AI Pioneer Yann LeCun: Next Breakthrough Not in Language Models

AI pioneer and AMI Labs founder Yann LeCun announced that the next major breakthrough in artificial intelligence will not come from large language models (LLMs), but from systems that understand and predict the physical world. LeCun believes these so-called "world models" will drive AI into its next phase.

LeCun highlighted the limitations of current language models in comprehending the real world, despite their proficiency in predicting words based on vast datasets. He cited examples such as the inability of robots to perform tasks without extensive training or the absence of L5-level autonomous vehicles, contrasting this with human drivers who learn to operate vehicles within hours of practice.

"This indicates a huge gap between understanding language and understanding the real world," LeCun stated. He explained that language models learn from patterns in massive training data to predict the next most likely word, generating text. While effective for predictable tasks like programming and mathematics, most real-world information is far less predictable.

LeCun advocates for the development of "world models" that would enable AI to learn the physical laws and causal relationships governing reality. These systems should be able to predict the outcomes of actions and interact with their environment. LeCun, a co-inventor of convolutional neural networks (CNN) and a recipient of the 2018 Turing Award, is considered one of the foremost figures in AI.