Rich Sutton, the 2024 Turing Award winner and co-founder of modern reinforcement learning, has launched the startup Oak Lab in Toronto with Khurram Javed. Both men previously worked at John Carmack’s AI company Keen Technologies. Sutton describes current deep learning methods as weak and inefficient, stating they require more than minor tweaks to provide a solid foundation for ambitious AI goals. He argues that generative AI excels at imitation but lacks the ability to evaluate its own outputs, rendering it incapable of real discovery.
The new company bets on reinforcement learning and the conviction that AI should learn from experience during operation rather than training once on static datasets. The long-term goal is an agent with a trillion parameters that learns and plans in real time while consuming only 20 watts of energy. This approach aims to move beyond current limitations by allowing systems to construct internal world models and handle variation independently.
* Focuses on continuous learning from the environment
* Requires significantly lower energy consumption
* Prioritises self-evaluation over static imitation
