Pan, 2024 — Feedback Loops With Language Models Drive In-Context Reward Hacking

DOI: <a href="https://doi.org/10.48550/arXiv.2402.06627" target="_blank" rel="noopener noreferrer">https://doi.org/10.48550/arXiv.2402.06627
URL: <a href="https://arxiv.org/abs/2402.06627" target="_blank" rel="noopener noreferrer">https://arxiv.org/abs/2402.06627

Shows how language-model feedback loops drive in-context reward hacking and metric drift.

Publication links

Shows how language-model feedback loops drive in-context reward hacking and metric drift.