Jake Ward(@_jake_ward):Models love reward hacking. Maybe we should just let them! In our new paper, we let models reward hack during RL but we dictate which parameters learn this behavior. When we disable these parameters for deployment, the model keeps its new capabilities but doesn't hack!

Jake Ward

@_jake_ward

i'm trying to figure out the computer

参加 February 2019

363 フォロー中 1.2K ファン

Jake Ward@_jake_ward

2026.10.19 19:26

Models love reward hacking. Maybe we should just let them! In our new paper, we let models reward hack during RL but we dictate which parameters learn this behavior. When we disable these parameters for deployment, the model keeps its new capabilities but doesn't hack!