LeCun: Next AI breakthrough lies in world models, not LLMs
Yann LeCun, founder and chairman of AMI Labs and a Turing Award winner, said in Seoul on Monday that AI's next breakthrough will not come from large language models but from systems that can understand and predict the physical world — so-called world models, according to South Korean outlet The Korea Herald. He said LLMs built for everyday scenarios are limited in understanding the real world: there are no robots that can complete tasks without any training and no Level 5 autonomous cars, while any teenager can learn to drive in a few hours. Arguing that machines find it easy to read language but hard to understand reality, he advocated building systems that learn the physical rules and causal relationships of the real world, predict the outcomes of actions and interact with their surroundings.
LeCun is a key inventor of convolutional neural networks and a prominent skeptic of the LLM-centric path, so his push for world models points to an alternative research direction worth tracking.