RESEARCH · RESEARCH · #3266
Richard Sutton: compute alone won't produce intelligence; calls for continuous learning
Turing Award winner Richard Sutton published a post arguing that raw compute and current LLM scaling are insufficient to produce intelligence, calling synthetic data a “big mistake” and urging agents with world models that continuously learn from experience; his remarks provoked discussion among researchers and industry leaders including Jerry Tworek, Jayden Teoh, Sam Altman and Dario Amodei.
KEY POINTS
- Turing Award winner Richard Sutton published a post arguing that raw compute and current LLM scaling are insufficient to produce intelligence, calling synthetic data a “big mistake” and urging agents with world models that continuously learn from experience; his remarks provoked discussion among researchers and industry leaders including Jerry Tworek, Jayden Teoh, Sam Altman and Dario Amodei.
- Sutton is a leading theorist whose critique challenges prevailing bets on LLM scaling and could shift research priorities toward continual-learning algorithms and world models.
- The inventor of the bitter lesson has a bitter lesson for LLM scaling enthusiasts
WHY IT MATTERS
Sutton is a leading theorist whose critique challenges prevailing bets on LLM scaling and could shift research priorities toward continual-learning algorithms and world models.