Yang Likun's recent criticism of LLMs is quite bold; whether the world model approach can succeed is another matter, but at least it has opened a new path for PhD students.

In an unsupervised learning podcast, Yann LeCun puts forward a non-mainstream view on the limitations of large language models, explores possible paths for robotics, explains why he left Meta, and disagrees with Hinton and Bengio on the perspective for the Turing Award. He also predicts 2027, introduces the new company AMI and its bet on world models, compares OpenAI and Anthropic to Sun Microsystems, and suggests that PhD students stop researching LLMs, sharing insights into AI safety, the mechanisms behind breakthrough research, and the gains and losses of Meta FAIR.