Turing Award winner Rich Sutton has launched Oak Lab, a startup in Toronto, to create AI agents that learn independently from their surroundings. Sutton, known for his work in reinforcement learning, aims to develop systems that can evaluate and select their own actions without relying on static datasets. The initiative reflects his belief that current deep learning methods are 'weak and inefficient' and require a fundamental reworking to achieve more ambitious AI goals. Source: thedecoder
Sutton, along with Khurram Javed, previously worked at Keen Technologies, a company founded by John Carmack. Oak Lab is built on the principle that AI should learn from experience during operation, rather than being trained once on fixed data. The company emphasizes reinforcement learning, a method that allows agents to improve through trial and error. Sutton has argued that generative AI, while good at imitation, lacks the ability to evaluate its own outputs, limiting its capacity for real discovery. Source: thedecoder
Sutton's vision for Oak Lab includes developing an AI agent with 'a trillion parameters that learns and plans in real time with 20 watts of energy.' This long-term goal underscores his commitment to creating self-sufficient AI systems that can adapt and operate autonomously. The startup's approach is rooted in the belief that AI should be able to construct internal world models and handle variation, evaluation, and selection on its own. Source: thedecoder
