通过定位神经元激活来提取大模型决策规则，效率比全量测试高97%以上。
Neuron-Anchored Rule Extraction for Large Language Models via Contrastive Hierarchical Ablation
- 基于对比分层消融，精准定位影响规则行为的稀疏激活神经元。
- 在算术和越狱任务中，仅用2.14%的计算成本召回97%关键神经元。
- 可精准修复错误或诱导越狱，适用于模型可解释性与安全控制研究。
可解释人工智能的核心目标是符号化表达大语言模型的决策逻辑，并将其锚定在内部机制上。现有规则提取方法通常生成无根基的符号代理，而机制可解释性虽能关联行为与神经元，但依赖手工假设且代价高昂。本文提出MechaRule，一种将规则提取锚定在大模型电路中的流程：通过定位稀疏激动激活（agonist activations），其消融会破坏规则相关行为。该方法基于两个发现：第一，在固定基线/翻转条件下，稀疏激动效应呈现“超越性”——少数高影响激活在群体中主导弱激活并翻转大量相同样本；在此类条件下，自适应组测试结合置信度引导的保守剪枝，只需O(k log(N/k) + k)次干预（当k << N时）即可完成检测。第二，激动激活的定位更可靠于贴近真实规则行为的数据划分；谱划分提供无规则备选，而不真实划分则降低定位精度。实验表明，在算术和越狱任务中，MechaRule在平均仅2.14%的全量消融成本下，对最高效应激动激活的召回率达97.0%，且消融定位激活可消除97.6%–100.0%的正确算术答案与越狱行为，并可通过操纵实现最多72.8%的算术纠错或32.5%的越狱诱导。
原文摘要 · Abstract (English)
A central goal of explainable AI is to express large language model (LLM) decision logic symbolically and ground it in internal mechanisms. Existing rule-extraction methods usually learn ungrounded symbolic surrogates, while mechanistic interpretability links behavior to neurons but often requires hand-crafted hypotheses and costly interventions. We introduce MechaRule, a pipeline that grounds rule extraction in LLM circuits by localizing sparse agonist activations whose ablation disrupts rule-related behavior. MechaRule rests on two findings. First, in a fixed baseline/flip regime, sparse agonist effects can exhibit overtopping: a few high-effect activations remain detectable within larger groups, dominate weaker ones, and flip many of the same examples. In such regimes, adaptive group testing with confidence-guided conservative pruning requires O(k log(N/k) + k) interventions over N candidates when k << N are agonists. Second, agonists are localized more reliably on data splits aligned with close-to-faithful rule behavior; spectral splits provide a rule-free fallback, whereas unfaithful splits degrade localization. Empirically, on arithmetic and jailbreaking, MechaRule recalls 97.0% of highest-effect agonists in matched brute-force validations at only 2.14% of exhaustive-ablation cost on average. Ablating the localized agonists eliminates 97.6--100.0% of eligible correct arithmetic answers and jailbreaks, and can correct arithmetic errors or induce jailbreaks by up to 72.8% and 32.5%.
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。