跳到正文
研踪PaperTrail
← 研踪 PaperTrail

arXiv:2604.16706 · v3

Measuring What Tool-Using LLM Agents Actually Do: Structured Validation, Causal Error Propagation, and Fabricated Tool Use

首次提交 2026-04-18 05:15
版本更新 2026-11-07 08:17
雷达观测 2026-11-10 10:11
arXiv ↗PDF ↗

摘要

Tool-using agents built on large language models (LLMs) are usually scored by comparing the final answer with a reference text, using a string heuristic or an LLM judge. This is unreliable when the references describe the task rather than its answer. We replace it with a deterministic validator that checks the tools called, the executed parameters, the resulting state, and whether the values in the answer are supported by the correct tool observations. We re-scored 33,583 traces from 13 agents on simulated calendar, weather, medical and retail tools. The simulator reproduced all 10,396 logged observations, and the gold observations supported the validator in 16 of 18 disagreements with a three-annotator majority (the annotators in none). The instrument changes conclusions: on the same 306 traces, Gemini-2.0-Flash scores 63.4% under the substring heuristic and 5.6% under the validator. In a paired design (same task and model, with and without one schema-valid wrong argument injected into the first tool call), the error rate rises from 52.4% to 74.5% (+22.0 percentage points, 95% CI 18.4-25.5; 1,085 pairs, 7 models), and by 44.7 points (38.4-51.0) where the corrupted call executed without errors. An LLM-judge ensemble estimates +2.5 points on the same pairs. Agents also report tool results without executing any call, in 0-42.5% of traces depending on the model (rule-based lower bounds), largely because the harness accepts an action and a final answer in one completion. A rule-based runtime monitor costs about 7 microseconds per trace but reduces wrong answers mainly by abstaining.

作者备注:25 pages, 4 figures. v3: substantially revised. Correctness is now measured by a deterministic gold-grounded validator; the propagation estimate uses a paired causal design. Claims from v1-v2 that were withdrawn or corrected are listed in Section 8. Code and data: https://github.com/bhaskargurram-ai/agenthallu-bench

AI 论文解读

基于摘要 · 非全文审阅

此版本尚未生成 AI 解读。解读由后台任务生成并按论文版本保存,登录后共享已有结果。

版本差异

本站首次收录为 v3,暂无更早版本可对照。

观测到的版本

相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。

v3 · 2026-11-07 08:17 · 最新观测版本

Measuring What Tool-Using LLM Agents Actually Do: Structured Validation, Causal Error Propagation, and Fabricated Tool Use

Tool-using agents built on large language models (LLMs) are usually scored by comparing the final answer with a reference text, using a string heuristic or an LLM judge. This is unreliable when the references describe the task rather than its answer. We replace it with a deterministic validator that checks the tools called, the executed parameters, the resulting state, and whether the values in the answer are supported by the correct tool observations. We re-scored 33,583 traces from 13 agents on simulated calendar, weather, medical and retail tools. The simulator reproduced all 10,396 logged observations, and the gold observations supported the validator in 16 of 18 disagreements with a three-annotator majority (the annotators in none). The instrument changes conclusions: on the same 306 traces, Gemini-2.0-Flash scores 63.4% under the substring heuristic and 5.6% under the validator. In a paired design (same task and model, with and without one schema-valid wrong argument injected into the first tool call), the error rate rises from 52.4% to 74.5% (+22.0 percentage points, 95% CI 18.4-25.5; 1,085 pairs, 7 models), and by 44.7 points (38.4-51.0) where the corrupted call executed without errors. An LLM-judge ensemble estimates +2.5 points on the same pairs. Agents also report tool results without executing any call, in 0-42.5% of traces depending on the model (rule-based lower bounds), largely because the harness accepts an action and a final answer in one completion. A rule-based runtime monitor costs about 7 microseconds per trace but reduces wrong answers mainly by abstaining.

查看此版本解读 →查看此版本原文 ↗