Turing winner Rich Sutton: Deep learning is weak, needs total reworking
rich sutton launched oak lab to build reinforcement learning agents that learn continuously in real-time rather than relying on static pre-training.
turing award winner rich sutton founded oak lab with khurram javed to pursue ai agents capable of continuous, lifelong learning from their environment. sutton critiques current generative ai for merely imitating data without evaluating outputs, arguing the industry needs fundamental changes, not incremental tweaks. the company aims to build a trillion-parameter agent that plans in real-time using only 20 watts of energy. this signals a research shift back toward reinforcement learning and world modeling as alternatives to llm scaling.