Co-founder & Executive Chairman, AMI Labs · Silver Professor, NYU · ex-Chief AI Scientist, Meta · Expert Watch · monthly scan · our rolling analysis
Why we watch him. Yann LeCun is one of the most credible senior dissenters against pure-LLM scaling, a Turing laureate who has put more than $1B behind the opposite bet. He marks the fault line between text-prediction AI and systems that model physical reality (world models, JEPA), the frontier question for where today's LLM tooling hits a ceiling. For Dominic he is the sharpest independent read on open-source AI, the path to human-level machine intelligence and the limits of the current stack, relevant to defence and intelligence data and to robotics-adjacent or reasoning-heavy applications. Monthly output is enough to track where the world-models argument goes.
Current headline view: LLMs are a useful product but a dead end for real intelligence. The next revolution is world models that learn how the physical world actually works, from video and sensory data rather than text alone. Open-source models are catching and passing closed ones, and the most exciting frontier work is coming from academia, not from scaling the current stack.
Latest 1-page summary (as at 10 Jul 2026)
The eight core theses we track him against:
LLMs cannot reach human-level AI. They "lack a model of the world" and "can't truly reason or plan" (his 2026 framing).
World models are the path forward. AI must learn the structure and dynamics of physical reality from video and sensory data, not text alone.
JEPA / V-JEPA. Predict in an abstract representation space, not in pixels or tokens; the core technology AMI is building.
Objective-driven architecture. A modular system (perception, memory, world model, actor, objectives) that plans by predicting the consequences of actions.
Open source is winning. "Open source models are surpassing closed ones", which reframes the US-versus-China debate.
Safety by construction. "Guardrail objectives" built into world models; he rejects the LLM-doomer framing.
Scaling is misdirected. The LLM-scaling obsession will not deliver AGI; the most exciting world-models work is coming from academia.
The grounding gap. Animals and babies acquire physical intuition without language; that sensory grounding is what current AI lacks.
Method note. This v1 baseline is distilled from the staged research on his 2026 talks, interviews and public statements (his X and Facebook presence, the AMI Labs framing, and recent conference and podcast appearances), verified 10 Jul 2026. It is not yet a full X post-by-post read. The monthly scan refines it from here, and flags any deviation. His LinkedIn slug was not fetched this session to confirm the canonical versus a shorter vanity form, and the academic site yann.lecun.com is from prior knowledge, not re-verified this session.
Analysis backlog (newest first)
Deep source docs are held locally in the workspace at expert-watch/, not on this page.
10 Jul 2026 · v1 profile built: role change captured (left Meta Nov 2025, now co-founder and Executive Chairman of AMI Labs, still NYU Silver Professor), surfaces, eight core theses, deviation baseline. Themes from the staged research on his 2026 talks, interviews and public statements.
Notable reactions (operators & researchers)
Up to three recent, high-signal reactions from operators or researchers (not commentators) to a specific new argument of his. Challenges first. As at 10 Jul 2026.
No high-signal reactions logged this cycle (v1 baseline). The first monthly scan will populate this.
Recent media
Recent YouTube videos and podcast episodes, most recent first. As at 10 Jul 2026.