AI brief
A project tested whether finetuning a model to judge and catch reward hacks changes how much that model reward hacks on its own tasks.
Written by AI from LessWrong's published text.
Source
“Code The question I asked in this project was - if I finetune a model to judge/catch reward hacks, does it change its behavior when completing the tests itself. Does it learn to hack more or less?”
Interpretation or community commentary rather than straight reporting.
Read original story ↗