研究：LLM 智能体蒸馏中思维监督并非必需，只在动作未披露思维信息时才关键

Agents, Think Only When it Helps

AI 导读

一项 arXiv 研究通过独立扰动教师模型的思维与动作来拆解 LLM 智能体蒸馏中各部分的作用，发现在多个主流智能体基准和不同模型规模上，改动或丢弃思维对学生模型的最终动作表现影响很小。但当训练中动作未披露对应思维信息时，思维监督才变得重要。该结果挑战了蒸馏完整教师轨迹的常见做法，指出可依据思维信息是否已在动作中披露，将思维监督作为提升效率的可选手段。

正文 · 原文

View PDF HTML (experimental)

Abstract:LLM agent distillation transfers agentic capabilities to student models by training on complete teacher trajectories that typically interleave thoughts and actions. We question how critical thought supervision is for agents' ultimate action performance. To answer this, we disentangle their impacts by independently perturbing teacher thoughts and actions during distillation. On widely adopted agent benchmarks across model scales, altering or discarding thoughts has little effect on student performance. We find, however, that thought supervision becomes important when information about the thought is not disclosed in the corresponding action during training. By identifying such discrepant cases, our findings challenge the common practice of distilling entire teacher trajectories for LLM agents, and offer a new insight that thought supervision can be a discernible choice for efficiency depending on whether the information of thought is disclosed in the action.

Subjects:	Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
Cite as:	arXiv:2611.05709 [cs.AI]
	(or arXiv:2611.05709v1 [cs.AI] for this version)
	https://doi.org/10.48550/arXiv.2611.05709 arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Jung-Eun Kim [view email]
[v1] Tue, 3 Nov 2026 22:35:21 UTC (70 KB)

来源：arXiv cs.AI · arxiv.org