跳到正文
研踪PaperTrail
← 研踪 PaperTrail

arXiv:2611.20441 · v1

Abstention Direction in Large Language Models: Mechanistic Interpretability of Abstention Across Query Ambiguity, Knowledge Limits, and Safety Alignment

首次提交 2026-11-16 11:12
版本更新 2026-11-16 11:12
雷达观测 2026-11-17 13:07
arXiv ↗PDF ↗

摘要

Abstention, the withholding of a response by a large language model (LLM) under uncertainty or safety risks, is essential for trustworthy models. The causes of abstention fall into three perspectives: ambiguity in the Query (Q), limits of Model Knowledge (MK), and misalignment with Human Values (HV). However, prior analyses of internal representations have focused on HV, using a refusal direction computed from harmful and harmless queries. No study has analyzed the three perspectives jointly. We therefore extend the refusal direction framework to all three perspectives. Specifically, for each perspective, we construct counterfactual pairs that share a template and differ only in the abstention-inducing factor, and compute an abstention direction by normalizing the difference in mean activations between the two polarities. Activation steering experiments show that intervening in middle layers increases or decreases abstention across all three perspectives. Moreover, steering yields cross-perspective effects, indicating that the abstention directions share a common directional component.

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

目前只收录了一个版本。收录下一版后,可以对照标题、作者、分类和摘要。

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。

v1 · 2026-11-16 11:12 · 最新观测版本

Abstention Direction in Large Language Models: Mechanistic Interpretability of Abstention Across Query Ambiguity, Knowledge Limits, and Safety Alignment

Abstention, the withholding of a response by a large language model (LLM) under uncertainty or safety risks, is essential for trustworthy models. The causes of abstention fall into three perspectives: ambiguity in the Query (Q), limits of Model Knowledge (MK), and misalignment with Human Values (HV). However, prior analyses of internal representations have focused on HV, using a refusal direction computed from harmful and harmless queries. No study has analyzed the three perspectives jointly. We therefore extend the refusal direction framework to all three perspectives. Specifically, for each perspective, we construct counterfactual pairs that share a template and differ only in the abstention-inducing factor, and compute an abstention direction by normalizing the difference in mean activations between the two polarities. Activation steering experiments show that intervening in middle layers increases or decreases abstention across all three perspectives. Moreover, steering yields cross-perspective effects, indicating that the abstention directions share a common directional component.

查看此版本解读 →查看此版本原文 ↗