Machine-learning pioneer Yann LeCun on why “a new revolution in AI” is coming

The launch of R1, an AI model by the Chinese startup DeepSeek, recently sent shockwaves through the technology world. R1 is a “reasoning” model—the most cutting-edge type of large language model (LLM)—and it performs about as well as the best-in-class Western models but for a fraction of the training cost. Like other LLMs, though, it still lacks many of the skills and types of intelligence that human brains achieve. For one, “reasoning” models still have a very limited understanding of the physical world in which they exist.

Our guest today wants to get beyond these hurdles. Yann LeCun, chief AI scientist at Meta and a professor at New York University, thinks LLMs are not the answer if we want truly useful personal assistants, humanoid robots and driverless cars in the future. For machine intelligence to get more interactive with the real world, he is fundamentally rethinking how AI models are built and trained.