English(EN) Component and Dimension Sparsity in Transformer Refusal Mechanisms

Transformer 拒绝机制被发现是稀疏且可识别的

作者 PulseAugur 编辑部 · [1 个来源] · 2026-10-07 04:00

研究人员发现，大型语言模型中的 Transformer 拒绝机制并非弥散编码，而是由结构化、可识别的机制组装而成。通过分解拒绝引导，他们发现注意力（attention）和 MLP 组件的稀疏子集，以及特定的残差流（residual stream）维度，足以重现完整的行为效果。这表明拒绝行为是通过这些集中的机制来表示和引导的，为理解其功能奠定了基础。 AI

影响识别出 LLM 中拒绝行为的具体、可引导的机制，可能有助于安全和控制研究。

排序理由该集群包含一篇详细介绍 Transformer 拒绝机制研究结果的研究论文。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.AI 阅读 →

AI 生成摘要 · Google Gemini · 来自 1 个来源。我们如何撰写摘要 →

本文如何被排名

Signal score: 15 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 该集群包含一篇详细介绍 Transformer 拒绝机制研究结果的研究论文。[lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration: Single-source cluster; Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics: paper, safety; Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance: High; Clearly on-topic for AI-industry coverage.
Story freshness: Breaking (< 6h); Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

完整方法见我们的编辑标准。

报道来源 [1]

arXiv cs.AI TIER_1 English(EN) · Vincent Siu, Glenn Grant-Richards, Vlad Pavlovich, Yizhou Sun, Dawn Song, Chenguang Wang · 2026-10-07 04:00

Transformer 拒绝机制中的组件和维度稀疏性

arXiv:2610.06903v1 Announce Type: cross Abstract: Activation steering manipulates large language model behavior by intervening on internal activations, but the mechanistic basis of these interventions remains poorly understood. We decompose refusal steering into component-level i…

本文如何被排名

报道来源 [1]

Transformer 拒绝机制中的组件和维度稀疏性

相关实体

相关话题