Sutton Launches Oak Lab to Advance OaK
Mindrust · reddit · 2026-07-14
Richard Sutton has launched Oak Lab with the goal of advancing his proposed OaK (Options and Knowledge) architecture.
The core idea is to stop having agents rely primarily on pre-training data. Instead, they should form concepts and skills by continuously learning from experience, and then use those skills for planning and improvement. Sutton believes the ideal form is an agent capable of real-time learning and planning, consuming only 20W of power, and scaling to a trillion parameters.
Lab collaborator Khurram Javed stated that the team will continue to share progress publicly over the next few years and hopes to build a complete prototype of the OaK architecture. This prototype will more closely resemble a "baby learning in its first year" rather than current mainstream AI systems.
Related event: Turing Award Winner Sutton Leaves Keen to Found OakLab(10 posts)→
More from AGI Musings
- DeepMind alignment researcher signs open letter urging coordinated AI slowdown — vkrakovna · 2026-09-11
- WIRED: recursive self-improvement and rogue agent swarms spook AI researchers — nordicinst · 2026-09-11
- People Neglect Human Agency Both Ways: Exaggerated Doom and Complacent Optimism — jankulveit · 2026-09-11
- Garrison Lovely's 'Obsolete' on AI Replacing Labor Lands September 2026 with Heavyweight Blurbs — GarrisonLovely · 2026-09-11
- AI Doom Skeptics Hit Back: EA-Driven Apocalypse Talk Doesn't Reflect Most Top-Tier Researchers — GarrisonLovely · 2026-09-11
- Over 1,000 AI Policy Initiatives Launched in 70+ Countries, but the Governance Gap Widens — CurieuxExplorer · 2026-09-11