Richard Sutton Tells Dwarkesh Patel That LLMs Are a Dead End — the Father of RL Versus the Entire Scaling Consensus

Dwarkesh Podcast·low signal

In a Dwarkesh Podcast episode, Sutton — 2024 Turing Award winner and the person who wrote The Bitter Lesson — argued that large language models are a dead end as a path to general intelligence. Coming from the author of the essay most often cited to justify scaling, this is the sharpest available insider dissent, and it lands alongside Yann LeCun's world-models bet at AMI Labs as a second Turing laureate defecting from the LLM paradigm. The counter-argument: Sutton has held roughly this position for years, so it's less a new signal than a persistently unresolved one.