用字典学习特征做分类器：比原始激活更准
Using Dictionary Learning Features as Classifiers A preliminary note comparing feature-based and raw-activation based harmfulness classifiers.
做 AI 安全和对齐的团队可以关注——字典学习特征让分类器更准更可解释，比直接用原始激活值靠谱，值得在内容审核场景里试试。
Anthropic 的 Transformer Circuits 团队发布了一篇技术笔记，探讨如何利用字典学习（Dictionary Learning）提取的稀疏特征来构建有害内容分类器。
研究发现，基于这些特征训练的分类器在检测有害性上优于直接使用模型原始激活值的分类器，且更具可解释性。这项工作为构建更安全、更可控的 AI 系统提供了新思路，尤其适用于需要细粒度内容审核的场景。
This report is categorized under AITOP's "Papers" desk. Key entities and technical topics under continuous tracking include 字典学习, 特征分类器, AI安全, 可解释性.
Primary source verification: published directly by Anthropic: Transformer Circuits with first-party authority. Cross-source corroboration: 4 independent sources have covered this development (including Dario Amodei Blog, The Rundown AI, arXiv: OpenAI), confirming industry-wide signal strength.