Yann LeCun has spent four decades shaping artificial intelligence, from early neural network research to leading Meta’s AI strategy. Now, after leaving the company, he is publicly challenging the direction Silicon Valley has taken. His core claim is direct. Large language models, despite massive investment and rapid gains, cannot deliver human-level intelligence and will eventually hit hard limits, says The New York Times.
LeCun’s career lends weight to the critique. As a young researcher, he championed neural networks when few believed in them. That work later enabled breakthroughs in handwriting recognition, computer vision, speech systems, and the technologies behind today’s AI boom. For more than a decade at Meta, he pushed not only technical progress but also the idea that research should remain open, published, and shared.
He argues that the industry’s current fixation on scaling language models reflects groupthink. Companies are pouring hundreds of billions into systems that lack core capabilities such as planning, grounded understanding, and real-world modeling. Trained largely on text, these models predict words well but struggle with causality, long-term reasoning, and reliable decision making. Errors compound as tasks grow more complex.
LeCun believes progress requires a different architecture, one that can model the world, anticipate outcomes, and plan actions. His new startup, AMI Labs, is built around that vision. The work extends ideas he pursued inside Meta, focused on systems that learn by predicting the consequences of their behavior rather than imitating language patterns.
The debate is not settled. Researchers tracking modern models point to continued gains in reasoning, coding, and science tasks. Others agree with LeCun that usefulness does not equal true intelligence and that current methods may plateau.
Beyond technology, he raises a strategic concern. As U.S. companies pull back from open research in pursuit of competitive advantage, he warns that more open ecosystems, including in China, could outpace them. In his view, openness accelerates progress and reduces concentrated control. The AI race, he argues, will be won by better ideas, not by louder hype.