arXiv：cs.AI· Shuhul Razdan, R Balamurugan·· 16 小时前
语言模型中的概念子补集增益：算子几何与行为归因
When a Conceptor Complement Becomes Gain: Operator Geometry and Behavioral Attribution in Language Models
AI 导读
研究针对以目标为中心、带缩放的 conceptor 补集，在保留目标专属中心与增益的前提下移除拟合协方差滤波器，考察转向干预中哪部分应归因于行为变化。在 α=.05、β=2 下，三个模型七个目标的 21 个拟合映射在生成干预层均扩展每一方向，矩阵修正为匹配控制位移的 0.093%–7.815%。
正文
Abstract:When a steering intervention changes a language model's behavior, which part of the intervention deserves the credit? We test this question for a target-centered, scaled conceptor complement by removing its fitted covariance filter while preserving the same target-specific center and gain. At the original aperture $\alpha=.05$ and gain $\beta=2$, all 21 fitted maps across three models and seven targets expand every direction at the generation intervention layers. The matrix correction is $0.093\%$--$7.815\%$ of the matched-control displacement on held-out extraction activations. In 480 paired comparisons within a 1,920-response study, retaining the matrix changes many individual texts but does not establish a consistent additional target-suppression benefit on the four-target panel. A geometric aperture sweep shows that the operating regime changes with the filter setting. Separately, an invertible soft filter preserves what an unrestricted affine readout can recover, whereas dedicated erasure methods reduce prediction by newly fitted probes. Together, the results show why a whole-intervention comparison cannot by itself attribute generated behavior to the fitted matrix: analyze the complete transformation, hold shared components fixed, and test its output consequences directly.
| Comments: | 50 pages. Accepted at the NeurIPS 2026 NeurReps Workshop (Symmetry and Geometry in Neural Representations). Code: this https URL |
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.16325 [cs.AI] |
| (or arXiv:2610.16325v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.16325 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Shuhul Razdan [view email]
[v1]
Mon, 12 Oct 2026 07:43:46 UTC (1,516 KB)
来源：arXiv：cs.AI · arxiv.org