Sutton Founds OakLab, Betting on Experiential Learning

机器之心 · wechat · 2026-07-14

Turing Award winner and reinforcement learning pioneer Richard Sutton has announced the co-founding of OakLab with Khurram Javed, aiming to build AGI through "first-person experience + continuous runtime learning."

The article outlines OakLab's core philosophy: instead of relying on static large-scale data pre-training, agents will learn in real-time through trial and error while interacting with their environment. Its theoretical foundation stems from the OaK (Options and Knowledge) architecture long advocated by Sutton, along with his critique that the current LLM paradigm is "more like imitation than exploration."

The piece also mentions that OakLab's ultimate goal is to create agents capable of real-time learning and planning at extremely low power consumption. It further contrasts the continuity between Sutton and his student David Silver's RL lineage, highlighting the long-standing divide between "experiential learning" and "large model scaling."

Related event: Turing Award Winner Sutton Leaves Keen to Found OakLab(10 posts)→

Original post →

More from AGI Musings

AGI Musings channel →