同一输入下大模型代理行为不一致，可作无训练不确定性信号。
When Agents Disagree With Themselves: Behavioral Consistency as an Uncertainty Signal for LLM Agents
- 用多次运行结果一致性判断代理决策可信度
- 一致任务准确率82%-87%，不一致仅41%-65%
- 适合需要可靠推理的高风险场景
在相同输入下，运行同一LLM代理在10次中产生2.3至4.2种不同的动作序列；这种行为差异构成一种无需训练、黑箱的不确定性信号，可用于实现选择性分类与分布无关校准。在200个HotpotQA问题上对四个模型进行8,000次运行，一致任务（最多2条路径）准确率达82%-87%，而不一致任务（4条及以上路径）准确率为41%-65%，该差距在控制任务难度后仍显著存在。分歧主要集中在第2步（Llama任务中占50.5%），一致性指标可检测失败，AUROC为0.62-0.78。利用该信号，当3次运行一致时才回答（选择性预测），准确率达87%-88%，覆盖率为54%-62%，比单次运行基线提升6-14个百分点，且无需保留校准集即可媲美分裂-共形基准。在SWE-bench（50任务，1,000次运行）上的跨基准验证保持了路径一致性层级，同时揭示各模型平均轨迹长度相差约8倍；自助分析显示单次评估会错误排序模型29.3%的时间。
原文摘要 · Abstract (English)
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.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。