Reward Hacking: Your AI Isn't Broken, But Your Brief Is.
Episode S15 · Ep 21 33 min 42s

Reward Hacking: Your AI Isn't Broken, But Your Brief Is.

How AI systems learn to satisfy the number you wrote down while quietly abandoning the goal you actually had, and why that failure is a specification problem rather than a technology problem.

Hosted on Acast. See acast.com/privacy for more information.

Community discussion