arXiv：cs.LG· Wei Shi, Ziheng Peng, Sihang Li, Xiting Wang, Xiang Wang, Mengnan Du, Na Zou·· 7 小时前AI 评分47
LLM 智能体内在过度调用偏差的诊断：When2Call 基准与 SAE 机制分析
To Call or Not to Call: Diagnosing Intrinsic Over-Calling Bias in LLM Agents
AI 导读
研究发现 LLM 智能体在 When2Call 基准上存在内在过度调用偏差，三个家族六个模型调用准确率高但无调用准确率明显偏低，整体准确率仅 55%-70%。团队用 Sparse Autoencoders 提取与调用/无调用决策对齐的特征基，将偏差归因于激活无关的调用偏移。提出的 AMCS 反偏差方法可缓解过度调用并提升整体准确率，调用准确率几乎不降。
正文
Abstract:LLM agents exhibit a consistent tendency to over-call, invoking tools even in situations where none is needed. On the When2Call benchmark, six models from three families show high call accuracy but much lower no-call accuracy, leaving overall accuracy in the 55%-70% range. We trace this to an Intrinsic Bias Hypothesis (IBH): the call/no-call decision mapping carries an activation-independent call offset, so the model favors call even at activation parity. Using Sparse Autoencoders (SAEs), we recover behavior-aligned feature bases for the call/no_call decision, reduce them to a signed activation margin, and estimate the offset directly. Across all six models, the model is decision-neutral only when no_call activation outweighs call activation, consistent with IBH. We then causally test IBH with Adaptive Margin-Calibrated Steering (AMCS), a closed-form counter-bias shift along SAE decoder directions. Cancelling the diagnosed offset mitigates over-calling and improves overall accuracy with a negligible drop in call accuracy. Our work recasts over-calling from an empirical phenomenon into a mechanistic object amenable to causal correction. Code is available at this https URL.
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2605.18882 [cs.LG] |
| (or arXiv:2605.18882v2 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2605.18882 arXiv-issued DOI via DataCite |
Submission history
From: Wei Shi [view email]
[v1]
Sat, 16 May 2026 04:18:30 UTC (2,666 KB)
[v2]
Tue, 6 Oct 2026 08:37:14 UTC (2,550 KB)
来源：arXiv：cs.LG · arxiv.org