arXiv cs.AI· Yuetian Du, Meng Wang, Yucheng Wang, Zhangzhe Chen, Wei Wang, Boyu Yang, Weiqi Zhai, Bing Zhao, Ming Kong, Zhan Zhou, Jie Liu, Qiang Zhu·· 3 天前AI 评分32
REVEAL：从前沿模型轨迹中逆向工程智能体隐变量
REVEAL: Reverse Engineering Agentic Latents from Frontier-Model Traces
AI 导读
研究者提出免训练框架 REVEAL，通过逆向工程思路从前沿 LLM 智能体的文本轨迹中提取"Agentic Latents"，为冻结的目标智能体生成局部激活控制方向。在 Terminal-Bench 2.1 和 Humanity's Last Exam with tools 上，该方法在三个目标模型上均取得高于未修改智能体的准确率，且无需目标模型训练或访问源模型内部。
正文 · 原文
Authors:Yuetian Du, Meng Wang, Yucheng Wang, Zhangzhe Chen, Wei Wang, Boyu Yang, Weiqi Zhai, Bing Zhao, Ming Kong, Zhan Zhou, Jie Liu, Qiang Zhu
Abstract:Frontier LLM agents perform complex interactive tasks, but their inference costs motivate transferring their experience to smaller open-source agents. Distillation and latent communication typically require target-model training or access to source-model internals. We study how textual frontier-model trajectories can provide reusable local controls for a frozen agent. Inspired by protocol reverse engineering, we introduce $\textit{Agentic Latents}$, comprising anchored states and dynamic profiles of feedback--response cycles elicited by replay in the target model. Our training-free framework, $\textbf{REVEAL}$ ($\textbf{REV}$erse $\textbf{E}$ngineering $\textbf{A}$gentic $\textbf{L}$atents), uses these profiles and trajectory outcomes to match and select local events from successful and failed rollouts of the same task and source model. Their response-transition contrasts yield a target-specific steering direction, applied before each response. Within-benchmark evaluations on Terminal-Bench 2.1 and Humanity's Last Exam with tools report higher accuracy than unmodified agents across three targets. Representation analyses and behavioral case studies support reusing frontier-model experience as local activation control.
| Subjects: | Artificial Intelligence (cs.AI); Computation and Language (cs.CL) |
| Cite as: | arXiv:2610.16288 [cs.AI] |
| (or arXiv:2610.16288v2 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.16288 arXiv-issued DOI via DataCite |
Submission history
From: Yuetian Du [view email]
[v1]
Mon, 12 Oct 2026 07:14:47 UTC (2,546 KB)
[v2]
Wed, 14 Oct 2026 02:36:15 UTC (2,546 KB)
来源：arXiv cs.AI · arxiv.org