Mark Zuckerberg called Yann LeCun one Sunday last November. Word had gotten out that LeCun, who built Meta’s Fundamental AI Research (FAIR) lab, was leaving to start his own venture. Zuckerberg urged him to stay. Good luck raising the money and good luck building a product the market would believe, LeCun recalls him saying. The following month, LeCun unveiled Advanced Machine Intelligence Labs. By March, it had raised $1.03 billion in seed funding—one of the largest rounds in history.
Much of the tech industry has converged on the idea that scaling large language models will eventually produce human-level intelligence. But LeCun, one of the field’s pioneers, believes they are chasing a dead end. Advanced Machine Intelligence is pursuing a fundamentally different approach to training, aiming instead to give AI an intuitive understanding of physical reality. So-called “world models,” he believes, will eventually make AI far more capable in areas like robotics, self-driving cars, and medicine.
LeCun was never one to follow the crowd. When he championed neural networks in the 1980s, the research community dismissed the idea for decades before it eventually became the foundation of the current AI boom, earning him a Turing Award in 2018. Being a contrarian takes a pinch of “confidence,” he says, and an ability to step back. “You just need to lift your nose from the trench you're digging, and look around.”