发现可操控大模型拒绝回答的隐藏向量，实现对AI“审查”行为的精准控制。
Steering the CensorShip: Uncovering Representation Vectors for LLM "Thought" Control
- 通过表示工程找到控制模型拒绝响应的向量。
- 在DeepSeek-R1衍生模型中发现思维抑制现象，拒绝率降低至5%以下。
- 只需反向应用该向量即可消除审查，适合安全可控研究者使用。
大型语言模型（LLMs）已改变我们获取信息的方式。这些模型通常被调优以拒绝有害请求，并生成更符合控制方偏好的回应。为理解这种“审查”机制，我们采用表示工程方法研究开源权重的安全调优模型。提出一种寻找拒绝-合规向量的方法，可检测并控制模型输出中的审查程度。我们还分析了从DeepSeek-R1蒸馏而来的近期推理型LLM，发现“思维抑制”这一额外审查维度。类似方法可用于定位抑制模型推理过程的向量，通过应用该向量的负值，可有效移除审查。代码已公开于：https://github.com/hannahxchen/llm-censorship-steering。
原文摘要 · Abstract (English)
Large language models (LLMs) have transformed the way we access information. These models are often tuned to refuse to comply with requests that are considered harmful and to produce responses that better align with the preferences of those who control the models. To understand how this "censorship" works. We use representation engineering techniques to study open-weights safety-tuned models. We present a method for finding a refusal--compliance vector that detects and controls the level of censorship in model outputs. We also analyze recent reasoning LLMs, distilled from DeepSeek-R1, and uncover an additional dimension of censorship through "thought suppression". We show a similar approach can be used to find a vector that suppresses the model's reasoning process, allowing us to remove censorship by applying the negative multiples of this vector. Our code is publicly available at: https://github.com/hannahxchen/llm-censorship-steering
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。