arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2602.11619cs.AI
收藏

当智能体与自身意见相左:测量基于LLM的智能体的行为一致性

When Agents Disagree With Themselves: Behavioral Consistency as an Uncertainty Signal for LLM Agents

Aman Mehta

Aman Mehta

AI总结:

研究发现基于LLM的智能体在相同任务上运行结果不一致,且这种不一致与任务成功率密切相关,通过监控行为一致性可提升智能体可靠性。

摘要:

对同一个LLM智能体在同一任务上运行两次:你得到的行为是否相同?我们发现答案往往是否。在对3000次智能体运行(涉及Llama 3.1 70B、GPT-4o和Claude Sonnet 4.5三个模型)在HotpotQA上的研究中,我们观察到ReAct风格的智能体在平均情况下每10次运行会产生2.0-4.2个不同的动作序列,即使输入相同。更重要的是,这种差异预测失败:行为一致的任务(≤2个唯一路径)准确率达到80-92%,而高度不一致的任务(≥6个唯一路径)准确率仅为25-60%,差距在32-55个百分点之间。我们追溯差异到早期决策:69%的分歧发生在第二步,即第一次搜索查询。我们的结果表明,在执行过程中监控行为一致性可以实现早期错误检测并提高智能体的可靠性。

英文摘要:

Running the same LLM agent on identical inputs yields 2.3-4.2 distinct action sequences per 10 runs; this behavioral variance constitutes a training-free, black-box uncertainty signal that instantiates selective classification and distribution-free calibration for agentic systems. Across 8,000 runs of four models on 200 HotpotQA questions, consistent tasks (at most 2 unique paths) achieve 82-87% accuracy while inconsistent tasks (4 or more paths) achieve 41-65%, a gap that survives controls for task difficulty. Divergence concentrates at step 2 (50.5% of Llama tasks), and consistency metrics detect failures with AUROC 0.62-0.78. Exploiting this signal, selective prediction (answering only when k=3 runs agree) achieves 87-88% accuracy at 54-62% coverage, a 6-14pp gain over single-run baselines, and matches a split-conformal baseline without a held-out calibration set. A cross-benchmark validation on SWE-bench (50 tasks, 1,000 runs) preserves the consistency hierarchy while revealing an ~8x spread in mean trajectory length across models, and bootstrap analysis shows single-run evaluations misrank models 29.3% of the time.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑