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