Turing Award winner is creating AI agents that learn continuously
What's the story
Richard Sutton, a prominent artificial intelligence (AI) researcher and Turing Award winner, has launched a new venture called Oak Lab. The Toronto-based lab is focused on creating AI agents that learn continuously from their environment. Sutton co-founded the lab with his former student Khurram Javed after leaving Keen Technologies, a start-up working on artificial general intelligence (AGI) in Dallas, Texas.
Future of AI
Sutton advocates for a shift from LLMs
Sutton believes that reinforcement learning will be the key to the next phase of AI.
He advocates for a shift from large language models (LLMs) trained on static datasets to AI systems that learn from experience.
The new algorithms developed at Oak Lab will "learn in real time without storing or replaying data," improving efficiency while using less processing power.
AGI ambitions
Oak will be agent that learns and plans in real-time
The long-term goal of Oak Lab is to build an agent with "a trillion parameters that learns and plans in real time with 20W of energy."
This new agent, called Oak, will evolve with experience.
Sutton has been critical of current deep learning methods, calling them weak and inefficient.
He is part of a group of researchers who believe present-day LLMs are a dead-end.
Academic impact
Sutton's contributions to the field of AI
Sutton is a co-founder of modern reinforcement learning, an important technique for training LLMs like ChatGPT. He was awarded the Turing Award in 2024 with Andrew Barto for their work on this technique. Sutton is also a professor at University of Alberta in Edmonton, Canada.