x
Thoughts on gradient hacking — AI Alignment Forum