arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2606.07313cs.CLcs.AI
收藏

SV-Detect: 基于引导向量的AI生成文本检测

SV-Detect: AI-generated Text Detection with Steering Vectors

Independent Researcher(独立研究者);Queen Mary University of London(伦敦女王学院)

Mikhail Vishnyakov, Tatiana Gaintseva

AI总结:

提出从冻结语言模型的隐藏表示中提取引导向量,通过层间投影特征训练轻量分类器,实现跨域、跨模型和编辑攻击下的机器生成文本检测。

摘要:

检测机器生成文本在分布偏移(如跨域、源模型和编辑攻击的迁移)下尤其困难。我们提出了一种基于从冻结语言模型的隐藏表示中提取的引导向量的假文本检测器。在每一层,我们构建一个分离人类编写文本和机器生成文本的方向,并通过每个输入与这些方向的逐层对齐来表示输入。在这些投影特征上训练的轻量分类器产生最终的检测分数。我们的方法在分布内和分布偏移下均表现出色,包括跨域、跨源模型以及机器编辑转换(如润色和重写)。解释分析表明,学习到的方向与可识别的风格线索一致,同时捕获了超越表面特征的显著额外信号。这些结果将假文本检测定位为表示空间探测问题,并表明引导向量提供了一种简单有效的解决方案。

英文摘要:

Detecting AI-generated text is especially difficult under distribution shift, such as transfer across domains, source models, and editing attacks. We propose an AI-generated text detector based on steering vectors extracted from the hidden representations of a frozen language model. At each layer, we construct a direction that separates human-written from AI-generated text, and represent each input by its layer-wise alignment with these directions. A lightweight classifier trained on these projection features yields the final detection score. Our method achieves strong performance both in-distribution and under distribution shift, including across domains, source models, and machine-editing transformations such as polishing and rewriting. Interpretation analyses show that the learned directions align with recognizable stylistic cues while capturing substantial additional signal beyond surface features. These results position AI-generated text detection as a representation-space probing problem and show that steering vectors provide a simple and effective solution.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑