PodcastsHear the voice. See the shape of the thought.

Sebastian Mallaby spent three years and over 30 hours with Demis Hassabis in a British pub to write *The Infinity Machine*, and this conversation pulls the most underreported threads from that access: the 2015 safety summit that accidentally spawned OpenAI, the secret billion-dollar spinout plan Demis never used as real leverage, and the quasi-spiritual conviction about God and science that Mallaby never expected to find. The throughline is a paradox — Demis understood the race was dangerous from day one, but as leader of one lab, even a Nobel Prize-winning one, he could not stop it. ## [00:00] Intro Jacob Effron sets up Sebastian Mallaby as someone who has spent more time with Demis Hassabis than almost any journalist alive — 30-plus hours across three years of pub sessions in London. Mallaby's book, *The Infinity Machine*, covers the full arc of DeepMind from its 2010 founding through the Nobel Prize. The clips previewed here — Demis banging the table about God and science, Reid Hoffman's billion-dollar pledge, and the Elon feud — all come from later in the conversation. > *"Demis has a Nobel Prize. Sam didn't finish his first degree. Therefore, Demis doesn't take Sam very seriously."* ## [02:04] Was the AI Race Inevitable? Mallaby's verdict: yes, inevitable. Any technology this powerful would attract multiple labs across multiple countries, and China's stack was already competitive despite semiconductor shortfalls. What makes the story poignant is that Demis didn't believe this in 2010. He genuinely hoped one lab could carry the AGI project safely to the finish line — a singleton scenario where DeepMind was the anointed team. By the mid-2020s he had swung to the opposite pole: safety is a collective action problem that only governments can solve, because no single lab's restraint can bind the others. > *"I think it was inevitable. When you have this sort of supremely strong technology, there's going to be multiple labs in multiple countries that are just desperate to try and build it."* ## [04:03] The 2015 Safety Summit Backfire Summer 2015, SpaceX headquarters: Demis convenes a small summit to bring Elon Musk inside the tent — the plan was for Elon to chair a safety oversight board and, critically, not launch a competitor. By end of year, OpenAI existed. Mallaby frames this as the moment Demis internalized that voluntary collaboration between lab leaders is structurally impossible. The only mechanism he now believes can work is a government enforcer setting uniform rules — mandatory pre-release testing, safety slow-downs — with US-China cooperation as the endpoint, however remote that prospect appears. Jacob pushes on whether lab leaders actually believe government intervention is achievable; Mallaby draws a parallel to the FDA: slow, imperfect, but it does adjudicate whether drugs are safe enough to ship. > *"You can't trust the other guys. The only way you get trust is if you have a government enforcer that comes along and says, 'Here's the rules for everybody. There's going to be a level playing field. You're all going to have to abide by some sort of safety slow-down.'"* ## [11:27] Why Google Doesn't Make As Concentrated Bets Jacob points to the two defining consumer-AI moments of the era — ChatGPT and Claude Code — and neither came from Google DeepMind despite its leaderboard dominance. Mallaby traces this directly to Demis' intellectual formation: a PhD in neuroscience, a broad theory of intelligence, a lab culture that says "whenever there are two paths, do both, find a third." The result is a heavily hedged research portfolio that is excellent at producing Nobel Prizes and state-of-the-art models but structurally slow to make the kind of one-directional product bet Anthropic made on coding. Gemini is bundled into Google Search, so usage is higher than it appears — but Mallaby concedes the product-zeitgeist gap is real. > *"Anthropic got to coding because it was willing to take a more concentrated bet. It never went into the whole field of, you know, everything at once."* ## [15:51] Project Mario: The Secret Spinout Plan The book's most explosive scoop: DeepMind had a secret plan — code-named Project Mario — to spin out of Google, backed by a $1 billion pledge from Reid Hoffman. Mallaby had to fight Google's general counsel to publish it. The motive was not entrepreneurial independence but safety leverage: Demis wanted formal safety oversight over DeepMind's models, Mountain View wasn't providing it, and a credible spinout threat was his negotiating chip. He never explicitly told Google about the Hoffman pledge, but pushed hard knowing the option existed. In the end he chose to stay — legal risk of the spinout fight, desire for compute access, and a preference for doing science over litigating corporate structure. A year later he shipped AlphaFold and won the Nobel Prize. > *"Demis really really wanted to get safety oversight over the Google DeepMind models. Google corporate in Mountain View wasn't doing that. So he had to have a credible threat of spinning out. He went to Reid Hoffman. Reid Hoffman pledged a billion dollars to finance a spinout — and Demis used that to kind of pressure Google."* ## [19:43] What Demis Actually Regrets On AlphaFold and AI-for-science: no regrets at all — Mallaby argues it was not only scientifically correct but politically necessary, because AI needs visible social benefits to survive the coming backlash against job disruption. The genuine regret is speed. Demis missed the transformer moment the way Ilya Sutskever did not: when the paper dropped, Ilya ran down the corridor to find Alec Radford to build a language model. Demis' broad-portfolio instinct meant DeepMind studied the transformer but didn't bet the lab on it. Missing that window — and the ChatGPT moment that followed — is a real failure, not just a stylistic difference. > *"Ilya is like jumping out of his chair, running down the corridor going to find Alec Radford saying, 'Hey, we're going to build a language model based on this transformer architecture.' On the day they won AlphaGo, Demis was already on to bio — and someone picked it up on a mic."* ## [23:46] Venture Startups vs. Tech Behemoths The broadest structural argument in the episode: does venture-backed concentration beat hyperscaler breadth in AI? Mallaby has written about both (his previous book covered venture capital) and calls it genuinely balanced. Hyperscalers have unlimited capital and can sustain a multi-year arms race; the problem is that unlimited resources breed portfolio thinking, which bleeds attention. Startups with one concentrated bet can move faster on that specific bet. Mallaby's live position: OpenAI has roughly 50/50 odds of being absorbed or failing before next summer — not because the tech is weak, but because the business model can't sustain indefinite losses against Google's balance sheet. He also floats that Anthropic should IPO right now while its brand is strongest. Jacob notes the robotics parallel: fifteen different approaches being funded simultaneously, and whoever picks the one that works the way transformers did will dominate. > *"I wrote in the New York Times in January that I thought OpenAI had a 50% chance of going bust by next summer. Is it still 50? Yeah. The tech is great. It's just the business model — and you're up against Google, which just has unlimited amounts of cash to spend you into the ground."* ## [34:08] David Silver and the RL True Believers David Silver — AlphaGo's lead researcher and co-author of the "reward is enough" paper with Rich Sutton — left DeepMind after the book came out to start a new company. Mallaby reads the departure as structurally inevitable: Silver is a pure reinforcement learning absolutist who believes learning from human data is fundamentally inferior because it encodes human errors. His thesis is that self-play and environment-generated experience is the only path to genuine superhuman performance. Demis told Mallaby this view may ultimately be correct *after* AGI is achieved — but the entire language model revolution showed that bootstrapping with human data is what gets you to AGI in the first place. Silver's RL purism was too far ahead of the current paradigm for his colleagues to follow. > *"David is just very very hard over on that vision — learning from data is inferior because the data includes mistakes. The machine needs to learn from its own experience, not rely on the crystallized knowledge of humans passed on through text."* ## [38:21] Demis, Elon, and the Evil Genius Feud The origin story: at a Founders Fund LP offsite in 2012, Elon argues that SpaceX matters most because even if AI wrecks Earth, humanity can move to Mars. Demis replies that his AI will eventually conquer space flight and follow them there. Elon goes quiet, then writes a $5 million check into DeepMind's Series B. Two years later, hearing Google was acquiring DeepMind, Elon and Luke Nosek Skyped Demis from a party closet in LA in the middle of the night, begging him not to sell to Larry Page. Demis said no, hung up, and Elon started calling him "evil genius" — the name of a video game Demis had designed. Mallaby characterizes Demis' view of Sam Altman as colored by the credential asymmetry: Nobel Prize winner vs. someone who didn't finish a degree. The relationships between these founders are less professional rivalries than a collection of specific personal slights and competitive provocations playing out over fifteen years. > *"Demis says, 'Yeah, but if you think you're going to be safe on Mars, remember that my AI will be able to conquer space flight, and it will just follow you to Mars. So then you won't be safe after all.' There's a silence. Then Elon goes, 'Hm.' And then: 'I'd like to invest in your Series B.'"* ## [42:39] Great Man Theory vs. Inevitability Jacob cites *The Economist*'s framing of the book as a test of great-man theory. Mallaby draws a parallel to his Greenspan biography: Greenspan understood bubbles were dangerous (literally the subject of his PhD), yet couldn't stop the 2008 crisis. He considered titling the Demis book *The Man Who Knew* for the same reason — Demis knew from the start this technology was dangerous, but one lab's restraint cannot bind the rest. Individual leaders do matter at the margin: Dario Amodei changed the safety narrative through the Anthropic mythos release; Sam Altman shaped the race by shipping ChatGPT while it was still hallucinating; Demis shaped it by persuading Rishi Sunak to host the UK AI Safety Summit. But the race itself? Structurally overdetermined. > *"I feel that one could have almost used the same title for the Demis book — 'the man who knew' — because Demis has known from the beginning that this thing is dangerous. But as the leader of one lab, even a very powerful rich lab, even he with his stature as a Nobel Prize winner — what can he do?"* ## [45:00] What Demis Didn't Want Published The detail Mallaby least expected: Demis is driven by something close to a spiritual conviction about science. In those two-hour pub sessions he would bang the table about the mystery of matter — why atoms cohere into a solid table, why silicon and copper can think — and say, unprompted, "Maybe if we approach science the right way, we will be getting closer to something that we could perhaps call God." Mallaby reads this as the psychological engine that lets Demis keep pushing a technology he knows to be dangerous: it's a quasi-spiritual quest, not just a commercial one. On what Demis blocked from publication: his family (he set that limit at the start), and his internal fights with Sundar Pichai — he didn't want to destabilize the Google relationship he still depends on. > *"He would start banging the table and saying, 'Maybe if we approach science the right way, we understand more about nature. We will be getting closer to something that we could perhaps call God.' I had no idea he would feel that way."* ## Entities - **Demis Hassabis** (Person): Co-founder and CEO of DeepMind / Google DeepMind; Nobel Prize winner in Chemistry (2024) for AlphaFold; central subject of *The Infinity Machine*. - **Sebastian Mallaby** (Person): Staff writer at *The New Yorker*; author of *The Infinity Machine* (Demis Hassabis biography) and a prior book on venture capital; spent 30+ hours with Hassabis over three years. - **Jacob Effron** (Person): Host of *Unsupervised Learning*; Managing Director at Redpoint Ventures. - **Reid Hoffman** (Person): LinkedIn co-founder; pledged $1 billion to finance DeepMind's potential spinout from Google under Project Mario. - **David Silver** (Person): Lead researcher on AlphaGo and AlphaZero at DeepMind; co-author of the "reward is enough" RL paper with Rich Sutton; departed DeepMind post-publication to start a new company. - **Elon Musk** (Person): Hosted the 2015 AI safety summit at SpaceX; early DeepMind investor; coined the "evil genius" nickname for Hassabis after DeepMind sold to Google. - **Sam Altman** (Person): CEO of OpenAI; shipped ChatGPT in late 2022 despite hallucination issues, which Mallaby argues irreversibly shaped the AI race's trajectory. - **Dario Amodei** (Person): CEO of Anthropic; credited with changing the AI safety narrative through the mythos paper release and his public Pentagon confrontation. - **DeepMind** (Organization): Google subsidiary; founded by Hassabis, Shane Legg, and Mustafa Suleyman in 2010; produced AlphaGo, AlphaFold, and Gemini. - **Project Mario** (Concept): Secret DeepMind plan to spin out of Google, backed by a Reid Hoffman $1B pledge; used as negotiating leverage for safety oversight, never executed as a real spinout. - **AlphaFold** (Software): DeepMind's protein-structure prediction model; won Hassabis the 2024 Nobel Prize in Chemistry; shipped in 2020, one year after he declined the spinout option. - **Reinforcement Learning** (Concept): Machine learning paradigm central to AlphaGo and AlphaZero; David Silver's absolutist commitment to RL (learning from environment experience over human data) created internal tension at DeepMind and ultimately led to his departure. - **The Infinity Machine** (Concept): Sebastian Mallaby's biography of Demis Hassabis; nearly titled *The Man Who Knew*; published with the full Project Mario scoop over Google's objections.

Oriol Vinyals（Google DeepMind VP of Research、Gemini 联合负责人）在 Google I/O 第二天坐下来，把 I/O 上发布的产品背后的研究路线一条条摊开：世界模型为什么是 Google 押向 AGI 的独特路径、视频 / 图像的"GPT moment"长什么样、Spark 和 agents 系统为什么必须和模型联合优化、scaffolding 终将由模型自己写、memory 应该走非参数 file-system 而不是塞进权重、当今 RL 在哪些维度上是数据受限的、为什么 math/code 上的训练能意外迁移、以及 Google 内部 Brain + DeepMind 合并后研究下注的取舍。 ## [00:00] Intro Jacob 用 60 秒铺垫了 Oriol 的背景（Gemini 联合负责人，与 Noam Shazeer、Jeff Dean 并列），以及 I/O 第二天访谈的优势：所有发布都还热乎，可以直接顺着 announcements 追到背后的研究。Oriol 进来打招呼，两人开始热身。 > *"I've been really excited for this because you're one of the people kind of most directly shaping the frontier of AI."* ## [01:36] Why World Models Jacob 先问"为什么是世界模型"。Oriol 把它拆成两层：一层是 self-improvement / coding 的角度，另一层是模型本身的对象——多模态、不止 closer 还包括 video / image 这种"world model"。Google 早就押了图像和视频路线，这次"显然押对了"，因为我们其实把整个世界都搬到了互联网上。 他也承认中间有一段时间这条路看似不性感：multimodal 模型在 LLM 风口下被边缘化过，但视频和图像里藏着语言抓不到的知识——"the GPT moment for video"还没真正发生，但拐点已经在视野里。 > *"There is lots of knowledge in videos and images, and what I would say is the GPT moment for that — I'm not sure we quite have seen that."* ## [04:21] The GPT Moment for Video Oriol 用 Omni（Google 的多模态产品线）当锚点解释：从单纯把视频喂进上下文，到能在长上下文里理解和生成视频，这段曲线已经很陡。下一步是问"能不能像 LLM 一样，在没有 paired text 的纯图像数据上预训练并依然提取出全部意义和细节"——这个 hard challenge 一旦解开，数据维度会从"被人类描述过的"跳到"所有视频"，量级差异巨大。 他特别承认现在 video 这块的标注数据相对 image 仍然稀缺，但解锁后的回报会"非常大"。 > *"Whether we agree with that or not is another question, but if it was to be unlocked, it would be massive."* ## [07:51] What Makes Omni a World Model "world model"这个词被滥用了，Oriol 给一个清晰定义：一个纯粹的 world model 必须做 representation learning——把世界压成紧致表征。在这之上，Omni 进一步成为可被语言驱动的 renderer：你用自然语言改一个 prompt，输出的视频内容随之改变，初始 image 之上能持续演化。这是从"被动建模"到"可控生成"的关键区别。 > *"The world model itself is acting as a renderer of the world, that you can really just change by language."* ## [10:04] World Models & Robotics 机器人是 world model 最直接的落地场景。Oriol 承认现在数据 mix 还在试错——sim 数据 vs 真机数据怎么配、什么时候 transfer 突然 click。世界模型本身的进步会带来一个 inflection point：一旦模型足够强，sim → real 的鸿沟会缩到 planning 和 gross motor 层面先打通，精细运动控制再慢慢跟上。 > *"At some level, maybe not at the precise motor control but at the kind of planning and gross, we are going to start seeing how things are going to fall into place."* ## [12:37] Evaluating Physics in AI 模型隐式学物理，但你怎么评估它学到没学到？Oriol 把它和无监督机器翻译做类比：如果模型内部确实表征了"重力"这个概念，应该能用某种 decode 把它翻译成显式 explanation。Stefano Gaus 等人 2014 年的早期 unsupervised translation 工作给了一条可借鉴的思路——把内部表征解码出来当 eval。 > *"You would need to somehow connect the concept of gravity which could be present or not in a world model to then decode that into an explanation."* ## [14:51] Consumer Agents & Spark I/O 发布的 Spark 是 Google 在 consumer agent 上的最新一步。Oriol 强调："action 作为一种 modality"已经被 DeepMind 早早识别为关键。但 agent 不是把模型塞进 generic scaffold 就行——模型能力必须先到某个门槛，你才能 dream 出下一阶段的产品形态。 他给一个工程判断：在 train 阶段就把"我有这些能力，怎么挑用哪些"内化进模型，比在 inference 时让外部 scaffold 临时决策更高效。 > *"It's useful to build kind of the system slightly more narrowly around something you care deeply about."* ## [18:39] Scaffolding & the Bitter Lesson Oriol 多年支持 Sutton 的 bitter lesson。Jacob 把它推到 agent 时代：scaffolding 看起来违背 bitter lesson 因为是手写的胶水。Oriol 的答案是——"scaffold 本身就是一段 code，最终应该是模型自己 on the fly 写出来"。短期内人写、长期模型写，bitter lesson 仍然站得住。同时优化 model 和 scaffold 两端，而不是把所有赌注押在一端。 > *"That system itself is a piece of code that eventually the model itself could write on the fly."* ## [22:06] Memory & Continual Learning Memory 这个话题 Oriol 谈得最深——他有 cognitive neuroscience 背景。他把 memory 分成两类：塞进权重（参数化）和挂在外部 file system（非参数化）。在 serving 规模下，把每次 user interaction 都 bake 进 weight 是不切实际的，非参数式 file-system memory 更可行。 真正的难点是"consolidate"：怎么把之前 session 的信息整合到新 session，让模型像人一样积累知识。这部分 momentum 很大但远未饱和，未来几年评估方式和工程实践都会迭代。 > *"The way that we'll see better evaluations and ways in which these models accumulate this knowledge as they go."* ## [26:54] Research Bets Inside Big Labs 在 Google 内部主导 Gemini 是什么体验？Oriol 谈三个维度的优势：TPU 联合设计（不用看 Nvidia 脸色）、广告/搜索带来的现金流稳定性、Brain + DeepMind 合并后端到端的研究强度。劣势是：组织太大没法对所有方向有全视野，必须靠直觉判断哪些早期研究值得 pull in，并接受"trade-off 不可能每次都做对"。 > *"Google is in a unique place. We have stability from hardware procurement and obviously like also investment of capital."* ## [32:30] Post-Training RL is Greenfield post-training 这块仍然是一片 greenfield。在 coding 和 math 上 LLM 已经走出指数曲线，但其他领域为什么没跟上？Oriol 的核心判断是"投入还远远不够"——相对预训练的算力消耗，post-training 至今只用了很小一部分。算法的 beauty 还在迭代，"cracking that recipe could be big"。 > *"Cracking that recipe could be big, at least in terms of the beauty of the algorithm."* ## [35:57] What Real Intelligence Looks Like 真智能长什么样？Oriol 用 2015 年的一个老 eval 来当锚——简单的 game-playing 任务，当时是 RL 的天花板，现在 LLM 一上来就能做。他想看到下一个数量级的跃迁：不是在熟悉的 benchmark 上推数字，而是在新的、人类没法立刻给出答案的问题上看到模型"主动产出洞察"。 > *"I like games."*（这句简单的自陈背后是他对 game-playing RL 长期偏爱的注脚） ## [39:11] RL Generalization 游戏曾经是 verifiable reward 的典型样板。现在的挑战是找新的 hard problem source，让 RL 在更广的领域诱发出深度推理和泛化。Oriol 抛出一个不对称观察：create solution 和 evaluate solution 之间存在 gap——如果 evaluation 比 generation 容易，RL 就有机会撬动。 让他意外的是：在 math/code 上的训练能 surprisingly 迁移到其他领域，"很多泛化能力可能其实来自 pre-training"。这是接下来几个月到几年研究者要破解的关键题。 > *"Possibly through pre-training — that's one of the quests for researchers to crack in the next few months and years."* ## [42:55] Advice for Founders 给 founder 的建议直白：evaluation 和 data 是绕不开的 moat。早期专注垂直产品、在 model 上叠一层 specialized scaffolding，等到 scale 起来再考虑 model layer 的差异化——这个路径"比较 scalable，也更适合早期玩家"。 > *"What I would tell folks is the value — and we discussed this a little bit — the value of evaluations and as a sequence of data."* ## [46:40] Can AI Truly Innovate? Oriol 2016 年加入 DeepMind 后最痴迷的方向是 meta-learning——模型自己产出 idea。但他承认到目前为止，"我没看到模型生成真正 outstanding 的 idea"。他比喻：你让一万个人尝试，挑出对的那个再 glorify，但模型真正自主提出方向的能力——quite limited。但他相信 "soon"。 > *"I don't think I've seen truly kind of outstanding ideas that a model has generated yet, but I am sure I will very soon."* ## [49:48] Recursive Self-Improvement 递归自我改进可以分层看：第一层是 researcher / engineer 用 AI 工具加速自己；第二层是模型直接自动化某些研究任务。当模型写英文比你好的那一天，下一个 ceiling 在哪里？Oriol 说："maybe there's no ceiling, or the ceiling is still far away" —— 我们甚至不一定能看到 ceiling 在哪里。 > *"At the point a model writes English better than you, maybe there's no ceiling, or the ceiling is still far away."* ## [52:14] Quickfire 最后 8 分钟快问快答覆盖了 TPU 投资历史、给年轻研究员的算力直觉、当下 AI 阶段的总体感受。Oriol 留下一句总结："I think it's a fascinating time as anything in AI"。Jacob 用 podcast 致谢和 outro 结束。 > *"I think it's a fascinating time as anything in AI."* ## Entities - **Jacob Effron**（人物）：Redpoint Ventures Managing Director，Unsupervised Learning 主持人。 - **Oriol Vinyals**（人物）：Google DeepMind VP of Research，Gemini 联合负责人（与 Noam Shazeer、Jeff Dean 并列）。 - **Gemini**（产品）：Google 的旗舰多模态 / agent 模型族；本期主要谈 I/O 第二天的发布。 - **Omni**（产品）：Google 的多模态产品线，被用作"video / image 的 GPT moment"参照系。 - **Spark**（产品）：I/O 发布的 consumer agent 产品。 - **World Model**（概念）：可被语言驱动的世界 renderer；representation learning 是其核心要素。 - **Bitter Lesson**（概念）：Sutton 的论点；本期延伸为"scaffold 长期应由模型自己写"。 - **Memory / Continual Learning**（概念）：非参数 file-system memory vs 把记忆塞进权重；consolidation 是关键难点。 - **Post-Training RL**（概念）：相对预训练的算力投入还很少，被定性为 greenfield。 - **Move 37**（概念）：AlphaGo 那一手；Oriol 用它指代"真正的 RL/research breakthrough"基准。

Yann LeCun, Turing Award winner and founder of AMI Labs, lays out his case that LLMs are a productive dead-end — genuinely useful products, but structurally incapable of modeling physical reality, planning, or predicting the consequences of actions. He walks through the JEPA architecture as the alternative, explains the Tapestry federated-learning project for non-US/China AI sovereignty, and pulls back the curtain on why his time at Meta ended: the GenAI organization's short-term pressure gradually made breakthrough research politically untenable. His predicted timeline for the paradigm shift: early 2027. ## [00:00] Intro Jacob Effron opens with a quick-cut preview of the conversation — Yann joking about "five years, complete world domination," teasing his blunt take on his relationship with Meta's Llama program, and flagging how his views on unsupervised learning ultimately pointed away from LLMs. Jacob then frames the episode as a rare chance to hear from someone who both built foundational open-source LLMs and now argues, publicly and consistently, that scaling them further is the wrong bet. > *"The best way to get breakthrough research is you hire the best people. You get the hell out of the way."* ## [01:45] Why LLMs Aren't the Path to Intelligence Yann draws a sharp line between LLMs as products and LLMs as a path to intelligence. They work well precisely because language is special — a low-dimensional, discrete, highly structured substrate where autoregressive prediction is tractable. Reality is not like that. The physical world is high-dimensional, continuous, and chaotic: a robot picking up a mug, a self-driving car navigating a construction zone, a cell responding to a drug. These are not language problems, and architectures optimized for language cannot acquire the internal models needed to reason about them. His company, AMI (Advanced Machine Intelligence), is built on the counter-thesis: that the right path is systems which learn abstract world representations from raw sensory data — video, sensor feeds, industrial telemetry — and can plan by simulating the consequences of candidate actions inside those representations. > *"They're just not a path towards human level or human like intelligence or even animal-like intelligence. That's my claim. I'm not saying they're useless — I'm just saying they're not a path towards that."* ## [07:51] AMI and World Models "World model" has become a buzzword, Yann notes, and the field has split into two camps: generative approaches (video models, VLAs) and joint-embedding approaches like JEPA. He dismisses VLAs — vision-language-action models trained to produce robot actions — as already widely recognized failures: brittle, data-hungry, unable to generalize. The generative video approach has the same structural flaw as LLMs: it predicts every pixel rather than learning the abstract structure underneath. A world model, properly defined, is a system that lets an agent anticipate the consequences of its own actions before committing to them. Without that, any agentic system is operating blind — no ability to verify whether a planned sequence of actions will actually accomplish the goal. > *"I cannot imagine how you can even think of building an agentic system without that system having the ability to predict the consequences of its actions."* ## [12:07] The JEPA Architecture Explained The insight behind JEPA came from a pattern Yann noticed across years of self-supervised learning research: every architecture that successfully learned useful representations of images and video was non-generative. Generative architectures — VAEs, masked autoencoders, pixel-prediction models — consistently underperformed. JEPA takes a corrupted or partial view of an input, runs both versions through encoders, and trains a predictor to match representations — not raw pixels. That abstraction is the point. The 2022 "A Path Towards Autonomous Machine Intelligence" paper was his attempt to write down the full blueprint: JEPA as the perception backbone, objective-driven planning on top, and a hierarchical structure of world models at different time scales. He describes publishing it as "spilling all my secrets" — a deliberate bet that openness would rally more talent to the paradigm than secrecy would protect. > *"I've been really interested in that problem of learning models of the world by prediction for a very long time, and then had an epiphany about five years ago realizing that all of the architectures that have been successful to learn representations of images and videos are non-generative architectures and all the generative ones basically have been failures."* ## [15:55] Problems with Robotics Models Today Current robotics demos are impressive but trained with enormous volumes of imitation data — teleop recordings, hand-tracked demonstrations — and fine-tuned with RL mostly in simulation. That pipeline produces brittle specialists. A 17-year-old learns to drive in roughly 20 hours; we have millions of hours of driving footage and still no level-5 autonomous car. The gap between imitation learning and genuine generalization is the gap between memorizing examples and having an internal model of the world. Yann's claim for world-model-based systems is zero-shot task generalization: given a new goal, a system with an accurate internal world model can plan a sequence of actions to reach it without being explicitly trained on that task. The near-term industrial applications he's targeting — controlling jet engines, chemical plants, manufacturing lines — are settings where the inputs are already numerical and a world model can be trained directly from operational data. > *"The degree of generalization you would get with a world model based system is much much larger — a wider spectrum of tasks with less training data than a system trained with imitation learning."* ## [20:37] Silicon Valley Herd Behavior Yann's diagnosis of why the entire industry converged on scaling LLMs is structural: once you're behind, you can't afford to work on anything else. The competitive race creates a rational incentive for every major lab to dig the same trench. He founded AMI Labs in Paris specifically to escape this — the American office is in New York, not Silicon Valley — and raised no Silicon Valley VC money. His predicted timeline for the paradigm shift is early 2027. "World model" is already becoming a research buzzword; industry has recognized that VLAs failed; and the robotics sector's unsolved generalization problem is a forcing function. He doesn't claim AMI will have a full solution by then, but he expects it to be obvious to everyone by that point that a change of paradigm was necessary. > *"I think the realization that you need a change of paradigm is happening as we speak and will become completely obvious to people by early 2027."* ## [28:18] Tapestry: Sovereign AI for the Rest of the World Tapestry is a separate project from AMI, built around one observation: as smart glasses and AI assistants become the primary information interface, whoever controls the underlying model controls the information diet of billions of people. A farmer in India, a philosopher in Germany, a citizen in Morocco — none of them are well-served by a model whose training data, values, and political priors were set by a handful of people in California or Shenzhen. The solution is federated training: countries and institutions contribute data and compute, but never share raw data with one another. They share parameter vectors. Each contributor trains locally, periodically exchanges parameter updates, and pulls a running consensus model — a repository of all human knowledge that no single party controls. Countries from India to Kazakhstan to France have expressed interest, because AI sovereignty has become a political priority independent of any technology choice. > *"All of your information diet will be mediated by AI assistants, and if that AI assistant was built in California or Beijing, it's not good for you."* ## [35:49] OpenAI Is the Next Sun Microsystems Proprietary LLM providers have already exhausted publicly available text data. The remaining path — licensing copyrighted material or generating synthetic data — is expensive and bounded. Open-source models have been closing the gap without that constraint. Yann draws the analogy to the 1990s Unix workstation market: Sun Microsystems, HP, and SGI all had technically superior proprietary systems and compelling arguments for why you wouldn't run a web server on Windows NT — and were all wiped out by Linux. The entire internet now runs on Linux. OpenAI and Anthropic, he says, are the Sun Microsystems of this cycle. > *"Basically, OpenAI, Anthropic, etc. of today are the Sun Microsystems and HPUX of yesterday."* ## [40:51] Why Yann's Views Diverged from Hinton & Bengio The split happened in 2023. Yann's position didn't change — Hinton's and Bengio's did. Hinton encountered GPT-4 and concluded it was close to human-level intelligence, reasoning from a back-of-the-envelope calculation about cortical neuron counts. Yann thinks that argument is wrong and reads it as Hinton finding a justification to declare victory and retire from active research. Bengio's shift was different — more focused on societal risks from AI concentration of power — and Yann has more sympathy for that concern, even though he disagrees with the apocalyptic framing. > *"I do not believe in this claim at all. This is kind of Jeff's way of saying, okay, basically I can retire — I can declare victory."* ## [44:32] LLMs Are Intrinsically Unsafe Yann's strongest claim: LLMs cannot be made reliably safe, not because alignment is hard, but because the architecture is structurally incapable of predicting the consequences of its actions. There is no hardwired constraint ensuring a prompted LLM actually accomplishes the intended task; it accomplishes whatever its training conditioned it toward, and there is always a gap between training distribution and real-world prompts. Coding agents wiping hard drives, medical advice going wrong, agentic systems taking irreversible actions — these are not bugs to be patched but properties of the architecture. His alternative, objective-driven AI, works differently: the system has an explicit world model, an explicit cost function representing the goal, and a set of hard safety constraints. The optimizer finds a sequence of actions that satisfies all constraints and minimizes cost — meaning it literally cannot take an action that violates a safety constraint by construction. That guarantee is impossible with an LLM. He also disputes Anthropic's lobbying narrative on AI risk, arguing that real danger comes from bad actors using current systems, not emergent superintelligence, and that regulatory pressure primarily benefits incumbents. > *"LLMs are intrinsically unsafe. I don't think they can be made reliable and safe. They cannot be made reliable because you can't stop them from hallucinating."* ## [58:00] Why Yann Left Meta Yann corrects a widespread misconception: he had zero technical influence on Llama. Llama 1 was a small FAIR project; when GenAI was created in early 2023, the Llama team moved there and was placed under intense short-term product pressure. Two of the Llama 1 authors left to found Mistral. GenAI became conservative and increasingly publication-restricted. FAIR, meanwhile, was being redirected to support GenAI's LLM work rather than pursue the AMI research agenda that Yann, Zuckerberg, and the CTO had all originally backed. By early 2024, the environment was no longer conducive to breakthrough research. > *"Here's a big misconception about my role, my relation to Alex, and how AI was run at Meta."* ## [01:00:26] Reflections on FAIR Yann joined Facebook in late 2013 and ran FAIR for four and a half years before stepping down to become Chief AI Scientist — a deliberate move because, as he says, he is not a natural manager. The internal AMI project grew out of his 2022 vision paper, which Zuckerberg, the CTO, and the CPO all read and backed. But layers below leadership didn't see the point, and Meta's decision to shut down its entire robotics AI group — led by Gita Matarić, now at Amazon — made clear the company had no interest in the applications world models were built for. Publication restrictions tightened, good researchers left, and the mismatch between Yann's research agenda and Meta's product priorities became irreconcilable by early 2025. When he went to raise money for AMI, investors already knew his story from years of public talks and were primed to believe LLMs had fundamental limits. > *"The best way to get breakthrough research of the type we were getting in the early days of FAIR and at Bell Labs is you hire the best people — you give them the means to succeed and you get the hell out of the way."* ## [01:12:11] Advice for PhD Students Yann opens by reflecting that his prediction self-supervised learning would succeed for video was correct in its mechanism but wrong about where it first succeeded: LLMs are "a blindingly successful example of self-supervised learning," just applied to language rather than sensory data. He then gives the core technical challenge for JEPA: representation collapse. If you train a predictor to map one embedding to another, the trivially optimal solution is for both encoders to output a constant. Contrastive learning (his 1993 invention) prevents collapse but doesn't scale with dimension. Distillation methods like DINO work but for poorly understood reasons. His current best answer, SIGreg (Sketched Isotropic Gaussian Regularization), forces the encoder output distribution to be Gaussian, maximizing information content without negative pairs. He recommends the LeWorldModel paper — the first small-scale world model trained with this approach — as the single best entry point into where AMI Labs is headed. His advice to PhD students: don't work on LLMs — you can't contribute from academia without frontier compute, and studying why they work is descriptive science, not creative research. > *"An LLM works because when you have a sequence of discrete symbols, making predictions is easy. If you have the real world, you can't use a generative model — you have to train a system that learns a representation and makes predictions in the representation space."* ## Entities - **Yann LeCun** (Person): Turing Award 2018 co-winner; former Chief AI Scientist at Meta FAIR; founder of AMI Labs; professor at NYU; inventor of convolutional neural networks and co-creator of JEPA - **Jacob Effron** (Person): Partner at Redpoint Ventures; host of Unsupervised Learning podcast - **Geoffrey Hinton** (Person): Turing Award co-winner; reversed position on LLM capabilities after GPT-4; less vocal on AI dangers since 2024 - **Yoshua Bengio** (Person): Turing Award co-winner; focused on societal risks from AI concentration rather than emergent superintelligence - **JEPA** (Concept): Joint Embedding Predictive Architecture — predicts in representation space rather than pixel space; forms the perceptual backbone of Yann's world-model framework - **World Model** (Concept): Internal model enabling an agent to predict the consequences of its own actions before committing to them; prerequisite for safe agentic AI in Yann's framework - **Tapestry** (Concept): Federated LLM training project enabling countries and institutions to train a shared foundation model while retaining data sovereignty through parameter-vector exchange - **AMI Labs** (Organization): Yann's company (Advanced Machine Intelligence); headquartered in Paris, US office in New York; focused on JEPA-based world models for robotics, industrial control, and healthcare - **Meta FAIR** (Organization): Facebook AI Research; origin of Llama 1, I-JEPA, V-JEPA, and the AMI internal research program; increasingly redirected toward GenAI LLM support before Yann's departure