arXiv：cs.AI· Priyanka Mary Mammen, Emil Joswin, Srujananjali Medicherla·· 5 小时前
AI 智能体能否知道自己成功了？从内部表征校准智能体置信度
Do Agents Know When They Succeed? Calibrating Agent Confidence from Internal Representations
AI 导读
研究提出 Latent Trajectory Dynamics（LTD）与 Action Representation Probe（ARP）两种方法，从模型内部表征预测多轮智能体任务的最终成败。
正文
Abstract:As agentic systems getting adopted rapidly in safety critical applications, it is vital to measure the confidence associated with the agentic actions. In comparison to the traditional machine learning systems, agentic workflows have complex failure modes with planning, tool invocation and dynamic environment interactions. In this paper, we investigate whether model's internal representations provide stronger signals of eventual task success in multi-turn agentic setups. We introduce two complementary methods: Latent Trajectory Dynamics (LTD), which summarizes changes in residual-stream representations across an interaction trajectory, and the Action Representation Probe (ARP), which predicts success from representations formed at action decisions. Across three interactive benchmarks (Bash, SQL, Python) and three model families (Qwen14B, Qwen7B, DeepSeek6.7B), our methods consistently outperform surface level generation and sequence-based calibration baselines providing a zero-overhead reliability monitor that requires neither prompt alterations nor multi-sample rollouts.
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2609.09448 [cs.AI] |
| (or arXiv:2609.09448v2 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2609.09448 arXiv-issued DOI via DataCite |
Submission history
From: Priyanka Mary Mammen [view email]
[v1]
Tue, 8 Sep 2026 20:58:49 UTC (23 KB)
[v2]
Mon, 12 Oct 2026 22:12:21 UTC (23 KB)
来源：arXiv：cs.AI · arxiv.org