The inventor of the bitter lesson has a bitter lesson for LLM scaling enthusiasts

Turing Award winner Richard Sutton argues in a new post that raw compute and current LLM scaling alone are insufficient to achieve true machine intelligence. Sutton contends that human expertise acts as a bottleneck and calls instead for continuous learning agents equipped with world models. While some AI researchers agree that algorithmic innovation remains crucial alongside scaling, industry leaders such as OpenAI CEO Sam Altman defend continued LLM scaling. Open article →