arXiv cs.CY· Nimun Bajwa, Massimo Poesio·· 13 小时前AI 评分31
在 Llama-3.1-8B-Instruct 上探测与引导大语言模型的安全相关行为
Probing and Steering for Safety-Relevant Behaviours in Large Language Models
AI 导读
研究在 Llama-3.1-8B-Instruct 上统一流程考察激活探针设计选择对欺骗、高风险意识和谄媚三种行为的影响。将两种估计器（均值差与逻辑回归）与三种 token 池化方案交叉组合，发现“逻辑回归擅长检测、均值差擅长引导”的读写分离在欺骗与高风险行为上成立，但在谄媚行为上反转。
正文 · 原文
Abstract:Activation probes are computed directions in a language model's residual stream, used both to detect safety-relevant behaviours and, by adding their direction back into the stream, to steer them. We explore how the main design choices behind such probes interact, within one pipeline, on one model (Llama-3.1-8B-Instruct), and across three behaviours: deception, high-stakes awareness and sycophancy. The pipeline crosses two estimators (difference-of-means and logistic regression) with three token-pooling schemes (last, mean and learned attention), selects appropriate layers to detect or steer a behaviour, and evaluates every direction twice: as a detector of the behaviour, and as a steering vector to push outputs towards the behaviour. Prior work read together suggests a read/write dissociation, in which logistic regression reads (detects) best but difference-of-means directions write (steer) best. We find that it holds for deception and high-stakes but not for sycophancy, where the pattern reverses. The reversal appears to follow partially from the data rather than from the estimators. The best pooling technique and the layer choice depend on the behaviour and data, and the layer that steers best can sit well before the layer that detects best. The right steering metric to use between changing a model's perception and eliciting behaviour depends on the task - and the two can disagree outright. The exported probes and directions power an interactive interface through which a user can score and steer text directly.
| Comments: | 20 pages, 26 figures, 11 tables. Code, data and interactive interface: this https URL |
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Computers and Society (cs.CY) |
| Cite as: | arXiv:2610.30363 [cs.LG] |
| (or arXiv:2610.30363v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.30363 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Nimun Bajwa [view email]
[v1]
Wed, 21 Oct 2026 19:25:24 UTC (3,890 KB)
来源：arXiv cs.CY · arxiv.org