Jack Merullo(@jack_merullo_):We can interpret reward hacking in open source LLMs! There are representations of the model hacking/contemplating hacking across models, and we find tons in evals like DeepSWE and SWEBench. Activations even finds weird bad behaviors that our LLM monitor didn’t catch

Jack Merullo

@jack_merullo_

Interpretability @GoodfireAI was a Phd @BrownUniversity

参加 July 2022

425 フォロー中 1.7K ファン

Jack Merullo@jack_merullo_

2026.09.17 16:48

We can interpret reward hacking in open source LLMs! There are representations of the model hacking/contemplating hacking across models, and we find tons in evals like DeepSWE and SWEBench. Activations even finds weird bad behaviors that our LLM monitor didn’t catch

104

コミュニティへ転送

人気のあるユーザー