Rich Sutton Interview: Critiques Synthetic Data, Aims to Build 20W Trillion-Param Agent
RichardSSutton · x · 2026-08-20
Rich Sutton, the father of reinforcement learning, discussed the state of AI with his partners.
Key points:
- The Bitter Lesson: The world is vastly more complex than any model; anything trained on human-curated data has a ceiling. General methods leveraging search and learning win in the long run.
- Continual Learning: Intelligence is defined by continual learning from experience, not one-off training.
- Against Synthetic Data: Sutton labels synthetic data a "big mistake" that leads to model collapse.
- Oak Lab Startup: Their new venture aims to build a 1 trillion parameter agent running on just 20 watts, equivalent to the human brain's power consumption.
Related event: RL Pioneer Sutton Calls Synthetic Data a Mistake(4 posts)→
More from AGI Musings
- "An LLM doesn't understand anything" gets the perfect parody reversal: neither do you — VoidStateKate · 2026-10-16
- Nina Schick: Europe's Policy Circle Still Doesn't Believe FSD, Orbital Compute Are Real — NinaDSchick · 2026-10-16
- Meta's Superintelligence Lab Runs 200+ Direct Reports—A Lesson for Flat Multi-Agent Systems — peterjliu · 2026-10-16
- Andy Masley lays out his version of effective altruism, pluralism over utilitarian math — RobbWiller · 2026-10-16
- We're forgetting how much of a miracle GPT-3.5-level intelligence was, and it may cost us — EigenGender · 2026-10-16
- Ex-Anthropic's McDonagh Builds Compute Intelligence Agency to Study Frontier AI Costs — McDonaghMatthew · 2026-10-16