PARIS, July 5, 2026 – The man once hailed as a founding father of modern artificial intelligence has a blunt message for the industry: today's chatbots are a dead end. Yann LeCun, who left Meta in 2025 after a decade as chief AI scientist, now warns that systems like ChatGPT, Claude, and Gemini lack the fundamental intelligence needed to navigate the real world. "We don't have robots that are nearly as good at understanding the physical world as a rat," LeCun told reporters on the sidelines of VivaTech in Paris this week.
LeCun's new venture, Advanced Machine Intelligence Labs (AMI Labs), is racing to build a radically different kind of AI. The core problem, he argues, is that Large Language Models (LLMs) are brilliant at regurgitating data but cannot reason about physical reality. To illustrate, he held a pen upright. "Even a toddler knows it will fall, but no human tries to predict which direction because it's impossible," LeCun explained. An LLM, he said, would waste resources generating a statistical guess—almost certainly wrong—rather than understanding the physics of the situation.
The stakes are enormous. In a stunning vote of confidence from Silicon Valley heavyweights, AMI Labs announced earlier this year it had raised over $1 billion in seed funding—one of the largest early-stage rounds in European history. Key investors include Nvidia, the dominant maker of AI chips, and the investment fund managing Amazon founder Jeff Bezos's private wealth. The cash injection signals a growing consensus that the current generation of AI has hit a glass ceiling, particularly in applications requiring physical interaction, such as household robotics or autonomous navigation.
LeCun's answer is a system called Joint Embedding Predictive Architecture (JEPA). Unlike LLMs that predict the next word in a sentence, JEPA is designed to create abstract models of the real world, allowing it to assess likely outcomes without needing to predict every detail. "They [LLMs] are not a path towards human-level intelligence, or even animal-like intelligence, because they cannot deal with real-world data," LeCun said. "They're not particularly smart. They just regurgitate." As the AI race enters a new phase, the question is no longer about scaling up data, but about building systems that actually understand the world they inhabit.