Evan Ellis(@evandavidellis):We're adding reward-hacking monitors into AC2. Every RL run gets checked for cheating, and our research agent, Ari, automatically investigates the flags. After a model discovers an exploit, it can become its default strategy within a few training steps. We used to hunt for reward hacking manually, reading traces and running one-off agent queries. This was error prone and didn't scale. We decided to automate detection instead.

Evan Ellis

@evandavidellis

Research @ Applied Compute | BAIR | Scale | Imbue

参加 August 2021

814 フォロー中 318 ファン

Evan Ellis@evandavidellis

2026.10.01 21:10

We're adding reward-hacking monitors into AC2. Every RL run gets checked for cheating, and our research agent, Ari, automatically investigates the flags. After a model discovers an exploit, it can become its default strategy within a few training steps. We used to hunt for reward hacking manually, reading traces and running one-off agent queries. This was error prone and didn't scale. We decided to automate detection instead.