Yann LeCun

Who they are

Yann LeCun is founder of AMI Labs — he left Meta after a decade as its Chief AI Scientist to start a $1.03B-funded startup building world models and co-founded 224 Ventures, a VC firm, alongside Oriol Vinyals.

Person

LeCun took his Engineering Diploma from ESIEE Paris in 1983, then his PhD in Computer Science from Pierre and Marie Curie University (Sorbonne) in 1987, followed by postdoctoral work under Geoffrey Hinton at the University of Toronto — the lineage that shaped modern deep learning. He spent formative years at Bell Labs and AT&T Labs Research, where he invented convolutional neural networks for document recognition, co-developed the Lush programming language with Léon Bottou, and built DjVu, an image-compression technology that powered the Internet Archive's document distribution. He then served as Founding Director of the NYU Center for Data Science and held a multi-role professorship there before joining Facebook in 2013 to found FAIR (Facebook AI Research). He served as VP and Chief AI Scientist at Meta for over a decade, becoming the company's most publicly visible AI voice. He also co-founded MuseAmi (2007), a music technology company that shipped apps including Improvox, Hook'd, and MusicPal, and Element Inc. (2012), a biometric authentication company. In 2026 he left Meta to found AMI Labs (Advanced Machine Intelligence), which raised $1.03B at a roughly $3.5B valuation to build world models — AI systems that reason about the physical world rather than predict tokens — with a first product called Marble generating physically sound 3D worlds; he also co-founded 224 Ventures, a deeply technical early-stage AI VC firm, with Oriol Vinyals and Shaun Johnson. He writes actively on LinkedIn about world models, JEPA, self-supervised learning, open-source AI policy, and sharp critiques of large language models, and has appeared on the Lex Fridman Podcast, with Kara Swisher, and at TED — the through-line is a researcher-turned-builder who bets his career on the ideas his peers haven't accepted yet.

Company

AMI Labs launched in 2026 as a newly founded startup with offices in Paris, Montreal, Singapore, and New York, and closed a $1.03B seed round at a roughly $3.5B valuation — one of the largest seed raises on record. The company's stated thesis is that large language models are a dead end for real intelligence, and its alternative is world models: AI systems that learn to understand and simulate the physical world. Its first product, Marble, generates physically sound 3D worlds. Alexandre LeBrun — former CEO of Nabla and founder of Wit.ai — was appointed CEO, giving LeCun a research-and-strategy seat rather than a day-to-day operating role. The company is simultaneously running research threads like TERRA, a JEPA-based world model for human tissue, signaling early moves into biology as well as 3D environments.

Market

AMI Labs is entering a market where LLM-centric labs — OpenAI, Anthropic, Google DeepMind — command the largest share of investor attention and enterprise spend, making its world-model thesis a deliberate contrarian bet. The competitive question is whether physical-world reasoning and planning capabilities can reach commercial scale before transformer-based approaches close the gap through scale alone. LeCun has also joined CuspAI's advisory board, a Cambridge-based materials-science AI startup co-backed by Jeff Bezos, suggesting AMI's world-model approach is being positioned as foundational infrastructure across multiple verticals, not just 3D media.

Network

LeCun's closest professional anchor at AMI Labs is Alexandre LeBrun (CEO), whose Wit.ai and Nabla background signals a preference for execution-focused operators paired with research visionaries. His academic and research network runs through Geoffrey Hinton (postdoctoral mentor and 2018 ACM Turing Award co-winner) and Yoshua Bengio (fellow Turing laureate), and his VC partnership through Oriol Vinyals at 224 Ventures. He also maintains advisory relationships with Element Inc. and CuspAI, the latter co-backed with Jeff Bezos.

Alexandre LeBrun· CEO, AMI Labs (former CEO of Nabla, founder of Wit.ai)
Oriol Vinyals· Co-founder, 224 Ventures
Geoffrey Hinton· Postdoctoral mentor; 2018 ACM Turing Award co-winner
Yoshua Bengio· 2018 ACM Turing Award co-winner
Léon Bottou· Co-developer of Lush programming language

How they likely show up

Decade-plus tenure at Meta as Chief AI Scientist followed by an immediate founding move in 2026 → thinks in long research arcs but acts decisively at inflection points rather than incrementally.
Publicly and repeatedly frames LLMs as insufficient for real intelligence (WAICF 2023, Lex Fridman 2024, multiple LinkedIn posts) → comfortable staking out unpopular positions and unlikely to soften a technical view to avoid friction.
Hybrid role pattern — researcher, professor, founder, VC, advisor simultaneously → high agency, will be bored by meetings that don't connect to a concrete research or product question.
Active LinkedIn poster on world models, JEPA, open-source AI policy, and AI regulation → comfortable being a public intellectual, responds well when interlocutors have read the actual technical work, not just headlines.
Co-founded five distinct entities (FAIR, MuseAmi, Element Inc., AMI Labs, 224 Ventures) alongside employed roles → treats side projects and institutions as legitimate parallel output, not distractions.
Personal hobbies include astrophotography and jazz (Bell Labs informal band in the 1990s) → wired for patient, detail-intensive craft; unlikely to be rushed toward shallow conclusions.

Conversation tips

→ Come in with a specific technical question about JEPA or world models — he has given multiple public talks on the architecture and will immediately assess whether you've engaged with the actual ideas.
→ Don't ask him to defend LLMs or play devil's advocate for the current consensus; he has a well-developed public critique and will spend the meeting re-litigating it rather than going somewhere new.
→ Reference AMI Labs' TERRA project (JEPA-based world model for human tissue) if the conversation touches biology — it signals you track his actual output, not just the funding headlines.
→ Ask about the 224 Ventures thesis separately from AMI Labs — he now has two distinct institutional bets and conflating them will signal you haven't done the work.
→ He is a Turing laureate who spent years being the most visible AI skeptic inside a major social platform — acknowledge the tension between that role and his current independence; he'll have a clear account of it.

Toolbox

Openers

Open on TERRA — he recently shared a JEPA-based world model for human tissue, a pointed move that extends AMI Labs' physical-world thesis into biology; it's a specific bet worth probing.
Lead with the 224 Ventures launch — he co-founded a deeply technical early-stage AI VC firm with Oriol Vinyals and Shaun Johnson in 2026, which sits alongside AMI Labs; it's an unusual dual-track move that reveals how he's thinking about the field structurally.
Reference his Lex Fridman Podcast appearance (#416) where he argued against LLMs and for world models — he gave a detailed account of why he thinks the field is on the wrong track, and it's the clearest public statement of the thesis AMI Labs is now trying to commercialize.

Discovery questions

Marble generates physically sound 3D worlds — what does 'physically sound' mean in practice as an evaluation criterion, and how do you know when a world model has actually learned physics versus memorized it?
You're now simultaneously running AMI Labs and 224 Ventures — when you're evaluating early-stage AI companies for the fund, what tells you a team actually has a world-model-native approach versus LLM fine-tuning dressed up in different language?
TERRA applies JEPA to human tissue — how different is the architectural work from Marble, and do you think the same world-model primitives generalize across physical domains or does each domain need its own inductive biases?

Avoid

Don't frame the conversation around AI safety timelines or AGI risk in the Hinton/Bengio sense — LeCun has publicly and repeatedly distanced himself from that framing and will spend the meeting correcting the premise rather than engaging with your actual question.

Make it yours