LLMs “are a dead end”: AI pioneer LeCun launches startup for ‘world models’

Yann LeCun, a “godfather of AI” and Turing Award winner, is reportedly considering leaving Meta to start a company focused on “world models”—a technology he believes is the true path to advanced AI.

He has been increasingly sidelined at Meta because he publicly critiques the company’s focus (led by Mark Zuckerberg) on Large Language Models (LLMs) like Llama, calling them a “dead end.” LeCun advocates for world models, which learn from visual information like a child, over LLMs, which are predictive models based on text.

LeCun, who has a history of pioneering work in machine learning (including at Bell Labs), stepped back from leading Meta’s AI division (FAIR) in 2018 to pursue long-term research. His division has recently seen job cuts and a shift in focus toward rapid product development, further diverging from LeCun’s vision.

TCE