1 year agoWhat Can We Do About Reward Hacking?: Concrete Problems in AI Safety Part 4Robert Miles Archive Channel