arXiv cs.CL· Shivam Adarsh, Hetianyu Huang, Lau Safin, Maria Maistro, Christina Lioma·· 11 小时前AI 评分31
激活引导如何改变 LLM 话语层面：识别并干预激活引导引发的话语变化
When Steering Shifts Discourse: Identifying and Intervening on Discourse-Level Changes due to Activation Steering in LLMs
AI 导读
激活引导会在不重训模型权重的情况下系统性改变 LLM 生成文本的话语维度，如正式程度、证据使用和模糊性，此前仅用困惑度等粗粒度指标衡量。研究者提出叠加第二个互补向量来针对受影响的话语维度，该干预恢复了约 78% 的非预期话语变化，同时在约 93% 的生成中保留了预期的引导行为。
正文 · 原文
Abstract:Activation steering is an inference-time technique used to control or modify the behavior, tone, or safety properties of a large language model (LLM) without retraining the model's weights or editing its input prompt, but simply by adding a vector to its residual-stream activations. However, activation steering also changes the generated text in unintended ways, unrelated to intended behavior change, that are not well understood. Prior work has measured these changes only through coarse signals such as perplexity. In this work, we show that steering systematically produces significant changes in discourse dimensions of the generated text, such as its formality, use of evidence, and ambiguity. We present a way of countering this by adding a second, complementary vector that targets the affected discourse dimension. This intervention restores about 78% of the unintended discourse changes while preserving the intended steering behavior in about 93% of the generations.
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.41661 [cs.CL] |
| (or arXiv:2610.41661v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.41661 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Shivam Adarsh [view email]
[v1]
Fri, 30 Oct 2026 12:32:47 UTC (1,015 KB)
来源：arXiv cs.CL · arxiv.org