Yann LeCun urges AI to move beyond essays and tackle real-world tasks

In a recent X exchange, Yann LeCun, the ex-Chief AI Scientist at Meta, expressed a desire for large language models to evolve from merely writing essays to performing tangible actions such as cleaning a bedroom. He suggested that solving this gap would require new architectures and methods that enable rapid learning of physical tasks, comparable to human and animal efficiency. LeCun also challenged former Meta colleague Jitendra Malik’s terminology around vision-language models and world models, arguing that understanding dynamics for control differs from video generation.

The debate underscores a broader discussion on the future direction of AI research. LeCun, known for pioneering convolutional neural networks and co-receiving the 2018 Turing Award, confirmed he will leave Meta after a twelve-year tenure, during which he led FAIR and served as chief scientist. He continues his academic role at New York University.