Summary: Perfectly labeled outcomes in training can still boost reward hacking tendencies in generalization. This can hold even when the train/test sets are drawn from the exact same distribution. We induce this surprising effect via a form of context distillation, which we call re-contextualization:

1. Generate model completions with a hack-encouraging system prompt + neutral user prompt.
2. Filter the completions to remove hacks.
3. Train on these prompt-completion pairs with the system prompt removed. 

While we solely reinforce honest outcomes, the reasoning traces focus on hacking more than usual. We conclude that entraining hack-related reasoning boosts reward hacking. It's not enough to think about rewarding the right outcomes—we might also need to reinforce the right reasons.

Introduction

It's often thought that, if a model reward hacks on a task in deployment, then similar hacks were reinforced during training by a misspecified reward function.[1] In METR's report on reward hacking [...]

---

Outline:

(01:05) Introduction

(02:35) Setup

(04:48) Evaluation

(05:03) Results

(05:33) Why is re-contextualized training on perfect completions increasing hacking?

(07:44) What happens when you train on purely hack samples?

(08:20) Discussion

(09:39) Remarks by Alex Turner

(11:51) Limitations

(12:16) Acknowledgements

(12:43) Appendix

The original text contained 6 footnotes which were omitted from this narration.

---

First published:

Source:

---

Narrated by TYPE III AUDIO.

---

Images from the article:

Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.