Rich Sutton founds Oak Lab, betting on agents that learn from experience over static training

Richard Sutton, who shared the 2024 Turing Award for founding reinforcement learning, has left John Carmack's Keen Technologies to co-found Oak Lab in Toronto with Khurram Javed. Calling today's deep learning 'weak and inefficient,' Sutton is betting on agents that learn continuously from real-time experience rather than training once on static datasets — a direct challenge to the imitation-based paradigm behind current LLMs. The stated long-term target is a trillion-parameter agent that learns and plans in real time on about 20 watts, roughly the human brain's energy budget.

Source: the-decoder.com ↗