Yann LeCun

Type: Person (AI Researcher)

Yann LeCun is a pioneer in deep learning — co-recipient of the 2018 ACM Turing Award with Geoffrey Hinton and Yoshua Bengio — and currently VP & Chief AI Scientist at Meta, where he leads Meta FAIR's research into Advanced Machine Intelligence (AMI). His research program since ~2022 has focused on world models as the architectural path to human-level intelligence, explicitly arguing that autoregressive LLMs are a dead end for AGI.

Key Positions

• LLMs alone cannot reach human-level intelligence. The argument rests on the bandwidth gap: a child processes ~10^14 bytes of high-bandwidth sensory data in 4 years; LLMs consume the same byte volume as low-bandwidth text, which LeCun argues is insufficient for physical common sense.
• World models are the required architecture. Systems must predict the consequences of their actions and plan by optimization, not just forward propagation.
• JEPA is the right recipe. Prediction should happen in an abstract representation space, not pixel space. Generative video prediction is computationally wasteful and produces blurry averages.
• Open-source AI is essential for sovereignty. LeCun has been a consistent, public advocate for open-weight models as a counter to regulatory capture by proprietary AI corporations.
• "AI doomer" existential-risk narratives are a marketing strategy. LeCun explicitly dismisses existential-risk claims as a cult-like framing used by proprietary labs to entrench their position.

Key Contributions to World Models

• "A Path Towards Autonomous Machine Intelligence" (Jun 2022, v0.9.2) — the canonical position paper proposing JEPA, configurable predictive world models, hierarchical planning, intrinsic motivation, and SSL as the blueprint for AMI. Load-bearing primary source for everything that follows. (paper)
• V-JEPA 2 (2025) — co-author; the flagship JEPA world model, 1M+ hours of video pre-training, zero-shot robotic manipulation. (paper)
• Long-standing critique of next-token-prediction scaling as the sole path to AGI

Mentioned In

• World Models — central figure in the JEPA camp
• JEPA — originator
• Self-Supervised Learning — articulated the bandwidth argument
• System 2 Reasoning — articulated the objective-driven AI framework

Related Entities

• Meta FAIR — LeCun's research organization
• Google DeepMind — produces Genie 3, the strongest counter-demonstration to LeCun's pixel-prediction critique

Notable Quotes

"We're never going to get to human-level intelligence by just training on text... This is just not going to happen."

"AI really sucks compared to animals and humans... [it] can't do very basic tasks in the real world like clear up the dinner table."

"As a path towards human level intelligence, LLMs are a dead end."

"The only way to have any level of sovereignty in AI is to have access to open source... powerful open source AI systems."

Changelog

• 2026-04-22 — Initial page. Sourced primarily from V-JEPA 2 paper + user-supplied lecture notes.