arXiv · cs.AI · Juan-Pablo Rivera, Ari Spiesberger, Juan J Vazquez, Arun Jose, Axel Ahlqvist, Adeline Kassler, Meru Gopalan, Sandy Tanwisuth, John Hughes · 10 小时前模型评分49
研究：微调监控模型可减少 Llama-3.3-70B 与 Qwen2.5-72B 的未发现合谋行为
Model Organisms of Collusion: Corrective Transfer in Untrusted Monitors
导读
研究者在 Llama-3.3-70B 和 Qwen2.5-72B 上安装五组策略-监控行为对，用 300 条已发现合谋的诚实评审微调监控模型，再在未见提示格式下测量其余四组合谋。
分类置信度
中 · 待人工复核
- 只有单一媒体来源，需要人工复核。
来源级别：媒体与个人
事件类型：研究发布
证据摘录
- “We ask whether this training also removes collusion that the control team has not yet found.”
- “We fine-tune Llama-3.3-70B and Qwen2.5-72B to install five policy-monitor behaviour pairs, correct the monitor on 300 honest reviews of one caught collusion, and measure collusion on the other four in unseen prompt formats.”
- “Under the tested training configuration, methods with similar caught-task accuracy differ in held-out transfer, from 58% to 83% average reduction, so the caught behaviour alone does not reveal how far a correction reaches.”
来源：arXiv · cs.AI · arxiv.org