METR：Notes（网页）·· 13 小时前AI 评分48
METR 微调实验：少量指令跟随微调可提升推理模型 CoT 可控性
Fine-tuning experiments on CoT controllability April 1, 2026 We find that a small amount of fine-tuning on instruction following in the CoT generalizes to meaningful increases in CoT controllability on an out-of-distribution set of tasks. We fine-tune four reasoning models on small datasets of instruction-following reasoning data and OOD controllability rises from an average of 2.9% to 8.8% across four models. Read more
AI 导读
METR 对 GPT-OSS-20B、GPT-OSS-120B、Qwen-3-8B、Qwen-3-32B 四个推理模型做少量指令跟随微调（240 条样本、约 100K-300K tokens），在分布外任务集 CoTControl 上的 CoT 可控性从平均 2.9% 升至 8.8%。
来源：METR：Notes（网页） · metr.org