x
Approaches to gradient hacking — AI Alignment Forum