人格视角：采纳观点会重组 LLM 中的表征
Persona Perspectives: Adopting Viewpoints Reorganizes Representations in LLMs
Austin C. Kozlowski · Avital Mintz · Madison Lee
中文摘要
线性探针（linear probes）和稀疏自编码器（SAEs）等常用的大语言模型（LLM）特征监测方法，假设特征在不同上下文中保持一致的表示。然而，LLM 可以灵活地采纳不同视角或“人格”（persona），这些视角暗示着不同的底层概念关系系统。本文考察提示 LLM 采纳某种人格是否会改变模型潜在空间的表示几何。我们从采纳自由派和保守派政治人格的模型中提取了超过 9,000 个常用词和 33 个语义轴的表示，并与两个中性基线进行比较。我们发现，人格条件会系统性地改变词语在语义轴上的投影以及语义轴之间的对齐方式，且变化方向与指定意识形态一致。随后，我们测试了这种重组对两个安全相关特征——真实性和道德性——的影响。线性探针、基于 SAE 的分类器以及由人工选取词表示构建的分类器，均未在人格条件下训练，但在应用于以政治人格为条件的模型时产生分歧结果，对有争议陈述的真实性和道德性给出带有政治倾向的评分。这些发现表明，按常规方式提取的内部表示在模型采纳人格时可能呈现不同的几何关系，从而可能削弱监测工作。
关键要点
- 01线性探针和 SAE 假设特征表示跨上下文一致，但 LLM 可灵活切换人格视角
- 02从自由派与保守派人格模型提取 9,000+ 词和 33 个语义轴表示，与中性基线对比
- 03人格条件系统性改变词在语义轴上的投影及轴间对齐，方向与意识形态一致
- 04未在人格条件下训练的分类器对政治人格模型给出有政治倾向的真实性与道德性评分
- 05常规提取的内部表示在人格切换下几何关系变化，可能削弱监测有效性
解读
尚无解读。
原始英文摘要
arXiv:2611.06848v1 Announce Type: new Abstract: Common methods for monitoring features in large language models (LLMs) such as linear probes and sparse autoencoders (SAEs) assume that features maintain a consistent representation across contexts. LLMs, however, can flexibly adopt different perspectives or "personas" that suggest different underlying systems of conceptual relations. In this paper, we examine whether prompting an LLM to adopt a persona alters the representational geometry of the model's latent space. We extract representations of over 9,000 common words and 33 semantic axes from models adopting liberal and conservative political personas and compare these against two neutral baselines. We find that persona conditioning systematically alters both the projections of words onto semantic axes and the alignments among the axes themselves in directions consistent with the specified ideology. We then test the implications of this reorganization for two safety-relevant features: truthfulness and morality. Linear probes, SAE-based classifiers, and classifiers built from manually selected word representations, all trained without a persona, produce divergent results when applied to models conditioned on political personas, producing politically slanted ratings of truth and morality on controversial statements. These findings suggest that internal representations, as they are conventionally extracted, can exhibit different geometric relations when the model adopts personas, potentially undermining monitoring efforts.