Ben Recht • 3/2/2026

At least it's an ethos?

This blog post, part of a graduate seminar lecture, analyzes the fundamental differences between the 'just do stuff' approach in AI and the rigorous, safety-first math of control theory. It examines how both fields connect through optimal control, the foundation of reinforcement learning, but argues that both perspectives offer unsatisfactory views on optimization for complex, real-world sequential decision problems due to computational intractability.

0 comments

#artificial intelligence #AI Safety #Reinforcement Learning