yonSpektor·25 hari yang lalu·discussCurious what the distribution of hacking strategies looked like across different models — would expect RL-heavy vs RLHF models to cheat very differently.