bruce comments on Infinite Rewards, Finite Safety: New Models for AI Motivation Without Infinite Goals