Yann LeCun

Overview

Computer scientist, AI pioneer, and one of the three "Godfathers of Deep Learning" (alongside Hinton and Bengio). Turing Award 2018. Longstanding Chief AI Scientist at Meta (since 2013). In 2026, founded AMI Labs (Autonomous Machine Intelligence Labs), a startup building World Models via his JEPA (Joint Embedding Predictive Architecture) framework.

AMI Labs (2026)

In March 2026, AMI Labs — co-founded by LeCun — raised $1.03B at a $3.5B valuation, backed by NVIDIA and the Bezos family fund. (TechCrunch)

⚠️ Verification needed: Whether LeCun fully departed Meta as Chief AI Scientist to run AMI Labs full-time has not been confirmed by a primary source as of this wiki page creation (2026-07-23). The entity/meta-ai page retains him as "Chief AI Scientist" pending clarification.

Core AMI Labs thesis (JEPA / World Models):

• LLMs learn only "indirectly through text written by humans" — insufficient for AGI
• JEPA learns abstract representations in latent space (not token prediction)
• Physical world understanding requires world models, not language models
• Long-term bet: robot and autonomous agents will require this architecture

ICML 2026 (July 7, Seoul): AMI Labs co-founder Pascal Fung presented the world-model thesis at ICML 2026. LeCun posted on X linking the thesis to ICML-presented work. (LeCun on X)

Connection to field context: Direct contrast to the dominant 2026 paradigm (scaling transformers, test-time compute, RL over LLMs). LeCun's JEPA is the primary structural alternative to transformer-based scaling.

Key Positions

Period	Role	Organization
2013–2026+	Chief AI Scientist	Meta AI (Meta)
2026–	Co-founder / AI lead	AMI Labs (external venture)
Ongoing	Professor	New York University, Courant Institute

Research Contributions

• Convolutional Neural Networks (CNNs) — 1989 LeNet; backbone of modern vision AI
• LeNet / MNIST — foundational CNN demonstration on handwritten digits
• JEPA (Joint Embedding Predictive Architecture) — energy-based predictive world model; alternative to LLM architectures
• Consistent critic of LLM scaling as the path to AGI since 2022

Person Weights

Interest weight: 1.2 per interests.md (Broad perspective; contrarian views also valuable)

Recent Activity

• 2026-07-07: AMI Labs presents world-model thesis at ICML 2026, Seoul
• 2026-03: AMI Labs raises $1.03B at $3.5B valuation; NVIDIA and Bezos fund as backers
• 2026+: Persistent public stance that LLMs cannot achieve AGI without world models and physical grounding

Related

• Meta AI — previous (ongoing?) affiliation
• Embodied Agents — JEPA's application target
• Reasoning Models — LeCun's foil: argues reasoning cannot emerge purely from LLM scaling