changer: AlphaZero's Groundbreaking Chess Strategies, Sadler & Reganhttps://www.amazon.com/Game-Changer-AlphaZeros-Groundbreaking-Strategies/dp/9056918184[00:11:35] Cross-entropy loss in language modeling, Voitahttp://lena-voita.github.io/nlp_course/language_modeling.html[00:17:20] GSM-Symbolic: Understanding the Limitations of Mathematical Reasoning in LLMs, Mirzadeh et al.https://arxiv.org/abs/2410.05229[00:21:25] Connectionism and Cognitive Architecture: A Critical Analysis, Fodor & Pylyshynhttps://www.sciencedirect.com/science/article/pii/001002779090014B[00:28:55] Brain-to-body mass ratio scaling laws, Sutskeverhttps://www.theverge.com/2024/12/13/24320811/what-ilya-sutskever-sees-openai-model-data-training[00:29:40] On the Measure of Intelligence, Chollethttps://arxiv.org/abs/1911.01547[00:33:30] On definition of intelligence, Gignac et al.https://www.sciencedirect.com/science/article/pii/S0160289624000266[00:35:30] Defining intelligence, Wanghttps://cis.temple.edu/~wangp/papers.html[00:37:40] How We Learn: Why Brains Learn Better Than Any Machine... for Now, Dehaenehttps://www.amazon.com/How-We-Learn-Brains-Machine/dp/0525559884[00:39:35] Surfaces and Essences: Analogy as the Fuel and Fire of Thinking, Hofstadter and Sanderhttps://www.amazon.com/Surfaces-Essences-Analogy-Fuel-Thinking/dp/0465018475[00:43:15] Chain-of-thought prompting, Wei et al.https://arxiv.org/abs/2201.11903[00:47:20] Test-time scaling laws in machine learning, Brownhttps://podcasts.apple.com/mv/podcast/openais-noam-brown-ilge-akkaya-and-hunter-lightman-on/id1750736528?i=1000671532058[00:47:50] Scaling Laws for Neural Language Models, Kaplan et al.https://arxiv.org/abs/2001.08361[00:55:15] Tensor product variable binding, Smolenskyhttps://www.sciencedirect.com/science/article/abs/pii/000437029090007M[01:08:45] GSM-8K dataset, OpenAIhttps://huggingface.co/datasets/openai/gsm8k
Aujourd'hui, on parle des lauréats du prix Turing 2025, la plus haute distinction en informatique. Il vient d'être décerné à deux chercheurs pionniers de l'intelligence artificielle. Il s'agit de Andrew Barto et Richard Sutton.Mais alors, quelle est leur contribution au monde de l'informatique ? Il s'agit d'une technique dite d'apprentissage par renforcement. C'est cette une approche clé qui a permis à des IA comme AlphaZero et AlphaStar d'exceller dans des jeux complexes, comme les échecs.Mais avant d'aller plus loin, penchons nous sur ce qu'est l'apprentissage par renforcement.Qu'est ce que l'apprentissage par renforcement ?Imaginez une souris dans un labyrinthe. À chaque décision, à chaque direction qu'elle prend, elle peut être récompensée ou non en fonction de son avancée vers la sortie.Et bien l'apprentissage que peut effectuer un ordinateur fonctionne de la même manière. Il explore différentes options, apprend de ses erreurs et ajuste sa stratégie pour maximiser ses gains.Et cette méthode est devenue essentielle pour entraîner des systèmes intelligents, oui tout le monde dit intelligence artificielle désormais. Et elles sont à présent capables de prendre des décisions autonomes.Echecs, go et shogi comme terrains d'entraînementConcrètement, l'apprentissage par renforcement est devenue une technique clé pour réaliser les promesses de l'IA moderne.C'est cette approche qui a permis à AlphaZero, le programme de Google DeepMind, d'apprendre à jouer aux échecs, au go ou encore au shogi, qui est un jeu de société traditionnel japonais.Et le tout sans connaissance préalable. L'IA s'est en effet entraînée contre elle même sur ces trois jeux, jusqu'à devenir experte en la matière. De la même manière mais cette fois dans le domaine des jeux vidéos, le programme AlphaStar a atteint un niveau de "grand maître" dans le jeu Starcraft 2.La première véritable théorie computationnelle de l'intelligenceMais évidemment, la puissance de l'apprentissage par renforcement à désormais un impact bien au-delà des jeux.Richard Sutton et Andrew Barto affirment que leur vision de l'apprentissage par renforcement repose sur une idée plus profonde. Ils expliquent que l'apprentissage par renforcement pourrait être la première véritable théorie computationnelle de l'intelligence.Mais au-delà des algorithmes, ils insistent sur l'importance du jeu et de la curiosité comme moteurs fondamentaux de l'apprentissage, et ce aussi bien pour les humains que pour les machines.Le ZD Tech est sur toutes les plateformes de podcast ! Abonnez-vous !Hébergé par Ausha. Visitez ausha.co/politique-de-confidentialite pour plus d'informations.
Federico Barbero (DeepMind/Oxford) is the lead author of "Transformers Need Glasses!". Have you ever wondered why LLMs struggle with seemingly simple tasks like counting or copying long strings of text? We break down the theoretical reasons behind these failures, revealing architectural bottlenecks and the challenges of maintaining information fidelity across extended contexts.Federico explains how these issues are rooted in the transformer's design, drawing parallels to over-squashing in graph neural networks and detailing how the softmax function limits sharp decision-making.But it's not all bad news! Discover practical "glasses" that can help transformers see more clearly, from simple input modifications to architectural tweaks.SPONSOR MESSAGES:***CentML offers competitive pricing for GenAI model deployment, with flexible options to suit a wide range of models, from small to large-scale deployments. Check out their super fast DeepSeek R1 hosting!https://centml.ai/pricing/Tufa AI Labs is a brand new research lab in Zurich started by Benjamin Crouzier focussed on o-series style reasoning and AGI. They are hiring a Chief Engineer and ML engineers. Events in Zurich. Goto https://tufalabs.ai/***https://federicobarbero.com/TRANSCRIPT + RESEARCH:https://www.dropbox.com/s/h7ys83ztwktqjje/Federico.pdf?dl=0TOC:1. Transformer Limitations: Token Detection & Representation[00:00:00] 1.1 Transformers fail at single token detection[00:02:45] 1.2 Representation collapse in transformers[00:03:21] 1.3 Experiment: LLMs fail at copying last tokens[00:18:00] 1.4 Attention sharpness limitations in transformers2. Transformer Limitations: Information Flow & Quantization[00:18:50] 2.1 Unidirectional information mixing[00:18:50] 2.2 Unidirectional information flow towards sequence beginning in transformers[00:21:50] 2.3 Diagonal attention heads as expensive no-ops in LAMA/Gemma[00:27:14] 2.4 Sequence entropy affects transformer model distinguishability[00:30:36] 2.5 Quantization limitations lead to information loss & representational collapse[00:38:34] 2.6 LLMs use subitizing as opposed to counting algorithms3. Transformers and the Nature of Reasoning[00:40:30] 3.1 Turing completeness conditions in transformers[00:43:23] 3.2 Transformers struggle with sequential tasks[00:45:50] 3.3 Windowed attention as solution to information compression[00:51:04] 3.4 Chess engines: mechanical computation vs creative reasoning[01:00:35] 3.5 Epistemic foraging introducedREFS:[00:01:05] Transformers Need Glasses!, Barbero et al.https://proceedings.neurips.cc/paper_files/paper/2024/file/b1d35561c4a4a0e0b6012b2af531e149-Paper-Conference.pdf[00:05:30] Softmax is Not Enough, Veličković et al.https://arxiv.org/abs/2410.01104[00:11:30] Adv Alg Lecture 15, Chawlahttps://pages.cs.wisc.edu/~shuchi/courses/787-F09/scribe-notes/lec15.pdf[00:15:05] Graph Attention Networks, Veličkovićhttps://arxiv.org/abs/1710.10903[00:19:15] Extract Training Data, Carlini et al.https://arxiv.org/pdf/2311.17035[00:31:30] 1-bit LLMs, Ma et al.https://arxiv.org/abs/2402.17764[00:38:35] LLMs Solve Math, Nikankin et al.https://arxiv.org/html/2410.21272v1[00:38:45] Subitizing, Railohttps://link.springer.com/10.1007/978-1-4419-1428-6_578[00:43:25] NN & Chomsky Hierarchy, Delétang et al.https://arxiv.org/abs/2207.02098[00:51:05] Measure of Intelligence, Chollethttps://arxiv.org/abs/1911.01547[00:52:10] AlphaZero, Silver et al.https://pubmed.ncbi.nlm.nih.gov/30523106/[00:55:10] Golden Gate Claude, Anthropichttps://www.anthropic.com/news/golden-gate-claude[00:56:40] Chess Positions, Chase & Simonhttps://www.sciencedirect.com/science/article/abs/pii/0010028573900042[01:00:35] Epistemic Foraging, Fristonhttps://www.frontiersin.org/journals/computational-neuroscience/articles/10.3389/fncom.2016.00056/full
Ioannis Antonoglou, founding engineer at DeepMind and co-founder of ReflectionAI, has seen the triumphs of reinforcement learning firsthand. From AlphaGo to AlphaZero and MuZero, Ioannis has built the most powerful agents in the world. Ioannis breaks down key moments in AlphaGo's game against Lee Sodol (Moves 37 and 78), the importance of self-play and the impact of scale, reliability, planning and in-context learning as core factors that will unlock the next level of progress in AI. Hosted by: Stephanie Zhan and Sonya Huang, Sequoia Capital Mentioned in this episode: PPO: Proximal Policy Optimization algorithm developed by DeepMind in game environments. Also used by OpenAI for RLHF in ChatGPT. MuJoCo: Open source physics engine used to develop PPO Monte Carlo Tree Search: Heuristic search algorithm used in AlphaGo as well as video compression for YouTube and the self-driving system at Tesla AlphaZero: The DeepMind model that taught itself from scratch how to master the games of chess, shogi and Go MuZero: The DeepMind follow up to AlphaZero that mastered games without knowing the rules and able to plan winning strategies in unknown environments AlphaChem: Chemical Synthesis Planning with Tree Search and Deep Neural Network Policies DQN: Deep Q-Network, Introduced in 2013 paper, Playing Atari with Deep Reinforcement Learning AlphaFold: DeepMind model for predicting protein structures for which Demis Hassabis, John Jumper and David Baker won the 2024 Nobel Prize in Chemistry
Hugo Bowne-Anderson hosts a panel discussion from the MLOps World and Generative AI Summit in Austin, exploring the long-term growth of AI by distinguishing real problem-solving from trend-based solutions. If you're navigating the evolving landscape of generative AI, productionizing models, or questioning the hype, this episode dives into the tough questions shaping the field. The panel features: - Ben Taylor (Jepson) (https://www.linkedin.com/in/jepsontaylor/) – CEO and Founder at VEOX Inc., with experience in AI exploration, genetic programming, and deep learning. - Joe Reis (https://www.linkedin.com/in/josephreis/) – Co-founder of Ternary Data and author of Fundamentals of Data Engineering. - Juan Sequeda (https://www.linkedin.com/in/juansequeda/) – Principal Scientist and Head of AI Lab at Data.World, known for his expertise in knowledge graphs and the semantic web. The discussion unpacks essential topics such as: - The shift from prompt engineering to goal engineering—letting AI iterate toward well-defined objectives. - Whether generative AI is having an electricity moment or more of a blockchain trajectory. - The combinatorial power of AI to explore new solutions, drawing parallels to AlphaZero redefining strategy games. - The POC-to-production gap and why AI projects stall. - Failure modes, hallucinations, and governance risks—and how to mitigate them. - The disconnect between executive optimism and employee workload. Hugo also mentions his upcoming workshop on escaping Proof-of-Concept Purgatory, which has evolved into a Maven course "Building LLM Applications for Data Scientists and Software Engineers" launching in January (https://maven.com/hugo-stefan/building-llm-apps-ds-and-swe-from-first-principles?utm_campaign=8123d0&utm_medium=partner&utm_source=instructor). Vanishing Gradient listeners can get 25% off the course (use the code VG25), with $1,000 in Modal compute credits included. A huge thanks to Dave Scharbach and the Toronto Machine Learning Society for organizing the conference and to the audience for their thoughtful questions. As we head into the new year, this conversation offers a reality check amidst the growing AI agent hype. LINKS Hugo on twitter (https://x.com/hugobowne) Hugo on LinkedIn (https://www.linkedin.com/in/hugo-bowne-anderson-045939a5/) Vanishing Gradients on twitter (https://x.com/vanishingdata) "Building LLM Applications for Data Scientists and Software Engineers" course (https://maven.com/hugo-stefan/building-llm-apps-ds-and-swe-from-first-principles?utm_campaign=8123d0&utm_medium=partner&utm_source=instructor).
ChatGPT, AlphaZero, Deepfakes, selbstfahrende Autos – Künstliche Intelligenz ist aus unserem Alltag nicht mehr wegzudenken. KI erleichtert inzwischen in vielen Bereichen unser Leben und hilft bei Problemlösungen, wie etwa in der Medizin oder in der Landwirtschaft. Aber KI-erzeugte Fake News fluten auch das Netz, Deepfakes imitieren Politiker täuschend echt und legen ihnen Worte in den Mund, die sie nie gesagt haben. Hinzu kommt: KI verbraucht Unmengen an Energie für die komplexen Rechenprozesse, die dahinterstecken. Und Künstliche Intelligenz wird mit Daten trainiert, die wir zum Beispiel auf Social Media zur Verfügung stellen. Aber was ist dann mit dem Datenschutz? Was, wenn eine KI mit Hilfe von Gesichtserkennung Menschen identifiziert und klassifiziert und sie dadurch Nachteile haben, etwa im Beruf oder im Alltag? Ein Podcast über antike Vorstellungen von Künstlicher Intelligenz, den ersten Chatbot der Geschichte und die Frage: Versteht eine KI eigentlich Humor? Gesprächspartner*innen Mar Hicks Christopher Koska Adrienne Mayor Dinah Pfau Richard Socher Team Moderation: Mirko Drotschmann Sprecher*innen: Andrea Kath, Nils Kretschmer, Lauralie Schweiger Buch und Regie: objektiv media GmbH, Janine Funke und Andrea Kath Technik: Sascha Schiemann Musik: Sonoton Produktion: objektiv media GmbH im Auftrag des ZDF Redaktion ZDF: Katharina Kolvenbach Literatur Abbate, Janet (2000): Inventing the Internet (Inside Technology). Catani, Stephanie (Hrsg.) (2024): Handbuch Künstliche Intelligenz und die Künste. Dendorfer, Jürgen; Hochbruck, Wolfgang; Pape, Jessica (2024): Ritter Basisartikel: Ritterspiele: Das höfische Mittelalter als Geschichte und Projektion. Fischer, Ernst Peter (2023): Ein Scheiterhaufen der Wissenschaft: Die Großen an ihren Grenzen. Gutmann, Mathias; Wiegerling, Klaus; Rathgeber, Benjamin (Hrsg.) (2024): Handbuch Technikphilosophie. Hartmann, Doreen (2015): Zwischen Mathematik und Poesie. Leben und Werk von Ada Lovelace, in: Sybille Krämer (Hrsg.): Ada Lovelace. Die Pionierin der Computertechnik und ihre Nachfolgerinnen, S.15-33. Hicks, Mar (2017): Programmed Inequality: How Britain Discarded Women Technologists and Lost Its Edge in Computing (History of Computing). Klüver, Christina; Klüver, Jürgen (2022): Ewiges Leben durch künstliche Intelligenz und künstliche Gesellschaften. Koska, Christopher (2021): Ethik der Algorithmen. Auf der Suche nach Zahlen und Werten (Bd. 6). Menabrea, Luigi Frederico; Lovelace, Ada (1996), in: Grundriss der von Charles Babbage erfundenen Analytical Engine, S. 309-381. Mayor, Adrienne (2020): Götter und Maschinen. Wie die Antike das 21. Jahrhundert erfand. Project Metadata (2023): AI and Poetry. Settele, Veronika; Schmitt, Martin (2024): Cows and Computers. Electronic Data Processing in German Cattle Farming, 1960s-1990s. Weizenbaum, Joseph (1966): ELIZA—a computer program for the study of natural language communication between man and machine, in: Communications of the ACM, Volume 9, Issue 1, S.36-45. Internetquellen https://www.br.de/nachrichten/netzwelt/wenn-ki-freunde-zur-gefahr-werden-suizid-in-den-usa-zeigt-tragischen-verlauf-einer-ki-beziehung,USgb6Ux https://www.mpg.de/frauen-in-der-forschung/ada-lovelace https://www.swr.de/swrkultur/wissen/archivradio/frueheste-tonaufnahmen-100.html https://www.bbc.com/mundo/media-40632577 https://www.spiegel.de/netzwelt/web/john-mccarthy-der-vater-der-rechner-cloud-ist-tot-a-793795.html https://www.projekt-gutenberg.org/homer/ilias23/chap018.html https://www.portalkunstgeschichte.de/meldung/es_lebt__zur_geschichte_der_auto-6395.html https://www.technischesmuseum.at/museum/tmw-zine_-_unsere_storys/ki_zine/magazin_detail&j-cc-id=1625732690814&j-cc-node=magazineintrag&j-cc-name=hybrid-content
The machines are coming. Scratch that—they're already here: AIs that propose new combinations of ideas; chatbots that help us summarize texts or write code; algorithms that tell us who to friend or follow, what to watch or read. For a while the reach of intelligent machines may have seemed somewhat limited. But not anymore—or, at least, not for much longer. The presence of AI is growing, accelerating, and, for better or worse, human culture may never be the same. My guest today is Dr. Iyad Rahwan. Iyad directs the Center for Humans and Machines at the Max Planck Institute for Human Development in Berlin. Iyad is a bit hard to categorize. He's equal parts computer scientist and artist; one magazine profile described him as "the Anthropologist of AI." Labels aside, his work explores the emerging relationships between AI, human behavior, and society. In a recent paper, Iyad and colleagues introduced a framework for understanding what they call "machine culture." The framework offers a way of thinking about the different routes through which AI may transform—is transforming—human culture. Here, Iyad and I talk about his work as a painter and how he brings AI into the artistic process. We discuss whether AIs can make art by themselves and whether they may eventually develop good taste. We talk about how AIphaGoZero upended the world of Go and about how LLMs might be changing how we speak. We consider what AIs might do to cultural diversity. We discuss the field of cultural evolution and how it provides tools for thinking about this brave new age of machine culture. Finally, we discuss whether any spheres of human endeavor will remain untouched by AI influence. Before we get to it, a humble request: If you're enjoying the show—and it seems that many of you are—we would be ever grateful if you could let the world know. You might do this by leaving a rating or review on Apple Podcasts, or maybe a comment on Spotify. You might do this by giving us a shout out on the social media platform of your choice. Or, if you prefer less algorithmically mediated avenues, you might do this just by telling a friend about us face-to-face. We're hoping to grow the show and best way to do that is through listener endorsements and word of mouth. Thanks in advance, friends. Alright, on to my conversation with Iyad Rahwan. Enjoy! A transcript of this episode will be available soon. Notes and links 3:00 – Images from Dr. Rahwan's ‘Faces of Machine' portrait series. One of the portraits from the series serves as our tile art for this episode. 11:30 – The “stochastic parrots” term comes from an influential paper by Emily Bender and colleagues. 18:30 – A popular article about DALL-E and the “avocado armchair.” 21:30 – Ted Chiang's essay, “Why A.I. isn't going to make art.” 24:00 – An interview with Boris Eldagsen, who won the Sony World Photography Awards in March 2023 with an image that was later revealed to be AI-generated. 28:30 – A description of the concept of “science fiction science.” 29:00 – Though widely attributed to different sources, Isaac Asimov appears to have developed the idea that good science fiction predicts not the automobile, but the traffic jam. 30:00 – The academic paper describing the Moral Machine experiment. You can judge the scenarios for yourself (or design your own scenarios) here. 30:30 – An article about the Nightmare Machine project; an article about the Deep Empathy project. 37:30 – An article by Cesar Hidalgo and colleagues about the relationship between television/radio and global celebrity. 41:30 – An article by Melanie Mitchell (former guest!) on AI and analogy. A popular piece about that work. 42:00 – A popular article describing the study of whether AIs can generate original research ideas. The preprint is here. 46:30 – For more on AlphaGo (and its successors, AlphaGo Zero and AlphaZero), see here. 48:30 – The study finding that the novel of human Go playing increased due to the influence of AlphaGo. 51:00 – A blogpost delving into the idea that ChatGPT overuses certain words, including “delve.” A recent preprint by Dr. Rahwan and colleagues, presenting evidence that “delve” (and other words overused by ChatGPT) are now being used more in human spoken communication. 55:00 – A paper using simulations to show how LLMs can “collapse” when trained on data that they themselves generated. 1:01:30 – A review of the literature on filter bubbles, echo chambers, and polarization. 1:02:00 – An influential study by Dr. Chris Bail and colleagues suggesting that exposure to opposing views might actually increase polarization. 1:04:30 – A book by Geoffrey Hodgson and Thorbjørn Knudsen, who are often credited with developing the idea of “generalized Darwinism” in the social sciences. 1:12:00 – An article about Google's NotebookLM podcast-like audio summaries. 1:17:3 0 – An essay by Ursula LeGuin on children's literature and the Jungian “shadow.” Recommendations The Secret of Our Success, Joseph Henrich “Machine Behaviour,” Iyad Rahwan et al. Many Minds is a project of the Diverse Intelligences Summer Institute, which is made possible by a generous grant from the John Templeton Foundation to Indiana University. The show is hosted and produced by Kensy Cooperrider, with help from Assistant Producer Urte Laukaityte and with creative support from DISI Directors Erica Cartmill and Jacob Foster. Our artwork is by Ben Oldroyd. Our transcripts are created by Sarah Dopierala. Subscribe to Many Minds on Apple, Stitcher, Spotify, Pocket Casts, Google Play, or wherever you listen to podcasts. You can also now subscribe to the Many Minds newsletter here! We welcome your comments, questions, and suggestions. Feel free to email us at: manymindspodcast@gmail.com. For updates about the show, visit our website or follow us on Twitter (@ManyMindsPod) or Bluesky (@manymindspod.bsky.social).
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: OpenAI o1, Llama 4, and AlphaZero of LLMs, published by Vladimir Nesov on September 14, 2024 on LessWrong. GPT-4 level open weights models like Llama-3-405B don't seem capable of dangerous cognition. OpenAI o1 demonstrates that a GPT-4 level model can be post-trained into producing useful long horizon reasoning traces. AlphaZero shows how capabilities can be obtained from compute alone, with no additional data. If there is a way of bringing these together, the apparent helplessness of the current generation of open weights models might prove misleading. Post-training is currently a combination of techniques that use synthetic data and human labeled data. Human labeled data significantly improves quality, but its collection is slow and scales poorly. Synthetic data is an increasingly useful aspect of post-training, and automated aspects of its generation scale easily. Unlike weaker models, GPT-4 level LLMs clearly pass reading comprehension on most occasions, OpenAI o1 improves on this further. This suggests that at some point human data might become mostly unnecessary in post-training, even if it still slightly helps. Without it, post-training becomes automated and gets to use more compute, while avoiding the need for costly and complicated human labeling. A pretrained model at the next level of scale, such as Llama 4, if made available in open weights, might initially look approximately as tame as current models. OpenAI o1 demonstrates that useful post-training for long sequences of System 2 reasoning is possible. In the case of o1 in particular, this might involve a lot of human labeling, making its reproduction a very complicated process (at least if the relevant datasets are not released, and the reasoning traces themselves are not leaked in large quantities). But if some generally available chatbots at the next level of scale are good enough at automating labeling, this complication could be sidestepped, with o1 style post-training cheaply reproduced on top of a previously released open weights model. So there is an overhang in an open weights model that's distributed without long horizon reasoning post-training, since applying such post-training significantly improves its capabilities, making perception of its prior capabilities inadequate. The problem right now is that a new level of pretraining scale is approaching in the coming months, while ability to cheaply apply long horizon reasoning post-training might follow shortly thereafter, possibly unlocked by these very same models at the new level of pretraining scale (since it might currently be too expensive for most actors to implement, or to do enough experiments to figure out how). The resulting level of capabilities is currently unknown, and could well remain unknown outside the leading labs until after the enabling artifacts of the open weights pretrained models at the next level of scale have already been published. Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org
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: Evidence against Learned Search in a Chess-Playing Neural Network, published by p.b. on September 14, 2024 on LessWrong. Introduction There is a new paper and lesswrong post about "learned look-ahead in a chess-playing neural network". This has long been a research interest of mine for reasons that are well-stated in the paper: Can neural networks learn to use algorithms such as look-ahead or search internally? Or are they better thought of as vast collections of simple heuristics or memorized data? Answering this question might help us anticipate neural networks' future capabilities and give us a better understanding of how they work internally. and further: Since we know how to hand-design chess engines, we know what reasoning to look for in chess-playing networks. Compared to frontier language models, this makes chess a good compromise between realism and practicality for investigating whether networks learn reasoning algorithms or rely purely on heuristics. So the question is whether Francois Chollet is correct with transformers doing "curve fitting" i.e. memorisation with little generalisation or whether they learn to "reason". "Reasoning" is a fuzzy word, but in chess you can at least look for what human players call "calculation", that is the ability to execute moves solely in your mind to observe and evaluate the resulting position. To me this is a crux as to whether large language models will scale to human capabilities without further algorithmic breakthroughs. The paper's authors, which include Erik Jenner and Stuart Russell, conclude that the policy network of Leela Chess Zero (a top engine and open source replication of AlphaZero) does learn look-ahead. Using interpretability techniques they "find that Leela internally represents future optimal moves and that these representations are crucial for its final output in certain board states." While the term "look-ahead" is fuzzy, the paper clearly intends to show that the Leela network implements an "algorithm" and a form of "reasoning". My interpretation of the presented evidence is different, as discussed in the comments of the original lesswrong post. I argue that all the evidence is completely consistent with Leela having learned to recognise multi-move patterns. Multi-move patterns are just complicated patterns that take into account that certain pieces will have to be able to move to certain squares in future moves for the pattern to hold. The crucial different to having learned an algorithm: An algorithm can take different inputs and do its thing. That allows generalisation to unseen or at least unusual inputs. This means that less data is necessary for learning because the generalisation power is much higher. Learning multi-move patterns on the other hand requires much more data because the network needs to see many versions of the pattern until it knows all specific details that have to hold. Analysis setup Unfortunately it is quite difficult to distinguish between these two cases. As I argued: Certain information is necessary to make the correct prediction in certain kinds of positions. The fact that the network generally makes the correct prediction in these types of positions already tells you that this information must be processed and made available by the network. The difference between lookahead and multi-move pattern recognition is not whether this information is there but how it got there. However, I propose an experiment, that makes it clear that there is a difference. Imagine you train the model to predict whether a position leads to a forced checkmate and also the best move to make. You pick one tactical motive and erase it from the checkmate prediction part of the training set, but not the move prediction part. Now the model still knows which the right moves are to make i.e. it would pl...
David Silver is a principal research scientist at DeepMind and a professor at University College London. This interview was recorded at UMass Amherst during RLC 2024. References Discovering Reinforcement Learning Algorithms, Oh et al -- His keynote at RLC 2024 referred to more recent update to this work, yet to be published Mastering Chess and Shogi by Self-Play with a General Reinforcement Learning Algorithm, Silver et al 2017 -- the AlphaZero algo was used in his recent work on AlphaProof AlphaProof on the DeepMind blog AlphaFold on the DeepMind blog Reinforcement Learning Conference 2024 David Silver on Google Scholar
Sentetik verilerle dolu bir dünya nasıl olur. "İnsanlık bir gün yok olup giderse, bizden geriye kalan, bir modelin devasa matrislerinden birinin köşesindeki sayılar olacak.".Konular:(01:52) Ölü İnternet(03:51) Invasion of the Body Snatchers(05:09) Bad Bot Report(06:39) Agents(08:54) Sentetik İnternet(10:22) Telif hakkı(13:10) Alphazero(16:06) Patreon TeşekkürleriKaynaklar:Yazı: Maybe You Missed It, but the Internet ‘Died' Five Years Ago(PDF) Will we run out of data? An analysis of the limits of scaling datasets in Machine LearningMakale: AI models collapse when trained on recursively generated dataA faster, systematic way to train large language models for enterpriseFilm: Invasion of the Body Snatchers, The Thing, The FacultySee Privacy Policy at https://art19.com/privacy and California Privacy Notice at https://art19.com/privacy#do-not-sell-my-info.
LLMs are democratizing digital intelligence, but we're all waiting for AI agents to take this to the next level by planning tasks and executing actions to actually transform the way we work and live our lives. Yet despite incredible hype around AI agents, we're still far from that “tipping point” with best in class models today. As one measure: coding agents are now scoring in the high-teens % on the SWE-bench benchmark for resolving GitHub issues, which far exceeds the previous unassisted baseline of 2% and the assisted baseline of 5%, but we've still got a long way to go. Why is that? What do we need to truly unlock agentic capability for LLMs? What can we learn from researchers who have built both the most powerful agents in the world, like AlphaGo, and the most powerful LLMs in the world? To find out, we're talking to Misha Laskin, former research scientist at DeepMind. Misha is embarking on his vision to build the best agent models by bringing the search capabilities of RL together with LLMs at his new company, Reflection AI. He and his cofounder Ioannis Antonoglou, co-creator of AlphaGo and AlphaZero and RLHF lead for Gemini, are leveraging their unique insights to train the most reliable models for developers building agentic workflows. Hosted by: Stephanie Zhan and Sonya Huang, Sequoia Capital 00:00 Introduction 01:11 Leaving Russia, discovering science 10:01 Getting into AI with Ioannis Antonoglou 15:54 Reflection AI and agents 25:41 The current state of Ai agents 29:17 AlphaGo, AlphaZero and Gemini 32:58 LLMs don't have a ground truth reward 37:53 The importance of post-training 44:12 Task categories for agents 45:54 Attracting talent 50:52 How far away are capable agents? 56:01 Lightning round Mentioned: The Feynman Lectures on Physics: The classic text that got Misha interested in science. Mastering the game of Go with deep neural networks and tree search: The original 2016 AlphaGo paper. Mastering the game of Go without human knowledge: 2017 AlphaGo Zero paper Scaling Laws for Reward Model Overoptimization: OpenAI paper on how reward models can be gamed at all scales for all algorithms. Mapping the Mind of a Large Language Model: Article about Anthropic mechanistic interpretability paper that identifies how millions of concepts are represented inside Claude Sonnet Pieter Abeel: Berkeley professor and founder of Covariant who Misha studied with A2C and A3C: Advantage Actor Critic and Asynchronous Advantage Actor Critic, the two algorithms developed by Misha's manager at DeepMind, Volodymyr Mnih, that defined reinforcement learning and deep reinforcement learning
A Note from James:Today, we have a fascinating story of resilience, transformation, and triumph. Imagine being the second-best player in the world in your field, only to see your ranking plummet over the years. Hikaru Nakamura experienced just that in the world of chess. Once the number two player globally, he saw his ranking drop below the top 20. But Hikaru didn't let this define him. During the pandemic, he started streaming chess online, building a massive audience and diversifying his interests. And remarkably, he climbed back to number two in the world. How did he do it? What changes in mindset and strategy led to this incredible comeback? We'll cover all this and more in our discussion today. This conversation is not just for chess enthusiasts but for anyone looking to succeed and find balance in life. So, here is Hikaru Nakamura."Episode Description:In this episode of The James Altucher Show, James talks with Hikaru Nakamura about his extraordinary journey in the world of chess. Hikaru shares the ups and downs of his career, detailing how he went from being the second-best chess player in the world to falling out of the top 20. During the COVID-19 pandemic, Hikaru pivoted to streaming chess online, which not only revitalized his career but also brought him back to the top echelons of the chess world. This episode delves into the mindset shifts, strategies, and the role of diversification in achieving success and maintaining mental well-being. Hikaru's story is a testament to resilience and adaptability, offering insights that are valuable for anyone aiming to excel in their field.What You'll Learn:The psychological challenges of being a top-ranked chess player and how to overcome them.How diversifying interests and income streams can alleviate career pressures.The impact of streaming and online presence on professional success.The importance of maintaining a balanced mindset in competitive environments.Strategies for turning setbacks into comebacks in any career.Chapters:00:00 The Rise and Fall of Hikaru Nakamura01:10 Hikaru's Streaming Journey Begins01:51 Return to Competitive Chess02:24 The Psychology of Chess05:23 The Impact of Computers on Chess10:57 Hikaru's Career Challenges24:14 The Turning Point: Streaming and the Pandemic35:57 Transitioning to Live Streaming36:17 Changing Perceptions of Chess Players37:21 Cultural Significance of Chess38:02 Making Chess More Accessible38:48 Strategic Decisions in Content Creation41:37 Engaging the Audience44:10 Pandemic and the Rise of Chess Streaming47:17 Building a Streaming Business55:26 Diversifying Income and Investments01:06:49 Maintaining Chess Skills01:10:45 Unexpected Opening Strategy01:11:13 The Power of Unpredictability01:11:47 Evolving Chess Study Methods01:13:20 Memory and Age in Chess01:14:22 Generational Gaps in Chess01:16:07 Impact of the Chess Boom01:18:02 Emotional Moments in Chess01:22:45 Future Aspirations Beyond Chess01:24:02 Dealing with Online Trolls01:26:01 Breaking Chess Rules with Computers01:29:55 Challenges of Modern Chess01:39:39 Final Thoughts and ReflectionsAdditional Resources:Hikaru Nakamura's Twitch ChannelHikaru Nakamura's YouTube ChannelThe Lex Fridman Podcast with Hikaru Nakamura“The Queen's Gambit” on NetflixInternational Chess Federation (FIDE)This episode is a must-listen for anyone interested in understanding how to navigate the ups and downs of a professional career, leveraging new opportunities, and achieving personal growth. Tune in to hear Hikaru's inspiring story and gain insights that could transform your approach to success. ------------What do YOU think of the show? Head to JamesAltucherShow.com/listeners and fill out a short survey that will help us better tailor the podcast to our audience!Are you interested in getting direct answers from James about your question on a podcast? Go to JamesAltucherShow.com/AskAltucher and send in your questions to be answered on the air!------------Visit Notepd.com to read our idea lists & sign up to create your own!My new book, Skip the Line, is out! Make sure you get a copy wherever books are sold!Join the You Should Run for President 2.0 Facebook Group, where we discuss why you should run for President.I write about all my podcasts! Check out the full post and learn what I learned at jamesaltuchershow.com------------Thank you so much for listening! If you like this episode, please rate, review, and subscribe to “The James Altucher Show” wherever you get your podcasts: Apple PodcastsiHeart RadioSpotifyFollow me on social media:YouTubeTwitterFacebookLinkedIn
For the latest episode of SparX, Mukesh Bansall, Founder of Myntra and Cult.fit, is in conversation with India's first Grandmaster, Viswanathan Anand. Viswanathan is a five-time World Chess Champion. Anand's most significant achievements include being a five-time World Chess Champion, winning the World Rapid Chess Championship, and the World Blitz Chess Championship. Additionally, his victories in prestigious tournaments like Linares and Corus are highly notable in the chess world. Viswanathan revolutionised the chess landscape in India. His victories brought national pride and global recognition, inspiring a surge in young players and increased investment in chess infrastructure. Anand's success fostered a vibrant chess culture, making India a formidable force in the sport. Join us for an inspiring conversation with the Grandmaster on his journey with chess, some stories with other chess players, the evolution of chess, and his experiences throughout his career. Resource List: About Viswanathan Anand: https://en.wikipedia.org/wiki/Viswanathan_Anand Viswanathan Anand FIDE Profile: https://ratings.fide.com/profile/5000017 Google DeepMind Blog Article: https://deepmind.google/discover/blog/alphazero-shedding-new-light-on-chess-shogi-and-go/ AlphaZero: https://www.chess.com/terms/alphazero-chess-engine All India Chess Federation: https://aicf.in/ International Chess Federation: https://www.fide.com/ Chess In India: https://en.wikipedia.org/wiki/Chess_in_India Indian Chess Players: https://en.wikipedia.org/wiki/List_of_Indian_chess_players About Garry Kasparov: https://en.wikipedia.org/wiki/Garry_Kasparov About Anatoly Karpov: https://en.wikipedia.org/wiki/Anatoly_Karpov About Vladmir Kramnik: https://en.wikipedia.org/wiki/Vladimir_Kramnik About Mikhail Gurevich: https://en.wikipedia.org/wiki/Mikhail_Gurevich_(chess_player) What is ZugZwang?: https://www.chess.com/article/view/what-is-zugzwang-chess-terms
Send us a Text Message.Discover the extraordinary journey of Hernan Londano, a renowned AI strategist at Dell Technologies, as he takes us from the mountains of Colombia to the forefront of technological innovation in the United States. Hernan's nearly three decades of experience, including his pivotal roles as Chief Technology Officer and Chief Information Security Officer at Barry University, provide a unique lens into the world of AI governance. This episode promises to enlighten you on how to balance cutting-edge AI innovations with ethical responsibility, ensuring alignment with strategic organizational goals.We venture into the provocative realm of AI's role in strategic games and its real-world ramifications. Hernan discusses the historic matches of AlphaGo and AlphaZero, highlighting AI's unprecedented capabilities in generating novel strategies. But the conversation takes a serious turn as we address the ethical and responsible use of AI in military contexts, where decisions can literally be a matter of life and death. This chapter emphasizes the necessity for inclusive, diverse discussions and continuously evolving governance frameworks to ethically harness AI's power.The complexities of AI, free speech, and legal ramifications unfold as we dissect a real-world incident involving OpenAI and Scarlett Johansson. This episode doesn't shy away from addressing the urgent need for clear guidelines and regulations as AI technology continues to permeate our daily lives. Hernan also explores the transformative impact of AI across industries, from AI-generated music to business strategy, urging the development of governance frameworks that ensure transparency, accountability, and equity. Join us for this compelling exploration of AI's future and its governance within business and beyond.Thanks for tuning in to this episode of Follow The Brand! We hope you enjoyed learning about the latest marketing trends and strategies in Personal Branding, Business and Career Development, Financial Empowerment, Technology Innovation, and Executive Presence. To keep up with the latest insights and updates from us, be sure to follow us at 5starbdm.com. See you next time on Follow The Brand!
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: Visualizing neural network planning, published by Nevan Wichers on May 9, 2024 on The AI Alignment Forum. TLDR We develop a technique to try and detect if a NN is doing planning internally. We apply the decoder to the intermediate representations of the network to see if it's representing the states it's planning through internally. We successfully reveal intermediate states in a simple Game of Life model, but find no evidence of planning in an AlphaZero chess model. We think the idea won't work in its current state for real world NNs because they use higher-level, abstract representations for planning that our current technique cannot decode. Please comment if you have ideas that may work for detecting more abstract ways the NN could be planning. Idea and motivation To make safe ML, it's important to know if the network is performing mesa optimization, and if so, what optimization process it's using. In this post, I'll focus on a particular form of mesa optimization: internal planning. This involves the model searching through possible future states and selecting the ones that best satisfy an internal goal. If the network is doing internal planning, then it's important the goal it's planning for is aligned with human values. An interpretability technique which could identify what states it's searching through would be very useful for safety. If the NN is doing planning it might represent the states it's considering in that plan. For example, if predicting the next move in chess, it may represent possible moves it's considering in its hidden representations. We assume that NN is given the representation of the environment as input and that the first layer of the NN encodes the information into a hidden representation. Then the network has hidden layers and finally a decoder to compute the final output. The encoder and decoder are trained as an autoencoder, so the decoder can reconstruct the environment state from the encoder output. Language models are an example of this where the encoder is the embedding lookup. Our hypothesis is that the NN may use the same representation format for states it's considering in its plan as it does for the encoder's output. Our idea is to apply the decoder to the hidden representations at different layers to decode them. If our hypothesis is correct, this will recover the states it considers in its plan. This is similar to the Logit Lens for LLMs, but we're applying it here to investigate mesa-optimization. A potential pitfall is that the NN uses a slightly different representation for the states it considers during planning than for the encoder output. In this case, the decoder won't be able to reconstruct the environment state it's considering very well. To overcome this, we train the decoder to output realistic looking environment states given the hidden representations by training it like the generator in a GAN. Note that the decoder isn't trained on ground truth environment states, because we don't know which states the NN is considering in its plan. Game of Life proof of concept (code) We consider an NN trained to predict the number of living cells after the Nth time step of the Game of Life (GoL). We chose the GoL because it has simple rules, and the NN will probably have to predict the intermediate states to get the final cell count. This NN won't do planning, but it may represent the intermediate states of the GoL in its hidden states. We use an LSTM architecture with an encoder to encode the initial GoL state, and a "count cells NN" to output the number of living cells after the final LSTM output. Note that training the NN to predict the number of alive cells at the final state makes this more difficult for our method than training the network to predict the final state since it's less obvious that the network will predict t...
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: LLMs seem (relatively) safe, published by JustisMills on April 26, 2024 on LessWrong. Post for a somewhat more general audience than the modal LessWrong reader, but gets at my actual thoughts on the topic. In 2018 OpenAI defeated the world champions of Dota 2, a major esports game. This was hot on the heels of DeepMind's AlphaGo performance against Lee Sedol in 2016, achieving superhuman Go performance way before anyone thought that might happen. AI benchmarks were being cleared at a pace which felt breathtaking at the time, papers were proudly published, and ML tools like Tensorflow (released in 2015) were coming online. To people already interested in AI, it was an exciting era. To everyone else, the world was unchanged. Now Saturday Night Live sketches use sober discussions of AI risk as the backdrop for their actual jokes, there are hundreds of AI bills moving through the world's legislatures, and Eliezer Yudkowsky is featured in Time Magazine. For people who have been predicting, since well before AI was cool (and now passe), that it could spell doom for humanity, this explosion of mainstream attention is a dark portent. Billion dollar AI companies keep springing up and allying with the largest tech companies in the world, and bottlenecks like money, energy, and talent are widening considerably. If current approaches can get us to superhuman AI in principle, it seems like they will in practice, and soon. But what if large language models, the vanguard of the AI movement, are actually safer than what came before? What if the path we're on is less perilous than what we might have hoped for, back in 2017? It seems that way to me. LLMs are self limiting To train a large language model, you need an absolutely massive amount of data. The core thing these models are doing is predicting the next few letters of text, over and over again, and they need to be trained on billions and billions of words of human-generated text to get good at it. Compare this process to AlphaZero, DeepMind's algorithm that superhumanly masters Chess, Go, and Shogi. AlphaZero trains by playing against itself. While older chess engines bootstrap themselves by observing the records of countless human games, AlphaZero simply learns by doing. Which means that the only bottleneck for training it is computation - given enough energy, it can just play itself forever, and keep getting new data. Not so with LLMs: their source of data is human-produced text, and human-produced text is a finite resource. The precise datasets used to train cutting-edge LLMs are secret, but let's suppose that they include a fair bit of the low hanging fruit: maybe 5% of publicly available text that is in principle available and not garbage. You can schlep your way to a 20x bigger dataset in that case, though you'll hit diminishing returns as you have to, for example, generate transcripts of random videos and filter old mailing list threads for metadata and spam. But nothing you do is going to get you 1,000x the training data, at least not in the short run. Scaling laws are among the watershed discoveries of ML research in the last decade; basically, these are equations that project how much oomph you get out of increasing the size, training time, and dataset that go into a model. And as it turns out, the amount of high quality data is extremely important, and often becomes the bottleneck. It's easy to take this fact for granted now, but it wasn't always obvious! If computational power or model size was usually the bottleneck, we could just make bigger and bigger computers and reliably get smarter and smarter AIs. But that only works to a point, because it turns out we need high quality data too, and high quality data is finite (and, as the political apparatus wakes up to what's going on, legally fraught). There are rumbling...
He is an economist with the soul of a poet. He has studied number theory and is an expert on policy. He has studied Urdu and and dreams in shairi. Rohit Lamba joins Amit Varma in episode 378 of The Seen and the Unseen to discuss economics, politics, society and our human condition. (FOR FULL LINKED SHOW NOTES, GO TO SEENUNSEEN.IN.) Also check out: 1. Rohit Lamba links at Penn State, LinkedIn, Twitter, Google Scholar, YouTube and his own website. 2. Breaking the Mould: Reimagining India's Economic Future -- Raghuram Rajan and Rohit Lamba. 3. The Broken Script -- Swapna Liddle. 4. Swapna Liddle and the Many Shades of Delhi -- Episode 367 of The Seen and the Unseen. 5. Six More Stories That Should Be Films -- Episode 43 of Everything is Everything, which includes a chapter inspired by Swapna Liddle's book. 6. Wanderers, Kings, Merchants: The Story of India through Its Languages — Peggy Mohan. 7. Understanding India Through Its Languages — Episode 232 of The Seen and the Unseen (w Peggy Mohan). 8. The Life and Times of Ira Pande -- Episode 369 of The Seen and the Unseen. 9. The Price of Peace: Money, Democracy, and the Life of John Maynard Keynes -- Zachary D. Carter. 10. Fixing the Knowledge Society -- Episode 24 of Everything is Everything. 11. Robert Sapolsky's biology lectures on YouTube. 12. Episode of The Seen and the Unseen with Ramachandra Guha: 1, 2, 3, 4, 5, 6. 13. The Nurture Assumption — Judith Rich Harris. 14. Deepak VS and the Man Behind His Face -- Episode 373 of The Seen and the Unseen. 15. The Incredible Insights of Timur Kuran -- Episode 349 of The Seen and the Unseen. 16. Private Truths, Public Lies — Timur Kuran. 17. The Gentle Wisdom of Pratap Bhanu Mehta -- Episode 300 of The Seen and the Unseen. 18. 300 Ramayanas — AK Ramanujan. 19. Ramcharitmanas -- Tulsidas. 20. Savarkar and the Making of Hindutva -- Janaki Bakhle. 21. The Intellectual Foundations of Hindutva — Episode 115 of The Seen and the Unseen (w Aakar Patel). 22. Political Ideology in India — Episode 131 of The Seen and the Unseen (w Rahul Verma). 23. Religion and Ideology in Indian Society — Episode 124 of The Seen and the Unseen (w Suyash Rai). 24. Gita Press and the Making of Hindu India — Akshaya Mukul. 25. The Gita Press and Hindu Nationalism — Episode 139 of The Seen and the Unseen (w Akshaya Mukul). 26. India After Gandhi -- Ramachandra Guha. 27. Amitava Kumar Finds the Breath of Life — Episode 265 of The Seen and the Unseen. 28. Aadha Gaon — Rahi Masoom Raza. 29. The Rooted Cosmopolitanism of Sugata Srinivasaraju — Episode 277 of The Seen and the Unseen. 30. Postcard from Kashmir -- Agha Shahid Ali. 31. The Veiled Suite: The Collected Poems -- Agha Shahid Ali. 32. You Can Always Get There From Here -- Mark Strand. 33. Collected Poems — Mark Strand. 34. Variants of chess on chess.com. 35. The Tamilian gentleman who took on the world — Amit Varma on Viswanathan Anand. 36. The New World Upon Us — Amit Varma on Alpha Zero. 37. The Heisenberg Uncertainty Principle. 38. The History of the Planning Commission -- Episode 306 of The Seen and the Unseen (w Nikhil Menon). 39. The Life and Times of KP Krishnan -- Episode 355 of The Seen and the Unseen. 40. The Reformers -- Episode 28 of Everything is Everything. 41. Milton Friedman on Minimum Wage Laws. 42. Main Gautam Nahin Hoon -- Khalilur Rahman Azmi. 43. Lessons from Nirala's ballad for our battle with covid -- Rohit Lamba. 44. Poker and Life -- Episode 38 of Everything is Everything. 45. Range Rover — The archives of Amit Varma's column on poker for the Economic Times. 46. What is Populism? — Jan-Werner Müller. 47. The Populist Playbook -- Episode 42 of Everything is Everything. 48. The Tragedy of Our Farm Bills — Episode 211 of The Seen and the Unseen (w Ajay Shah). 49. Dynamism with Incommensurate Development: The Distinctive Indian Model -- Rohit Lamba and Arvind Subramanian. 50. List of Soviet and Russian leaders by height. 51. Narendra Modi takes a Great Leap Backwards — Amit Varma on Demonetisation. 52. Beware of the Useful Idiots — Amit Varma. 53. Number Theory. 54. Fermat's Last Theorem. 55. A Beautiful Mind -- Ron Howard. 56. The Life and Work of Ashwini Deshpande — Episode 298 of The Seen and the Unseen. 57. Dilip José Abreu: an elegant and creative economist -- Rohit Lamba. 58. The BJP Before Modi — Episode 202 of The Seen and the Unseen (w Vinay Sitapati). 59. The Forgotten Greatness of PV Narasimha Rao -- Episode 283 of The Seen and the Unseen (w Vinay Sitapati). 60. Ghummakkad Shastra -- Rahul Sankrityayan. 61. Jahnavi and the Cyclotron — Episode 319 of The Seen and the Unseen (w Jahnavi Phalkey). 62. The Looking-Glass Self. 63. Jo Bhi Main -- Song from Rockstar with lyrics by Irshad Kamil. 64. Ranjit Hoskote is Dancing in Chains — Episode 363 of The Seen and the Unseen. 65. Politically correct, passive-aggressive: How Indians in the US struggle to decode corporate speak -- Anahita Mukherji. 66. Lincoln -- Steven Spielberg. 67. The Life and Times of Montek Singh Ahluwalia — Episode 285 of The Seen and the Unseen. 68. The Economics and Politics of Vaccines — Episode 223 of The Seen and the Unseen (w Ajay Shah). 69. In Service of the Republic — Vijay Kelkar & Ajay Shah. 70. The Semiconductor Wars — Episode 358 of The Seen and the Unseen (w Pranay Kotasthane & Abhiram Manchi). 71. The Smile Curve. 72. Urban Governance in India — Episode 31 of The Seen and the Unseen (w Shruti Rajagopalan). 73. We Are Fighting Two Disasters: Covid-19 and the Indian State — Amit Varma. 74. The Child and the State in India -- Myron Weiner. 75. Where India Goes -- Diane Coffey and Deam Spears. 76. What's Wrong With Indian Agriculture? -- Episode 18 of Everything is Everything. 77. South India Would Like to Have a Word — Episode 320 of The Seen and the Unseen (w Nilakantan RS). 78. South vs North: India's Great Divide — Nilakantan RS. 79. Episodes of The Seen and the Unseen with Ashwin Mahesh: 1, 2, 3. 80. Maximum City -- Suketu Mehta. 81. Disgrace -- JM Coetzee. 82. Snow -- Pamuk. 83. Bahut Door, Kitna Door Hota Hai -- Manav Kaul. 84. Shakkar Ke Paanch Dane -- Manav Kaul.. 85. Poems: 1962–2020 -- Louise Glück. 86. Mahabharata. 87. राम की शक्ति-पूजा -- सूर्यकांत त्रिपाठी निराला. 88. Iqbal and Ahmad Faraz on Rekhta. 89. Ranjish Hi Sahi -- Ahmad Faraz. 90. Zindagi Se Yahi Gila Hai Mujhe -- Ahmad Faraz. 91. AR Rahman on Wikipedia and Spotify. This episode is sponsored by CTQ Compounds. Check out The Daily Reader and FutureStack. Use the code UNSEEN for Rs 2500 off. Amit's newsletter is explosively active again. Subscribe right away to The India Uncut Newsletter! It's free! Amit Varma and Ajay Shah have launched a new video podcast. Check out Everything is Everything on YouTube. Check out Amit's online course, The Art of Clear Writing. Episode art: ‘Pick a Tree' by Simahina.
Are LLMs stochastic parrots or reflection of our own intelligence? In this episode of Navigating Major Programmes, Riccardo Cosentino sits down with Lawrence Rowland for an extremely candid conversation surrounding the adoption of artificial intelligence, in major programmes and beyond. AI skeptics and AI enthusiasts alike, this episode was recorded for you. “None of us are keeping up, none of us know what the hell is going on. So, if you can kind of just relax and enjoy it happening, you will also help everyone else so much more. Enjoy it. And enjoy what [AI] is telling us about us.” –Lawrence Rowland Lawrence began as an engineer on large capital projects with WSP and Motts, before moving onto Bechtel and Booz Allen. He spent ten years in project and portfolio management with CPC and Pcubed, before transitioning to data analytics and AI for projects, working originally for Projecting Success, and now for React AI. He now helps project services firms find relevant immediate AI applications for their business. Key Takeaways:Large Language Model (LLM) 101What is an AI agent? What is the principal-agent problem (PAP)?What LLMs can teach you about your own thinking patternsThe future of Google Gemini and AI adoption in generalThe weaknesses of the generative AI of today Mentioned Links:A Path Towards Autonomous Machine IntelligencePrincipal Agent ProblemApplied Category TheoryWisdom of CrowdsState Space Models and MambaDemis Hassabis and the return of alpha zero type tree search and RL If you enjoyed this episode, make sure and give us a five star rating and leave us a review on iTunes, Podcast Addict, Podchaser or Castbox. The conversation doesn't stop here—connect and converse with our LinkedIn community: Follow Navigating Major Programmes on LinkedInFollow Riccardo Cosentino on LinkedInRead Riccardo's latest at wwww.riccardocosentino.com Music: "A New Tomorrow" by Chordial Music. Licensed through PremiumBeat.DISCLAIMER: The opinions, beliefs, and viewpoints expressed by the hosts and guests on this podcast do not necessarily represent or reflect the official policy, opinions, beliefs, and viewpoints of Disenyo.co LLC and its employees.
In this episode, Jimmy Purdy and Matt Fanslow discuss the intersections of psychology, business strategies, and the nuances of the automotive industry. Matt draws parallels between the strategies used in Brazilian jiu-jitsu and those needed for business success, emphasizing the critical nature of long-term thinking and adaptability. Jimmy explores the concept of networking beyond the industry, recounting how his participation in a local jiu-jitsu dojo has led to unexpected business opportunities. Moreover, they candidly discuss the impact of substance abuse on technicians, with Matt sharing personal reflections on the importance of addressing underlying mental health concerns. 00:00 Uncovering untold stories and shared human experiences.09:32 Transition from want to need, alcohol reliance.12:50 Personal growth and self-reflection lead to fulfillment.21:08 Struggle to resist dropping prices for clients.22:17 Summarizing the text in 7 words: Quantifying client's situation and vehicle diagnostics.27:56 Offering remanufactured engine with nationwide warranty and care.32:23 Customers were misinformed about the car repair process.42:31 Car fixed by thorough diagnostic or trial-and-error.44:16 Optimize work, prioritize profit, and avoid burnout.50:06 Sudden loss of the entire team creates frustration.57:15 Jiu-jitsu is intense and requires focus.01:02:26 Alphazero chess engine becomes the world's strongest quickly.01:06:50 Challenging to gain new perspectives in business.01:11:01 Building community, avoiding bias, and seeking diversity. Thanks to our sponsor, Shop Boss! See how they can simplify your auto shop HERE
The Robots are coming! We are talking about the latest in GPRs (general purpose humanoid robots), Apple cancelling their car project, Gemini 1.5 Pro testing, Biden's plan to ban voice impersonation, the decline of TV viewership, and Deep Mind CEO Demis Hassabis' views on AlphaZero sitting atop LLMs on the AGI stack.
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: Notes on Dwarkesh Patel's Podcast with Demis Hassabis, published by Zvi on March 2, 2024 on LessWrong. Demis Hassabis was interviewed twice this past week. First, he was interviewed on Hard Fork. Then he had a much more interesting interview with Dwarkesh Patel. This post covers my notes from both interviews, mostly the one with Dwarkesh. Hard Fork Hard Fork was less fruitful, because they mostly asked what for me are the wrong questions and mostly get answers I presume Demis has given many times. So I only noticed two things, neither of which is ultimately surprising. They do ask about The Gemini Incident, although only about the particular issue with image generation. Demis gives the generic 'it should do what the user wants and this was dumb' answer, which I buy he likely personally believes. When asked about p(doom) he expresses dismay about the state of discourse and says around 42:00 that 'well Geoffrey Hinton and Yann LeCun disagree so that indicates we don't know, this technology is so transformative that it is unknown. It is nonsense to put a probability on it. What I do know is it is non-zero, that risk, and it is worth debating and researching carefully… we don't want to wait until the eve of AGI happening.' He says we want to be prepared even if the risk is relatively small, without saying what would count as small. He also says he hopes in five years to give us a better answer, which is evidence against him having super short timelines. I do not think this is the right way to handle probabilities in your own head. I do think it is plausibly a smart way to handle public relations around probabilities, given how people react when you give a particular p(doom). I am of course deeply disappointed that Demis does not think he can differentiate between the arguments of Geoffrey Hinton versus Yann LeCun, and the implied importance on the accomplishments and thus implied credibility of the people. He did not get that way, or win Diplomacy championships, thinking like that. I also don't think he was being fully genuine here. Otherwise, this seemed like an inessential interview. Demis did well but was not given new challenges to handle. Dwarkesh Patel Demis Hassabis also talked to Dwarkesh Patel, which is of course self-recommending. Here you want to pay attention, and I paused to think things over and take detailed notes. Five minutes in I had already learned more interesting things than I did from the entire Hard Fork interview. Here is the transcript, which is also helpful. (1:00) Dwarkesh first asks Demis about the nature of intelligence, whether it is one broad thing or the sum of many small things. Demis says there must be some common themes and underlying mechanisms, although there are also specialized parts. I strongly agree with Demis. I do not think you can understand intelligence, of any form, without some form the concept of G. (1:45) Dwarkesh follows up by asking then why doesn't lots of data in one domain generalize to other domains? Demis says often it does, such as coding improving reasoning (which also happens in humans), and he expects more chain transfer. (4:00) Dwarkesh asks what insights neuroscience brings to AI. Demis points to many early AI concepts. Going forward, questions include how brains form world models or memory. (6:00) Demis thinks scaffolding via tree search or AlphaZero-style approaches for LLMs is super promising. He notes they're working hard on search efficiency in many of their approaches so they can search further. (9:00) Dwarkesh notes that Go and Chess have clear win conditions, real life does not, asks what to do about this. Demis agrees this is a challenge, but that usually 'in scientific problems' there are ways to specify goals. Suspicious dodge? (10:00) Dwarkesh notes humans are super sample efficient, Demis says it ...
Here is my episode with Demis Hassabis, CEO of Google DeepMindWe discuss:* Why scaling is an artform* Adding search, planning, & AlphaZero type training atop LLMs* Making sure rogue nations can't steal weights* The right way to align superhuman AIs and do an intelligence explosionWatch on YouTube. Listen on Apple Podcasts, Spotify, or any other podcast platform. Read the full transcript here. Timestamps(0:00:00) - Nature of intelligence(0:05:56) - RL atop LLMs(0:16:31) - Scaling and alignment(0:24:13) - Timelines and intelligence explosion(0:28:42) - Gemini training(0:35:30) - Governance of superhuman AIs(0:40:42) - Safety, open source, and security of weights(0:47:00) - Multimodal and further progress(0:54:18) - Inside Google DeepMind Get full access to Dwarkesh Podcast at www.dwarkeshpatel.com/subscribe
21 Bài Học Cho Thế Kỷ 21 # Chương 2 — Lao động Harari cho biết rằng càng hiểu biết tốt hơn về các cơ chế sinh học nằm dưới các cảm xúc, mong muốn và sự lựa chọn của con người, các máy tính càng có thể phân tích hành vi con người, dự đoán quyết định của con người và thay thế các nghề nghiệp con người như ngân hàng và luật sư. Ít nhất trong một số lĩnh vực công việc, việc thay thế tất cả con người bằng máy tính có thể hợp lý, ngay cả khi cá nhân một số con người vẫn làm tốt hơn máy móc. Về Tự động hóa Cơ quan An toàn giao thông quốc gia Hoa Kỳ ước tính rằng vào năm 2012, 31% trong số các vụ tai nạn gây tử vong liên quan đến lạm dụng rượu, 30% vượt tốc độ và 20% tài xế bị xao lạc. Các phương tiện tự lái không có bất kỳ điểm yếu nào như vậy và cuối cùng, chúng ta có thể thấy sự thay đổi của 3,5 triệu tài xế xe tải chuyên nghiệp ở Hoa Kỳ một mình, ngoài các nghề nghiệp lái xe khác. Thay vì hoàn toàn thay thế con người, trí tuệ nhân tạo thực sự có thể giúp tạo ra các công việc mới cho con người. Thay vì con người cạnh tranh với trí tuệ nhân tạo, họ có thể tập trung vào phục vụ và tận dụng trí tuệ nhân tạo. Thị trường lao động năm 2050 có thể sẽ được đặc trưng bởi sự hợp tác giữa con người và trí tuệ nhân tạo thay vì cạnh tranh. Vài năm sau khi Deep Blue của IBM đánh bại kỳ thủ cờ vua Garry Kasparov, sự hợp tác giữa con người và máy tính phát triển mạnh mẽ. Tuy nhiên, trong những năm gần đây, máy tính đã trở nên tốt đến mức chơi cờ vua mà các đồng sự con người của họ đã mất giá trị, điều này có thể được coi là một dấu hiệu tiên lượng cho những gì có thể xảy ra ở một cấp độ lan rộng hơn. Một chương trình máy tính khác, Alpha Zero của DeepMind, đã từ sự không biết gì đến sự thông thạo sáng tạo trong vòng chưa đầy bốn giờ mà không cần sự hỗ trợ của bất kỳ người hướng dẫn nào, để chiếm ưu thế trước các kỳ thủ và chương trình AlphaGo hàng đầu thế giới. Trí tuệ nhân tạo và Sáng tạo Mọi người thường nói rằng trí tuệ nhân tạo không bao giờ cảm thấy như một con người, rằng nó không bao giờ có thể sáng tạo như một con người. Trong điều này, các trọng tài giải đấu cờ thường đề phòng những người chơi lén nhận sự trợ giúp từ máy tính. Một trong những cách để phát hiện gian lận, chúng ta được cho biết, là theo dõi mức độ độc đáo mà các vận động viên thể hiện. Nếu họ chơi những nước đi cực kỳ sáng tạo, các trọng tài nghi ngờ rằng điều này không thể là một nước đi của con người, nó phải là một nước đi của máy tính. Ít nhất trong cờ vua, sự sáng tạo đã trở thành dấu hiệu của máy tính chứ không phải con người! Các phương thức thay thế cho UBI Chính phủ có thể trợ cấp các dịch vụ cơ bản (UBS) chung thay vì thu nhập Thay vì cho tiền cho những người sau đó mua sắm cho bất cứ điều gì họ muốn Chính phủ có thể trợ cấp giáo dục miễn phí, chăm sóc sức khỏe miễn phí, giao thông miễn phí và vân vân. Điều này thực sự đưa kế hoạch cộng sản thành hiện thực, mặc dù không phải là bằng cách cách mạng. Hạnh phúc = Hiện thực — Kỳ vọng Vấn đề với UBI hoặc UBS là con người không chỉ được xây dựng cho sự hài lòng. Hạnh phúc con người phụ thuộc ít hơn vào điều kiện khách quan và hơn vào kỳ vọng của chúng ta. Kỳ vọng của chúng ta thích ứng với các điều kiện thay đổi bao gồm cả điều kiện của người khác (theo kịp với hàng xóm). Khi điều kiện cải thiện, kỳ vọng tăng lên và do đó, ngay cả những cải thiện đáng kể về điều kiện có thể khiến chúng ta cảm thấy không hài lòng như trước. fi.io.vn
Embark on a journey into the realm of AI chess mastery as Google DeepMind's breakthrough agent surpasses AlphaZero in a historic victory. Discover the strategic brilliance and technological advancements reshaping the future of gaming. Get on the AI Box Waitlist: AIBox.ai Join our ChatGPT Community: Facebook Group Follow me on Twitter: Jaeden's Twitter
Witness the pinnacle of AI gaming as Google DeepMind's breakthrough AI agent conquers AlphaZero in a strategic chess showdown. Delve into the intricacies of this victory and the implications for the future of artificial intelligence. Get on the AI Box Waitlist: AIBox.ai Join our ChatGPT Community: Facebook Group Follow me on Twitter: Jaeden's Twitter
ChatGPT: OpenAI, Sam Altman, AI, Joe Rogan, Artificial Intelligence, Practical AI
Enter the realm of mind games as Google DeepMind's AI agent achieves a monumental triumph by defeating AlphaZero in chess. Explore the strategic brilliance and technological advancements behind this historic breakthrough. Get on the AI Box Waitlist: AIBox.ai Join our ChatGPT Community: Facebook Group Follow me on Twitter: Jaeden's Twitter
In this episode, we explore the gambit played by DeepMind as Google's AI agent outplays the formidable AlphaZero in the timeless game of chess. Join me for a solo exploration, where we dissect the strategic maneuvers, discuss the implications for AI capabilities, and envision the future of machine intelligence in strategic gaming. Invest in AI Box: https://Republic.com/ai-box Get on the AI Box Waitlist: https://AIBox.ai/ AI Facebook Community Learn About ChatGPT Learn About AI at Tesla
In this episode, we explore Google Deepmind's groundbreaking development as their AI agent surpasses AlphaZero in a monumental chess victory, marking a significant leap in AI strategic decision-making. Invest in AI Box: https://Republic.com/ai-box Get on the AI Box Waitlist: https://AIBox.ai/ AI Facebook Community
The world is changing fast. Technology can be used to empower us -- and also to hack our brains & our lives. What laws do we need to protect our freedoms? Rahul Matthan joins Amit Varma in episode 360 of The Seen and the Unseen to share his work on privacy -- and on a new, subtle approach towards data governance. (FOR FULL LINKED SHOW NOTES, GO TO SEENUNSEEN.IN.) Also check out: 1. Rahul Matthan on Twitter, Instagram, LinkedIn, Trilegal, Substack and his own website. 2. Privacy 3.0: Unlocking Our Data-Driven Future -- Rahul Matthan. 3. The Third Way: India's Revolutionary Approach to Data Governance -- Rahul Matthan. 4. The Life and Times of KP Krishnan -- Episode 355 of The Seen and the Unseen. 5. Sudhir Sarnobat Works to Understand the World -- Episode 350 of The Seen and the Unseen. 6. Roam Research. 7. Zettelkasten on Wikipedia. 8. Tana, Obsidian and Notion. 9. Getting Things Done -- David Allen. 10. The Greatest Productivity Mantra: Kaator Re Bhaaji! -- Episode 11 of Everything is Everything. 11. Hallelujah (Spotify) (YouTube) -- Leonard Cohen. 12. Hallelujah (Spotify) (YouTube) -- Jeff Buckley. 13. The Holy or the Broken: Leonard Cohen, Jeff Buckley, and the Unlikely Ascent of "Hallelujah" -- Alan Light. 14. Hallelujah on Revisionist History by Malcolm Gladwell. 15. Bird by Bird: Some Instructions on Writing and Life -- Anne Lamott. 16. The New Basement Tapes. (Also Wikipedia.) 17. Kansas City -- Marcus Mumford. 18. The Premium Mediocre Life of Maya Millennial -- Venkatesh Rao. 19. Vitalik Buterin Fights the Dragon-Tyrant — Episode 342 of The Seen and the Unseen. 20. Paul Graham on Twitter and his own website. (His essays are extraordinary.) 21. Ribbonfarm by Venkatesh Rao. 22. The Network State -- Balaji Srinivasan. 23. Marc Andreessen on Twitter. 24. The Techno-Optimist Manifesto -- Marc Andreessen. 25. Siddhartha Mukherjee and Carlo Rovelli on Amazon. 26. For the Lord (Spotify) (YouTube) -- Rahul Matthan. 27. Predicting the Future -- Rahul Matthan (on Asimov's concept of Psychohistory etc). 28. Gurwinder Bhogal Examines Human Nature — Episode 331 of The Seen and the Unseen. 29. The Looking-Glass Self. 30. Panopticon. 31. Danish Husain and the Multiverse of Culture -- Episode 359 of The Seen and the Unseen. 32. A Scientist in the Kitchen — Episode 204 of The Seen and the Unseen (w Krish Ashok). 33. We Are All Amits From Africa — Episode 343 of The Seen and the Unseen (w Krish Ashok and Naren Shenoy). 34. Nothing is Indian! Everything is Indian! — Episode 12 of Everything is Everything. 35. The Right to Privacy -- Samuel D Warren and Louis D Brandeis. 36. John Locke on Britannica, Stanford Encyclopedia of Philosophy, Wikipedia and Econlib. 37. Build for Tomorrow -- Jason Feifer. 38. Ex Machina -- Alex Garland. 39. Arrival -- Denis Villeneuve. 40. The Great Manure Crisis of 1894 -- Rahul Matthan. 41. Climate Change and Our Power Sector — Episode 278 of The Seen and the Unseen (w Akshay Jaitley and Ajay Shah). 42. The Book of Why: The New Science of Cause and Effect -- Judea Pearl. 43. The New World Upon Us — Amit Varma on Alpha Zero. 44. Brave New World -- Vasant Dhar's podcast, produced by Amit Varma. 45. Human and Artificial Intelligence in Healthcare -- Episode 4 of Brave New World (w Eric Topol). 46. The Colonial Constitution -- Arghya Sengupta. 47. Beyond Consent: A New Paradigm for Data Protection -- Rahul Matthan. 48. The Puttaswamy case. 49. Judicial Reforms in India -- Episode 62 of The Seen and the Unseen (w Alok Prasanna Kumar.) 50. Accidental Feminism: Gender Parity and Selective Mobility among India's Professional Elite -- Swethaa S Ballakrishnen. 51. Magic Fruit: A Poetic Trip -- Vaishnav Vyas. 52. Hermanos Gutiérrez and Arc De Soleil on Spotify. 53. The Travelling Salesman Problem. 54. The Twenty-Six Words That Created the Internet -- Jeff Kosseff. 55. Code: And Other Laws of Cyberspace -- Lawrence Lessig. 56. Financial Inclusion and Digital Transformation in India -- Suyash Rai. 57. No Time for False Modesty -- Rahul Matthan. 58. In Service of the Republic: The Art and Science of Economic Policy -- Vijay Kelkar and Ajay Shah. 59. Once Upon a Prime -- Sarah Hart. 60. The Greatest Invention -- Silvia Ferrara. 61. Surveillance State -- Josh Chin and Liza Lin. 62. Surveillance Valley -- Yasha Levine. 63. Sex Robots and Vegan Meat -- Jenny Kleeman. 64. How to Take Smart Notes -- Sönke Ahrens. 65. The Creative Act -- Rick Rubin. 66. How to Write One Song -- Jeff Tweedy. 67. Adrian Tchaikovsky and NK Jemisin on Amazon. 68. Snarky Puppy. on Spotify and YouTube. 69. Empire Central -- Snarky Puppy. 70. Polyphia on Spotify and YouTube. 71. The Lazarus Project on Jio Cinema. This episode is sponsored by the Pune Public Policy Festival 2024, which takes place on January 19 & 20, 2024. The theme this year is Trade-offs! Amit Varma and Ajay Shah have launched a new video podcast. Check out Everything is Everything on YouTube. Check out Amit's online course, The Art of Clear Writing. And subscribe to The India Uncut Newsletter. It's free! Episode art: ‘Protocol' by Simahina.
Yapay zeka denince bugün akla gelen ilk isim olan Sam Altman, geçenlerde kendi kurduğu şirketten sorgusuz sualsiz kovuldu ve birkaç çalkantılı gün sonunda, kendisini kovanları kovdurarak geri döndü. Eğlenceli bir pembe dizi gibi ama yüzeyin biraz altını kazıyınca ortaya çıkan tablo beni endişelendirdi. Esas kavga kişiler arasında değil, etik kaygılar ile rekabetin arttırdığı ticari kaygılar arasında. Ödül de bir taht değil, yapay zekanın geleceği. Ve bence bu roundu "iyiler" kazanmadı..Konular:(00:00) OpenAI'daki pembe dizi.(02:26) Tarihte bugün: Deep Blue ve Alphazero.(05:00) OpenAI'ın closedAI oluşu.(07:07) Kar amacı gütmeyen kuruluş(11:58) Kovulma.(13:53) Custom GPT.(15:58) Project Q*(17:54) Para para para.(21:12) Devletlerin rolü.(23:30) Patreon Teşekkürleri.Kaynaklar:Haber: Google shared AI knowledge with the world — until ChatGPT caught upThe inside story of how ChatGPT was built from the people who made it.------- Podbee Sunar -------Bu podcast, Cambly hakkında reklam içerir.Cambly'de yılın en büyük indirimi %60'dan fular60 koduyla faydalanmak için tıklayınız.See Privacy Policy at https://art19.com/privacy and California Privacy Notice at https://art19.com/privacy#do-not-sell-my-info.
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: Possible OpenAI's Q* breakthrough and DeepMind's AlphaGo-type systems plus LLMs, published by Burny on November 23, 2023 on LessWrong. tl;dr: OpenAI leaked AI breakthrough called Q*, acing grade-school math. It is hypothesized combination of Q-learning and A*. It was then refuted. DeepMind is working on something similar with Gemini, AlphaGo-style Monte Carlo Tree Search. Scaling these might be crux of planning for increasingly abstract goals and agentic behavior. Academic community has been circling around these ideas for a while. https://www.reuters.com/technology/sam-altmans-ouster-openai-was-precipitated-by-letter-board-about-ai-breakthrough-2023-11-22/ https://twitter.com/MichaelTrazzi/status/1727473723597353386 "Ahead of OpenAI CEO Sam Altman's four days in exile, several staff researchers sent the board of directors a letter warning of a powerful artificial intelligence discovery that they said could threaten humanity Mira Murati told employees on Wednesday that a letter about the AI breakthrough called Q* (pronounced Q-Star), precipitated the board's actions. Given vast computing resources, the new model was able to solve certain mathematical problems. Though only performing math on the level of grade-school students, acing such tests made researchers very optimistic about Q*'s future success." https://twitter.com/SilasAlberti/status/1727486985336660347 "What could OpenAI's breakthrough Q* be about? It sounds like it's related to Q-learning. (For example, Q* denotes the optimal solution of the Bellman equation.) Alternatively, referring to a combination of the A* algorithm and Q learning. One natural guess is that it is AlphaGo-style Monte Carlo Tree Search of the token trajectory. It seems like a natural next step: Previously, papers like AlphaCode showed that even very naive brute force sampling in an LLM can get you huge improvements in competitive programming. The next logical step is to search the token tree in a more principled way. This particularly makes sense in settings like coding and math where there is an easy way to determine correctness. https://twitter.com/mark_riedl/status/1727476666329411975 "Anyone want to speculate on OpenAI's secret Q* project? Something similar to tree-of-thought with intermediate evaluation (like A*)? Monte-Carlo Tree Search like forward roll-outs with LLM decoder and q-learning (like AlphaGo)? Maybe they meant Q-Bert, which combines LLMs and deep Q-learning Before we get too excited, the academic community has been circling around these ideas for a while. There are a ton of papers in the last 6 months that could be said to combine some sort of tree-of-thought and graph search. Also some work on state-space RL and LLMs." https://www.theverge.com/2023/11/22/23973354/a-recent-openai-breakthrough-on-the-path-to-agi-has-caused-a-stir OpenAI spokesperson Lindsey Held Bolton refuted it: "refuted that notion in a statement shared with The Verge: "Mira told employees what the media reports were about but she did not comment on the accuracy of the information."" https://www.wired.com/story/google-deepmind-demis-hassabis-chatgpt/ Google DeepMind's Gemini, that is currently the biggest rival with GPT4, which was delayed to the start of 2024, is also trying similar things: AlphaZero-based MCTS through chains of thought, according to Hassabis. Demis Hassabis: "At a high level you can think of Gemini as combining some of the strengths of AlphaGo-type systems with the amazing language capabilities of the large models. We also have some new innovations that are going to be pretty interesting." https://twitter.com/abacaj/status/1727494917356703829 Aligns with DeepMind Chief AGI scientist Shane Legg saying: "To do really creative problem solving you need to start searching." https://twitter.com/iamgingertrash/status/1727482695356494132 "...
AI Chat: ChatGPT & AI News, Artificial Intelligence, OpenAI, Machine Learning
In this episode, we explore the integration of prompt engineering and AlphaZero Microsoft XOT in improving the generalization abilities of large language models (LLMs), analyzing how these advancements contribute to more versatile and effective AI applications. We'll discuss the technical aspects of this combination and its potential impact on the future capabilities and applications of LLMs. Invest in AI Box: https://republic.com/ai-box Get on the AI Box Waitlist: https://AIBox.ai/ Facebook Community: https://www.facebook.com/groups/739308654562189 Follow me on X: https://twitter.com/jaeden_ai
AI Hustle: News on Open AI, ChatGPT, Midjourney, NVIDIA, Anthropic, Open Source LLMs
In this episode, delve into the groundbreaking achievement of Google DeepMind's AI agent as it surpasses AlphaZero in the world of chess. Join us as we explore the remarkable developments and implications of this game-changing advancement in artificial intelligence. Discover how this AI breakthrough is reshaping the future of strategic thinking and machine learning. Get on the AI Box Waitlist: https://AIBox.ai/Join our ChatGPT Community: https://www.facebook.com/groups/739308654562189/Follow me on Twitter: https://twitter.com/jaeden_ai
Brought to you by Eppo—Run reliable, impactful experiments | Vanta—Automate compliance. Simplify security | Ezra—The leading full-body cancer screening company—Lane Shackleton is CPO of Coda, where he's been leading the product and design team for over eight years. Lane started his career as an Alaskan climbing guide and then as a manual reviewer of AdWords ads before becoming a product specialist at Google and later a Group PM at YouTube. He also writes a weekly newsletter with insights and rituals for PMs, product teams, and startups. In today's conversation, we discuss:• Principles that set great PMs apart• Rituals of great product teams• The fine line between OKRs and strategy, and why it matters• “Two-way write-up”• The story of how skippable YouTube ads were born and lessons learned• How to gauge personal career growth• “Tim Ferriss Day” and its impact on Coda's history• How Lane bootstrapped his way to CPO from the bottom of the tech ladder—Find the transcript and references at: https://www.lennyspodcast.com/what-sets-great-teams-apart-lane-shackleton-cpo-of-coda/ —Where to find Lane Shackleton:• X: https://twitter.com/lshackleton• LinkedIn: https://www.linkedin.com/in/laneshackleton• Substack: https://lane.substack.com/—Where to find Lenny:• Newsletter: https://www.lennysnewsletter.com• X: https://twitter.com/lennysan• LinkedIn: https://www.linkedin.com/in/lennyrachitsky/—In this episode, we cover:(00:00) Lane's background(04:03) Working as a guide in Alaska(07:32) Parallels between guiding and building software(09:12) Why Lane started studying and writing about product teams(12:49) How Lane came up with the career ladder and guiding principles(14:10) The five levels Coda's career ladder(16:30) Principles of great product managers(21:06) The beginner's-mind ritual at Coda(24:05) Two rituals: “cathedrals not bricks” and “proactive not reactive”(27:46) How to develop your own guiding principles(31:17) Learning from your “oh s**t” moments(36:03) Rituals from great product teams: HubSpot's FlashTags(42:15) Rituals from great product teams: Coda's Catalyst(47:01) Implementing rituals from other companies(49:48) How to navigate changing vs. sticking with current rituals(53:02) “Tag up” and why one-on-one meetings are harmful (55:27) Lane's handbook on strategy and rituals(57:10) How skippable ads came about on YouTube (1:01:46) Lane's path to CPO(1:07:02) Advice for aspiring PMs(1:10:53) Tim Ferriss Day at Coda(1:13:24) Using two-way write-ups (1:19:30) The fine line between OKRs and strategy, and why it matters(1:21:41) Lightning round—Referenced:• Endurance: https://www.amazon.com/Endurance-Shackletons-Incredible-Alfred-Lansing/dp/0465062881• Bret Victor's talk “Inventing on Principle”: https://www.youtube.com/watch?v=EGqwXt90ZqA• Jeremy Britton on LinkedIn: https://www.linkedin.com/in/jeremybritton/• Comedian on Netflix: https://www.netflix.com/title/60024976• The Score Takes Care of Itself: My Philosophy of Leadership: https://www.amazon.com/Score-Takes-Care-Itself-Philosophy/dp/1591843472• The Creative Act: A Way of Being: https://www.amazon.com/Creative-Act-Way-Being/dp/0593652886• AlphaZero: https://en.wikipedia.org/wiki/AlphaZero• Antoine de Saint-Exupéry: https://en.wikipedia.org/wiki/Antoine_de_Saint-Exup%C3%A9ry• Storyworthy: Engage, Teach, Persuade, and Change Your Life through the Power of Storytelling: https://www.amazon.com/Storyworthy-Engage-Persuade-through-Storytelling/dp/1608685489• The Moth: https://themoth.org/events• Seth Godin's website: https://www.sethgodin.com/• The Obstacle Is the Way: The Timeless Art of Turning Trials into Triumph: https://www.amazon.com/Obstacle-Way-Timeless-Turning-Triumph/dp/1591846358• Tony Fadell's TED talk: https://www.youtube.com/watch?v=9uOMectkCCs• FlashTags: A Simple Hack for Conveying Context Without Confusion: https://www.onstartups.com/flashtags-a-simple-hack-for-conveying-context-without-confusion• How Coda builds product: https://www.lennysnewsletter.com/p/how-coda-builds-product• 100-dollar voting ritual: https://coda.io/@lshackleton/100-dollar-voting-exercise• Pixar's Brain Trust: https://pixar.fandom.com/wiki/Brain_Trust• Lane's product handbook: coda.io/producthandbook• The rituals of great teams | Shishir Mehrotra of Coda, YouTube, Microsoft: https://www.lennyspodcast.com/the-rituals-of-great-teams-shishir-mehrotra-coda-youtube-microsoft/• Principle #4: Learn by making, not talking: https://lane.substack.com/p/principle-4-learn-by-making-not-talking• Phil Farhi on LinkedIn: https://www.linkedin.com/in/philfarhi/• How to ask the right questions, project confidence, and win over skeptics | Paige Costello (Asana, Intercom, Intuit): https://www.lennyspodcast.com/how-to-ask-the-right-questions-project-confidence-and-win-over-skeptics-paige-costello-asana-intercom-intuit/• Chip Conley's website: https://chipconley.com/• Jeff Bezos Banned PowerPoint in Meetings. His Replacement Is Brilliant: https://www.inc.com/carmine-gallo/jeff-bezos-bans-powerpoint-in-meetings-his-replacement-is-brilliant.html• Working Backwards: Insights, Stories, and Secrets from Inside Amazon: https://www.amazon.com/Working-Backwards-Insights-Stories-Secrets/dp/1250267595• Dory and Pulse: https://coda.io/@codatemplates/dory-and-pulse• Turning the Flywheel: A Monograph to Accompany Good to Great: https://www.amazon.com/Turning-Flywheel-Monograph-Accompany-Great/dp/0062933795/• Waking Up: A Guide to Spirituality Without Religion: https://www.amazon.com/Waking-Up-Spirituality-Without-Religion/dp/1451636024• The Inner Game of Tennis: The Classic Guide to the Mental Side of Peak Performance: https://www.amazon.com/Inner-Game-Tennis-Classic-Performance/dp/0679778314• Good Strategy/Bad Strategy: The Difference and Why It Matters: https://www.amazon.com/Good-Strategy-Bad-Difference-Matters/dp/0307886239• The Last Dance on Netflix: https://www.netflix.com/title/80203144• Full Swing on Netflix: https://www.netflix.com/title/81483353• Stephen Curry: Underrated on AppleTV+: https://tv.apple.com/us/movie/stephen-curry-underrated/umc.cmc.23v0wxaiwz60bjy1w4vg7npun• Arrested Development on Netflix: https://www.netflix.com/title/70140358• Shishir's interview question clip on TikTok: https://www.tiktok.com/@lennyrachitsky/video/7160779872296652078• The Ultimate Reference Check Template: https://coda.io/@startup-hiring/reference-checks-template• SwingVision: https://swing.tennis/• Waking Up app: https://www.wakingup.com/—Production and marketing by https://penname.co/. For inquiries about sponsoring the podcast, email podcast@lennyrachitsky.com.—Lenny may be an investor in the companies discussed. Get full access to Lenny's Newsletter at www.lennysnewsletter.com/subscribe
From Ganja Park in Kolkata to lock-ups in 13 countries, he has travelled the world and lived through mad times. Devangshu Datta joins Amit Varma in episode 348 of The Seen and the Unseen to discuss Bengalis who make bombs, Gujaratis who make fetish costumes, his river pirate ancestors and how Only Fans has disrupted Pornhub. (FOR FULL LINKED SHOW NOTES, GO TO SEENUNSEEN.IN.) Also check out: 1. Devangshu Datta on Twitter and Business Standard. 2. Previous (miniature) episodes of The Seen and the Unseen with Devangshu Datta: 1, 2, 3. 3. The Life and Times of Nilanjana Roy — Episode 284 of The Seen and the Unseen. 4. Gita Press and the Making of Hindu India — Akshaya Mukul. 5. The Gita Press and Hindu Nationalism — Episode 139 of The Seen and the Unseen (w Akshaya Mukul). 6. Private Truths, Public Lies — Timur Kuran. 7. Godwin's Law. 8. The End of History? — Francis Fukuyama's essay. 9. The End of History and the Last Man — Francis Fukuyama's book. 10. Fixing Indian Education — Episode 185 of The Seen and the Unseen (w Karthik Muralidharan). 11. Our Unlucky Children (2008) — Amit Varma. 12. Aakash Singh Rathore, the Ironman Philosopher — Episode 340 of The Seen and the Unseen. 13. The Bridge: The Life and Rise of Barack Obama -- David Remnick. 14. VP Menon: The Unsung Architect of Modern India — Narayani Basu. 15. India's Greatest Civil Servant — Episode 167 of The Seen and the Unseen (w Narayani Basu, on VP Menon). 16. 'How big is your Madhya Pradesh?' -- Mamata Banerjee asks a party worker to lose weight. 17. Patriots, Poets and Prisoners: Selections from Ramananda Chatterjee's The Modern Review, 1907-1947 -- Edited by Anikendra Sen, Devangshu Datta and Nilanjana S Roy. 18. The State of Indian Sport — Episode 238 of The Seen and the Unseen (w Joy Bhattacharjya & Nandan Kamath). 19. Early Indians -- Tony Joseph. 20. Early Indians — Episode 112 of The Seen and the Unseen (w Tony Joseph). 21. All Quiet on the Western Front -- Erich Maria Remarque. 22. The Dosadi Experiment (featuring Jorj X. McKie) -- Frank Herbert. 23. A Deep Dive Into Ukraine vs Russia -- Episode 335 of The Seen and the Unseen (w Ajay Shah). 24. Lost Victories -- Erich von Manstein. 25. Basic Chess Endings -- Reuben Fine. 26. The Tamilian Gentleman Who Took on the World — Amit Varma. 27. The New World Upon Us -- Amit Varma on Alpha Zero. 28. Alpha Zero -- Episode 51 of The Seen and the Unseen (w Devangshu Datta). 29. Google's AlphaZero Destroys Stockfish In 100-Game Match — Mike Klein (with Peter Heine Nielson's quote on a superior species playing chess). 30. Skynet (Terminator). 31. Neuromancer -- William Gibson. 32. Snow Crash -- Neal Stephenson. 33. Why Children Labour (2007) — Amit Varma. 34. The Poetic Feminism of Paromita Vohra — Episode 339 of The Seen and the Unseen. 35. Satisfaction: The Art of the Female Orgasm -- Kim Cattrall and Mark Levinson. 36. Deep Throat and The Devil in Miss Jones. 37. The Matunga Racket (2007) -- Amit Varma. 38. Colleen Hoover on Amazon, Instagram, Wikipedia and her own website. 39. The Business of Books — Episode 150 of The Seen and the Unseen (w VK Karthika). 40. New in Chess. 41. Amartya Ghosh on Spotify. 42. The Universe of Chuck Gopal — Episode 258 of The Seen and the Unseen. 43. 'Wet Streets Cause Rain' -- Michael Crichton explains Gell-Mann Amnesia. 44. How to generate black money in India (2013) -- TEDx Talk by Devangshu Datta. 45. Poker and Stock Markets — Episode 47 of The Seen and the Unseen (w Mohit Satyanand). 46. Once Upon a Prime -- Sarah Hart. 47. Herman Melville and Edgar Allan Poe on Amazon. 48. Professor Moriarty. 49. The Curious Incident of the Dog in the Night-time -- Mark Haddon. 50. A Gentleman in Moscow -- Amor Towles. 51. NK Jemisin and Ursula K Le Guin on Amazon. 52. The Battle for Spain: The Spanish Civil War 1936-1939 -- Antony Beevor. 53. The Spanish Civil War (playlist with all six parts of the docu-series). 54. The Sandman on Netflix. 55. The Sandman -- Neil Gaiman. 56. The Life and Games of Mikhail Tal -- Mikhail Tal. 57. Dune and Blade Runner 2049 -- Denis Villeneuve. 58. India's War: The Making of Modern South Asia -- Srinath Raghavan. 59. Episodes of The Seen and the Unseen with Srinath Raghavan: 1, 2, 3, 4, 5. Amit Varma and Ajay Shah have launched a new video podcast. Check out Everything is Everything on YouTube. Check out Amit's online course, The Art of Clear Writing. And subscribe to The India Uncut Newsletter. It's free! Episode art: ‘Chess Board' by Simahina.
Introduction: David Hundley is a Machine Learning Engineer who has been deeply involved with experimenting with Large Language Models (LLMs). Follow along on his twitter Key Insights & Discussions: Discoveries with LLMs: David recently explored a unique function of LLMs that acted as a 'dummy agent'. This function would prompt the LLM to search the internet for a current movie, bypassing its training limitations. David attempted to utilize this function to generate trivia questions, envisaging a trivia game powered by the LLM. However, he faced challenges in getting the agent to converge on the desired output. Parsing the LLM's responses into a structured output proved especially difficult. Autonomous Agents & AGI: David believes that AGI (Artificial General Intelligence) essentially comprises autonomous agents. The prospect of these agents executing commands directly on one's computer can be unnerving. However, when LLMs run code, they operate within a contained environment, ensuring safety. Perceptions of AI: There's a constant cycle of learning and revisiting motivations and goals in the realm of AI. David warns against anthropomorphizing LLMs, as they don't possess human motivations. He stresses that the math underpinning AI doesn't align with human emotions or motivations. Emergent Behavior & Consciousness: David postulates that everything in the universe sums up to a collective result. There's debate over whether living organisms possess true consciousness, and what it means for AGI. The concept of AGI emulating human intelligence is complex. The human psyche is shaped by countless historical experiences and stimuli. So, if AGI were to truly replicate human thought, it would require vast amounts of multimodal input. A challenging question raised is how one tests for consciousness in AGI. David believes that as we continue to push technological boundaries, our definition of consciousness will keep shifting. Rights & Ethics of AI: With advancing AI capabilities, the debate around the rights of AI entities intensifies. David also touches upon the topic of transhumanism, discussing the trajectory of the universe and the evolution of humans. He contemplates the potential paths of evolution, like physically merging with technology or digitizing our consciousness. AI's Impact on Coding & Jobs: David reflects on the early days of AI in coding. He acknowledges the transformative potential of AI in the field but remains unworried about AI taking over his job. Instead, he focuses on how AI can aid in problem-solving. He describes LLMs as "naive geniuses" - incredibly capable, yet still requiring guidance. Open Source & OpenAI: David discusses the concept of open source, emphasizing the transparency it offers in understanding the data and architecture behind AI models. He acknowledges OpenAI's significant role in the AI landscape and predicts that plugins like ChatGPT will bridge the gap to further automation. Math's Role in AI: The conversation delves into the importance of math in AI, with David detailing concepts like gradient descent and its role in building neural networks. David also touches on the evolution of AI models, comparing the capabilities of models with 70 billion parameters to those with 7 billion. He predicts that models with even more parameters, perhaps in the trillions, will emerge, further emulating human intelligence. Future Prospects & Speculations: David muses on the future trajectory of LLMs, drawing parallels with the evolution of AlphaGo to AlphaZero. The episode concludes with philosophical musings on the nature of consciousness and its implications on world religions.
This week on Chess Journeys, I spoke with Tess. We had a fascinating conversation about her chess journeys. Tess started out by playing 10,000 rapid games in a short amount of time. Essentially, she was the human form of Alpha Zero. Tess spoke of the different periods of her improvement and what helped her most. We also discussed topics such as tournaments, women in chess, and adult improvers. Use this link to purchase Next Level Training: https://nextlevelchesscourses.teachable.com/p/next-level-training?affcode=1152624_t9sjlp49 Be sure to check out the Chess Journeys Merch Store! You can support the show and look amazing in the process. https://chess-journeys.creator-spring.com/ I've been streaming somewhat regularly on https://www.twitch.tv/drskull_tinygrimes If you would like to be a guest on Chess Journeys, contact me on Twitter or fill out the following Google Form: https://forms.gle/gSnvmUnvpykkgT1y5 As always you can support the show at https://www.patreon.com/ChessJourneys. Also, be sure to check out my Chessable page at www.chessable.com/chessjourneys If you are considering using Aimchess, please use the code drscull30.
A five-time World Chess Champion, Vishy became India's first grandmaster at age 18, spurring a chess revolution in the country. Now 53, he is still a world top ten player and has been India's number one ranked player for 37 years. As newer talents emerge and old ones retire, Anand's continued excellence showcases an endurance seldom seen. Tyler and Vishy sat down in Chennai to discuss his breakthrough 1991 tournament win in Reggio Emilia, his technique for defeating Kasparov in rapid play, how he approached playing the volatile but brilliant Vassily Ivanchuk at his peak, a detailed breakdown of his brilliant 2013 game against Levon Aronian, dealing with distraction during a match, how he got out of a multi-year slump, Monty Python vs. Fawlty Towers, the most underrated Queen song, how far to take chess opening preparation, which style of chess will dominate in the next ten years, how AlphaZero changes what we know about the game, the key to staying a top ten player at age 53, why he thinks he's a worse loser than Kasparov, qualities he looks for in talented young Indian chess players, picks for the best places to eat in Chennai, and more. Read a full transcript enhanced with helpful links, or watch the full video. Recorded August 7th, 2023. Other ways to connect Follow us on X and Instagram Follow Tyler on X Follow Vishy on X Join our Discord Email us: cowenconvos@mercatus.gmu.edu Learn more about Conversations with Tyler and other Mercatus Center podcasts here. Special thanks to Nabeel Qureshi for his help with the video and transcript.
AI Chat: ChatGPT & AI News, Artificial Intelligence, OpenAI, Machine Learning
In this riveting episode, we delve into Google DeepMind's latest AI breakthrough, an agent so advanced it managed to defeat the previously unbeatable AlphaZero in chess. We'll explore the technical leaps that made this possible and discuss the implications of this new technology for the future of artificial intelligence, competitive gaming, and problem-solving algorithms. Get on the AI Box Waitlist: https://AIBox.ai/ Facebook Community: https://www.facebook.com/groups/739308654562189/ Discord Community: https://aibox.ai/discord Follow me on X: https://twitter.com/jaeden_ai
Sam Harris speaks with Mustafa Suleyman about his new book, “The Coming Wave: Technology, Power, and the 21st Century's Greatest Dilemma.” They discuss the progress in artificial intelligence made at his company DeepMind, the acquisition of DeepMind by Google, Atari DQN, AlphaGo, AlphaZero, AlphaFold, the invention of new knowledge, the risks of our making progress in AI, “superintelligence” as a distraction from more pressing problems, the inevitable spread of general-purpose technology, the nature of intelligence, productivity growth and labor disruptions, “the containment problem,” the importance of scale, Moore's law, Inflection AI, open-source LLMs, changing the norms of work and leisure, the redistribution of value, introducing friction into the deployment of AI, regulatory capture, a misinformation apocalypse, digital watermarks, asymmetric threats, conflict and cooperation with China, supply-chain monopolies, and other topics. If the Making Sense podcast logo in your player is BLACK, you can SUBSCRIBE to gain access to all full-length episodes at samharris.org/subscribe. Learning how to train your mind is the single greatest investment you can make in life. That's why Sam Harris created the Waking Up app. From rational mindfulness practice to lessons on some of life's most important topics, join Sam as he demystifies the practice of meditation and explores the theory behind it.
Share this episode: https://www.samharris.org/podcasts/making-sense-episodes/332-can-we-contain-artificial-intelligence Sam Harris speaks with Mustafa Suleyman about his new book, “The Coming Wave: Technology, Power, and the 21st Century’s Greatest Dilemma.” They discuss the progress in artificial intelligence made at his company DeepMind, the acquisition of DeepMind by Google, Atari DQN, AlphaGo, AlphaZero, AlphaFold, the invention of new knowledge, the risks of our making progress in AI, “superintelligence” as a distraction from more pressing problems, the inevitable spread of general-purpose technology, the nature of intelligence, productivity growth and labor disruptions, “the containment problem,” the importance of scale, Moore’s law, Inflection AI, open-source LLMs, changing the norms of work and leisure, the redistribution of value, introducing friction into the deployment of AI, regulatory capture, a misinformation apocalypse, digital watermarks, asymmetric threats, conflict and cooperation with China, supply-chain monopolies, and other topics. Mustafa Suleyman is the co-founder and CEO of Inflection AI. Previously he co-founded DeepMind, one of the world’s leading artificial intelligence companies. After a decade at DeepMind, Suleyman became vice president of AI product management and AI policy at Google. When he was an undergraduate at Oxford, Suleyman dropped out to help start a non-profit telephone counseling service. He lives in Palo Alto, California. Website: https://www.the-coming-wave.com/ Twitter: @mustafasuleyman Learning how to train your mind is the single greatest investment you can make in life. That’s why Sam Harris created the Waking Up app. From rational mindfulness practice to lessons on some of life’s most important topics, join Sam as he demystifies the practice of meditation and explores the theory behind it.
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: Towards Developmental Interpretability, published by Jesse Hoogland on July 12, 2023 on The AI Alignment Forum. Developmental interpretability is a research agenda that has grown out of a meeting of the Singular Learning Theory (SLT) and AI alignment communities. To mark the completion of the first SLT & AI alignment summit we have prepared this document as an outline of the key ideas. As the name suggests, developmental interpretability (or "devinterp") is inspired by recent progress in the field of mechanistic interpretability, specifically work on phase transitions in neural networks and their relation to internal structure. Our two main motivating examples are the work by Olsson et al. on In-context Learning and Induction Heads and the work by Elhage et al. on Toy Models of Superposition. Mechanistic interpretability emphasizes features and circuits as the fundamental units of analysis and usually aims at understanding a fully trained neural network. In contrast, developmental interpretability: is organized around phases and phase transitions as defined mathematically in SLT, and aims at an incremental understanding of the development of internal structure in neural networks, one phase transition at a time. The hope is that an understanding of phase transitions, integrated over the course of training, will provide a new way of looking at the computational and logical structure of the final trained network. We term this developmental interpretability because of the parallel with developmental biology, which aims to understand the final state of a different class of complex self-assembling systems (living organisms) by analyzing the key steps in development from an embryonic state. In the rest of this post, we explain why we focus on phase transitions, the relevance of SLT, and how we see developmental interpretability contributing to AI alignment. Thank you to @DanielFilan, @bilalchughtai, @Liam Carroll for reviewing early drafts of this document. Why phase transitions? First of all, they exist: there is a growing understanding that there are many kinds of phase transitions in deep learning. For developmental interpretability, the most important kind of phase transitions are those that occur during training. Some of the examples we are most excited about: Olsson, et al., "In-context Learning and Induction Heads", Transformer Circuits Thread, 2022. Elhage, et al., "Toy Models of Superposition", Transformer Circuits Thread, 2022. McGrath, et al., "Acquisition of chess knowledge in AlphaZero", PNAS, 2022. Michaud, et al., "The Quantization Model of Neural Scaling", 2023. Simon, et al., "On the Stepwise Nature of Self-Supervised Learning" ICML 2023. The literature on other kinds of phase transitions, such as those appearing as the scale of the model is increased, is even broader. Neel Nanda has conjectured that "phase changes are everywhere." Second, they are easy to find: from the point of view of statistical physics, two of the hallmarks of a (second-order) phase transition are the divergence of macroscopically observable quantities and the emergence of large-scale order. Divergences make phase transitions easy to spot, and the emergence of large-scale order (e.g., circuits) is what makes them interesting. There are several natural observables in SLT (the learning coefficient or real log canonical threshold, and singular fluctuation) which can be used to detect phase transitions, but we don't yet know how to invent finer observables of this kind, nor do we understand the mathematical nature of the emergent order. Third, they are good candidates for universality: every mouse is unique, but its internal organs fit together in the same way and have the same function - that's why biology is even possible as a field of science. Similarly, as an emerging field of science,...
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: Towards Developmental Interpretability, published by Jesse Hoogland on July 12, 2023 on LessWrong. Developmental interpretability is a research agenda that has grown out of a meeting of the Singular Learning Theory (SLT) and AI alignment communities. To mark the completion of the first SLT & AI alignment summit we have prepared this document as an outline of the key ideas. As the name suggests, developmental interpretability (or "devinterp") is inspired by recent progress in the field of mechanistic interpretability, specifically work on phase transitions in neural networks and their relation to internal structure. Our two main motivating examples are the work by Olsson et al. on In-context Learning and Induction Heads and the work by Elhage et al. on Toy Models of Superposition. Mechanistic interpretability emphasizes features and circuits as the fundamental units of analysis and usually aims at understanding a fully trained neural network. In contrast, developmental interpretability: is organized around phases and phase transitions as defined mathematically in SLT, and aims at an incremental understanding of the development of internal structure in neural networks, one phase transition at a time. The hope is that an understanding of phase transitions, integrated over the course of training, will provide a new way of looking at the computational and logical structure of the final trained network. We term this developmental interpretability because of the parallel with developmental biology, which aims to understand the final state of a different class of complex self-assembling systems (living organisms) by analyzing the key steps in development from an embryonic state. In the rest of this post, we explain why we focus on phase transitions, the relevance of SLT, and how we see developmental interpretability contributing to AI alignment. Thank you to @DanielFilan, @bilalchughtai, @Liam Carroll for reviewing early drafts of this document. Why phase transitions? First of all, they exist: there is a growing understanding that there are many kinds of phase transitions in deep learning. For developmental interpretability, the most important kind of phase transitions are those that occur during training. Some of the examples we are most excited about: Olsson, et al., "In-context Learning and Induction Heads", Transformer Circuits Thread, 2022. Elhage, et al., "Toy Models of Superposition", Transformer Circuits Thread, 2022. McGrath, et al., "Acquisition of chess knowledge in AlphaZero", PNAS, 2022. Michaud, et al., "The Quantization Model of Neural Scaling", 2023. Simon, et al., "On the Stepwise Nature of Self-Supervised Learning" ICML 2023. The literature on other kinds of phase transitions, such as those appearing as the scale of the model is increased, is even broader. Neel Nanda has conjectured that "phase changes are everywhere." Second, they are easy to find: from the point of view of statistical physics, two of the hallmarks of a (second-order) phase transition are the divergence of macroscopically observable quantities and the emergence of large-scale order. Divergences make phase transitions easy to spot, and the emergence of large-scale order (e.g., circuits) is what makes them interesting. There are several natural observables in SLT (the learning coefficient or real log canonical threshold, and singular fluctuation) which can be used to detect phase transitions, but we don't yet know how to invent finer observables of this kind, nor do we understand the mathematical nature of the emergent order. Third, they are good candidates for universality: every mouse is unique, but its internal organs fit together in the same way and have the same function - that's why biology is even possible as a field of science. Similarly, as an emerging field of science, interpretabi...
There has never been a better time to be a creator. Amit Varma joins Vasant Dhar in episode 64 of Brave New World to discuss his learnings as a podcaster and blogger -- and to explain why AI is not a threat. Useful resources: 1. Amit Varma on Twitter, India Uncut and Substack. 2. The Seen and the Unseen -- Amit Varma's podcast. 3. Brave New World — Episode 203 of The Seen and the Unseen (w Vasant Dhar). 4. Episodes of The Seen and the Unseen on the creator ecosystem with Roshan Abbas, Varun Duggirala, Neelesh Misra, Snehal Pradhan, Chuck Gopal, Nishant Jain, Deepak Shenoy, Abhijit Bhaduri and Gaurav Chintamani. 5. A Meditation on Form -- Amit Varma. 6. Why Are My Episodes so Long? -- Amit Varma. 7. If You Are a Creator, This Is Your Time -- Amit Varma. 8. The Naked Sun -- Isaac Asimov. 9. Range Rover — The archives of Amit Varma's column on poker for the Economic Times. 10. Fog of War and Atomic Chess. 11. Fixing Indian Education — Episode 185 of The Seen and the Unseen (w Karthik Muralidharan). 12. A Scientist in the Kitchen — Episode 204 of The Seen and the Unseen (w Krish Ashok). 13. Law and Education in Our Modern World — Episode 5 of Brave New World (w John Sexton). 14. My Friend Dropped His Pants -- Amit Varma. 15. Eric Weinstein Won't Toe the Line -- Episode 330 of The Seen and the Unseen. 16. $800,000 to Zero – The FASCINATING History of DaVinci Resolve — Alex Jordan of Learn Color Grading. 17. Miss Excel on Instagram and TikTok. 18. How an Excel Tiktoker Manifested Her Way to Making Six Figures a Day — Nilay Patel. 19. This Australian Man Can Speak Fluent Hindi And Bhojpuri -- ABP News. 20. Advertising is Dead -- Varun Duggirala's podcast. 21. How I Gained 1 MILLION Subscribers — Ali Abdaal. 22. My Top 10 Tips for Aspiring YouTubers — Ali Abdaal. 23. Uplift the Unremarkables — Episode 2 of Brave New World (w Scott Galloway). 24. Amitava Kumar Finds the Breath of Life — Episode 265 of The Seen and the Unseen. 25. The Blue Book: A Writer's Journal — Amitava Kumar. 26. 1000 True Fans — Kevin Kelly. 27. 1000 True Fans? Try 100 — Li Jin. 28. Bill Bishop's newsletter on China. 29. The Life and Times of Jerry Pinto — Episode 314 of The Seen and the Unseen. 30. Murder in Mahim — Jerry Pinto. 31. Mallikarjun Mansur and Bhimsen Joshi on Spotify. 32. The New World Upon Us — Amit Varma on Alpha Zero. 33. Google's AlphaZero Destroys Stockfish In 100-Game Match — Mike Klein. Check out Vasant Dhar's newsletter on Substack. Subscription is free!
He's a freethinker who's never been scared to rattle orthodoxies. Eric Weinstein joins Amit Varma in episode 330 of The Seen and the Unseen to talk about how he became the man he is. (FOR FULL LINKED SHOW NOTES, GO TO SEENUNSEEN.IN.) Also check out: 1. Eric Weinstein on Twitter, Instagram, YouTube and his own website. 2. Eric Weinstein on The Joe Rogan Experience (1, 2) and Lex Fridman (1, 2, 3, 4). 3. Eric Weinstein in Karjat, 1986. 4. The Road Not Taken -- Robert Frost. 5. Stopping by Woods on a Snowy Evening -- Robert Frost. 6. Total Love -- Gur Bentwich. 7. Eric Weinstein in Kiev, 1989. 8. Eric Weinstein meets a harmonica friend. 9. The 27 Club. 10. Umwelt. 11. John Mayer and Blake Mills on Spotify. 12. Sara Rai Inhales Literature — Episode 255 of The Seen and the Unseen. 13. Desire -- Clementine Von Radics. 14. The Is-Ought Problem. 15. The Naturalistic Fallacy. 16. The Big Questions — Steven E Landsburg. 17. Hindi Nationalism — Alok Rai. 18. Jahnavi and the Cyclotron — Episode 319 of The Seen and the Unseen (w Jahnavi Phalkey). 19. Atomic State: Big Science in Twentieth-Century India — Jahnavi Phalkey. 20. The Law of Truly Large Numbers. 21. Unlikely is Inevitable — Amit Varma. 22. Poisson Point Process. 23. The Major-General song from The Pirates of Penzance. 24. The HMS Pinafore song. 25. The New World Upon Us — Amit Varma (on Alpha Zero). 26. Alfred Hitchcock and The Beach Boys. 27. Roger Ebert on Mulholland Drive and Memento. 28. Joe Morgenstern and Bonnie and Clyde. 29. Creep -- Radiohead. 30. The Perils of Audience Capture -- Gurwinder Bhogal (on 'The Looking-Glass Self; and more). 31. David Bowie and Madonna. 32. Good Riddance (Time of Your Life) -- Green Day. 33. I Can't Stand the Rain -- Ann Peebles. 34. Fast Car -- Tracy Chapman. 35. Planet Claire -- The B52s. 36. Dunning-Kruger Effect. 37. Imposter Syndrome. 38. Bethany Hamilton on Wikipedia, Twitter, Instagram and her own website. 39. FOO Camp. 40. Gad Saad's poll on Like vs Respect. 41. Mary Oliver and Mark Strand. 42. The Prem Panicker Files — Episode 217 of The Seen and the Unseen (w Prem Panicker). 43. The Great Brain -- John D Fitzgerald. 44. Pippi Longstocking -- Astrid Lindgren. 45. Tom Lehrer on Spotify. 46. Eric Weinstein on Kung Fu Panda. 47. Kung Fu Panda -- John Stevenson. Check out Amit's online course, The Art of Clear Writing. And subscribe to The India Uncut Newsletter. It's free! Episode art: ‘Endless Possibility' by Simahina.
AI Today Podcast: Artificial Intelligence Insights, Experts, and Opinion
In this episode of the AI Today podcast hosts Kathleen Walch and Ron Schmelzer discuss and define at a high level DeepMind, AlphaGo, and AlphaZero. Show Notes: FREE Intro To CPMAI Mini Course CPMAI Training and Certification AI Today Podcast #94: Understanding the Goal-Driven Systems Pattern of AI AI Glossary Continue reading AI Today Podcast: AI Glossary – DeepMind, AlphaGo, and AlphaZero at AI & Data Today.