Yann LeCun

Who they are

Yann LeCun is founder and CEO of AMI Labs — the Turing Award-winning deep learning pioneer who invented convolutional neural networks, co-developed the Lush programming language with Léon Bottou, and built DjVu, the image-compression technology that powers the Internet Archive's document collection.

Person

Yann LeCun took his Engineering Diploma from ESIEE Paris in 1983, then his PhD in Computer Science from Pierre and Marie Curie University in 1987, followed by postdoctoral work under Geoffrey Hinton at the University of Toronto — the lineage that produced the modern deep learning field. He spent the formative years of his research career at AT&T Bell Labs, where he invented convolutional neural networks for handwritten digit recognition, served as Head of Image Processing Research, and played jazz informally with colleagues; he also co-developed the Lush programming language with Léon Bottou and built DjVu, an image-compression format that became the backbone of the Internet Archive's scanned document collection. He moved to NYU as a professor across Computer Science, Neural Science, Electrical Engineering, and Data Science, and founded the NYU Center for Data Science. In 2013 he joined Facebook to found FAIR (Facebook AI Research) and eventually became VP and Chief AI Scientist at Meta, a role he held through the company's transformation into an AI infrastructure giant. He co-founded MuseAmi (music technology) and Element Inc. (biometric authentication) alongside his employed track. In January 2026, he left Meta to found AMI Labs (Advanced Machine Intelligence Labs), a direct bet that world models — AI systems that understand physical reality and can plan — will outrun the large-language-model path the rest of the industry is on. He posts prolifically on LinkedIn about world models, JEPA, self-supervised learning, LLM limitations, open-source AI, and AI safety skepticism; he has spoken at TED, AAAI, the AI Action Summit (2025), VivaTech 2026, and the IPAM UCLA Long Program (2026), and keeps a personal site at yann.lecun.com.

Company

LeCun founded AMI Labs (Advanced Machine Intelligence Labs) in January 2026 immediately after departing Meta, co-founding it with Alexandre LeBrun — former CEO of Nabla and founder of Wit.ai. In March 2026, AMI raised a $1.03 billion seed round at a valuation of approximately $3.5 billion, one of the largest seed rounds ever recorded. The company's first product is Marble, which generates physically sound 3D worlds — a demonstration of the world-model thesis that AI must internalize the structure of physical reality, not just predict tokens. The raise reflects a direct contrarian bet against the LLM-centric path dominant at OpenAI, Google, and LeCun's former employer Meta.

Market

AMI Labs enters a field where the dominant paradigm — large language models trained on text — is under genuine technical scrutiny: LeCun has argued publicly and at length that LLMs cannot reason, plan, or model the physical world, and AMI's world-model approach is a structural alternative, not an iteration. The closest conceptual rivals are research programs at DeepMind (world modeling for agents) and academic groups working on JEPA-style architectures, though no well-funded startup has staked out this exact position at scale. LeCun is simultaneously an advisor and investor in CuspAI, a Cambridge-based materials-science AI startup co-backed with Jeff Bezos, signaling that the world-model thesis has adjacent commercial applications beyond 3D content generation.

Network

LeCun's closest professional anchor at AMI Labs is co-founder and CEO Alexandre LeBrun, who previously founded Wit.ai and ran Nabla. His deepest scientific relationships are with Geoffrey Hinton (postdoc mentor, co-Turing laureate) and Yoshua Bengio (co-Turing laureate), and his longest research collaboration is with Léon Bottou, with whom he co-developed the Lush programming language at Bell Labs. He holds advisory roles at CuspAI and Element Inc., and was a senior leader alongside Mark Zuckerberg for over a decade at Meta.

Alexandre LeBrun· Co-founder and CEO, AMI Labs (former CEO of Nabla, founder of Wit.ai)
Geoffrey Hinton· Postdoctoral mentor; co-winner of 2018 ACM Turing Award
Yoshua Bengio· Co-winner of 2018 ACM Turing Award
Léon Bottou· Long-time collaborator; co-developed Lush programming language
Mark Zuckerberg· CEO, Meta — senior leadership counterpart during LeCun's tenure at Meta/FAIR

How they likely show up

Multi-decade tenures (Bell Labs, NYU, Meta) punctuated by a decisive break in January 2026 → thinks in long arcs but acts sharply when conviction is high enough.
Founded AMI Labs, MuseAmi, and Element Inc. alongside full-time research and executive roles → high agency; builds rather than advises when he sees a gap.
Prolific LinkedIn poster on world models, LLM limitations, open-source AI, and AI safety skepticism → comfortable with public disagreement, expects interlocutors to engage with the technical argument, not the social consensus.
Hybrid role pattern (researcher + executive + founder simultaneously) → likely impatient with meetings that don't advance either a research question or a product decision.
Co-developed a programming language (Lush) and an image-compression format (DjVu) as side outputs of primary research → works at the intersection of theory and artifact; values things that ship.
Active speaker at venues ranging from AAAI to French national television in 2026 → adjusts register for audience but the core thesis (LLMs are not enough) stays constant.

Conversation tips

→ Come in with a specific technical position on world models or JEPA — he engages with arguments, not with flattery about his Turing Award.
→ If you disagree with his LLM skepticism, say so directly and be ready to defend it; he debates publicly and respects a well-formed counter.
→ Reference Marble specifically (AMI's first product) rather than generic 'world model' language — it signals you've tracked the company past the funding announcement.
→ Don't conflate open-source advocacy with Meta loyalty; he left Meta to pursue a fundamentally different technical direction.
→ He has spoken at venues from TED to French TV in 2026 — he's not hard to reach, but he values people who have done the reading.

Toolbox

Openers

Open on Marble — AMI's first product generates physically sound 3D worlds, which is a concrete instantiation of the world-model thesis he's argued for years; ask what made that the right first product to ship rather than a research demo.
Reference his April 2026 Brown University lecture or the IPAM UCLA Long Program talk on world models — he's been on the road making the case publicly, and naming a specific recent appearance signals you've tracked the argument beyond the Lex Fridman episode.
He co-developed DjVu at Bell Labs, which ended up powering the Internet Archive — a side output that outlasted the research program that produced it. That arc from curiosity-driven artifact to infrastructure is directly relevant to what AMI is building now.

Discovery questions

JEPA (Joint Embedding Predictive Architecture) has been your proposed alternative to autoregressive LLMs for years — how does Marble's 3D world generation demonstrate the JEPA thesis in a way that a text benchmark can't?
You founded AMI in January 2026, raised $1.03 billion at seed, and shipped Marble — what does the research-to-product feedback loop look like inside a world-model startup at this stage versus FAIR, where the timeline pressure was different?
You're an advisory investor in CuspAI, which is applying AI to materials science for chips — how does that bet relate to the AMI thesis, and do you see world models as a general substrate for physical-world reasoning across domains?

Avoid

Don't lead with AGI timelines or ask whether LLMs will eventually reason — he has staked his career and $1.03 billion on the answer being no, and the question will read as you not having engaged with his public work.

Make it yours