Yann LeCun says industry’s focus on L.L.M.s risks a dead end

Yann LeCun, a veteran A.I. researcher and former chief A.I. scientist at Meta, warned in an interview from his home in Paris that many technology companies are on the wrong path to creating intelligent machines and are “marching into a dead end.”

Dr. LeCun, a Turing Award winner whose early work helped develop neural networks, said large language models, or L.L.M.s, ‘‘can get only so powerful’’ and that the industry has been ‘‘L.L.M.-pilled.’’ He criticized a Silicon Valley ‘‘herd effect’’ that he said leaves little room for other approaches and suggested that ‘‘more creative Chinese companies’’ could make advances first.

While at Meta, Dr. LeCun pushed for open research and worked on systems that try to predict the outcomes of actions so A.I. can plan ahead. He said L.L.M.s ‘‘do not plan ahead’’ and ‘‘absolutely cannot do that,’’ and he will pursue prediction-and-planning work at his new start-up, Advanced Machine Intelligence Labs, which he formed after leaving Meta in November.

Others cited in the discussion said language models have continued to improve and are useful in areas like coding, while critics note Dr. LeCun’s newer methods remain unproven. What comes next remains uncertain: Dr. LeCun is betting on alternative approaches at his start-up, and he warned that open sharing of work is crucial if the field is to progress faster.

Key Topics

Tech, Yann Lecun, Ami Labs, Meta, Large Language Models, Neural Networks