发现大模型代理过度调用工具的内在偏见，并提出可直接纠正的方法。
To Call or Not to Call: Diagnosing Intrinsic Over-Calling Bias in LLM Agents
- 通过稀疏自编码器定位调用决策的特征表示，揭示其固有偏向性。
- 在六个模型上验证：仅当不调用激活更强时，模型才中立，符合偏见假设。
- 设计可解释的校正方法，显著减少误调用，提升整体准确率。
大模型代理普遍存在不必要的工具调用倾向。在When2Call基准上，六个来自三个模型家族的模型表现出高调用准确率但低不调用准确率，整体准确率仅55%-70%。我们提出内在偏见假说（IBH）：调用/不调用决策映射存在与激活无关的调用偏移，使模型在激活相等时仍倾向于调用。利用稀疏自编码器（SAEs），我们恢复了与行为对齐的特征基，将其简化为带符号的激活差值，并直接估计该偏移。所有六种模型均显示：仅当不调用激活强于调用激活时，模型才保持中立，支持IBH。随后，我们通过自适应边际校准引导（AMCS）进行因果验证，沿SAE解码方向施加闭式反偏移。消除诊断出的偏移后，过调用现象缓解，整体准确率提升，调用准确率几乎无损。本工作将过调用从经验现象转化为可因果干预的机制对象。代码已公开于https://github.com/SKURA502/agent-sae/。
原文摘要 · Abstract (English)
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 https://github.com/SKURA502/agent-sae/.
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。