Le scientifique Français qui va détruire l’IA Américaine

Yann LeCun’s Billion-Dollar Bet Against Large Language Models

This video follows Yann LeCun, a Turing Award winner and one of the founding fathers of modern AI, as he challenges the industry’s trillion-dollar race to scale large language models (LLMs). LeCun argues that LLMs are a dead end for achieving human-level intelligence and proposes an entirely different path: world models that understand physics and causality. After a leadership clash at Meta, he left to build his own startup, raising $1 billion to prove his vision.

The Case Against LLMs

LeCun contends that LLMs, despite their impressive fluency, are fundamentally dumb. They master language but lack any grasp of the physical world. He gives a simple example: ask an LLM whether to walk to a car wash 100 meters away, and it says yes—ignoring that you need the car to drive there. Such “hallucinations” are not bugs, he insists, but symptoms of a flawed architecture.

• LLMs are trained only to predict the next word; they have no intuitive physics, memory, or planning ability.
• He states bluntly: “The best AI systems are not as intelligent as a cat. A cat or a 4‑year‑old instinctively understands gravity, friction, and how objects fall.”
• Scaling up LLMs with more data and compute, LeCun believes, will never yield an artificial general intelligence (AGI).

World Models: An Alternative

LeCun’s solution is world models—AI systems that learn the world the way a baby does: by observing, interacting, and building an internal representation of reality. His research, now years old, recently produced a proof‑of‑concept that runs on a single GPU (versus the tens of thousands needed for LLMs). The model learns to ignore irrelevant details (e.g., the sky in driving scenes) and focuses on causal relationships.

• Key capabilities: physical intuition, long‑term memory, planning, and reasoning.
• Goal: enable truly autonomous robots and self‑driving cars, where an error can be life‑threatening.
• LeCun has been working on this for 15 years; the breakthrough came only months ago.

The Clash with Meta and a New Chapter

LeCun was Chief AI Scientist at Meta (Facebook) for over a decade, personally recruited by Mark Zuckerberg. But in June 2025, Zuckerberg invested billions in Scale AI and put its founder, Alexander Wang (age 28), in charge of Meta’s AI strategy. Wang shifted the company toward short‑term LLM competition, sidelining fundamental research. LeCun publicly disagreed: “You don’t tell a researcher what to do.” Unable to stay, he left Meta and launched Amabs, raising $1 billion (double his original target) from investors including Xavier Niel and Jeff Bezos. He took many top Meta researchers with him.

Key Takeaway

Yann LeCun is betting that the entire AI industry is wrong. While giants pour billions into scaling LLMs, he is building world models that understand the physical world. His track record (pioneering neural networks when they were dismissed) gives weight to his claim. If he succeeds, AI could move from being a sophisticated parrot to a truly intelligent system—and the next revolution in robotics and autonomous driving will begin.