用线性探针可高精度识别大模型生成的欺骗性回答。
Caught in the Act: a mechanistic approach to detecting deception
- 通过分析模型内部激活值，用线性探针检测欺骗行为。
- 70-80%准确率（>7B模型），推理类回答超90%准确。
- 小模型难识别，大模型多方向编码欺骗特征，适合安全研究者。
先进的AI系统可能具备提示价值观错位的指标，如同汽车的“检查发动机”灯。其中一种错位表现是生成回答时的欺骗性。未来AI系统或可检测语言模型在推理看似合理但错误的事实问题时产生的欺骗性回应。本文证明，对语言模型内部激活值进行线性探测，可极高精度识别欺骗性回答。探测器在1.5B至14B参数的Llama与Qwen系列模型（包括DeepSeek-r1微调版本）上达到超过90%的准确率。小模型（1.5B）探测准确率接近随机（50%），而大模型（>7B）准确率达70-80%，其推理类回答更超90%。逐层探测显示三阶段模式：早期层接近随机（50%），中层峰值，后期略有下降。进一步通过迭代空投影法发现，欺骗性信息以多重线性方向编码，从Qwen 3B的20个到DeepSeek 7B和Qwen 14B的近100个不等。
原文摘要 · Abstract (English)
Sophisticated instrumentation for AI systems might have indicators that signal misalignment from human values, not unlike a "check engine" light in cars. One such indicator of misalignment is deceptiveness in generated responses. Future AI instrumentation may have the ability to detect when an LLM generates deceptive responses while reasoning about seemingly plausible but incorrect answers to factual questions. In this work, we demonstrate that linear probes on LLMs internal activations can detect deception in their responses with extremely high accuracy. Our probes reach a maximum of greater than 90% accuracy in distinguishing between deceptive and non-deceptive arguments generated by llama and qwen models ranging from 1.5B to 14B parameters, including their DeepSeek-r1 finetuned variants. We observe that probes on smaller models (1.5B) achieve chance accuracy at detecting deception, while larger models (greater than 7B) reach 70-80%, with their reasoning counterparts exceeding 90%. The layer-wise probe accuracy follows a three-stage pattern across layers: near-random (50%) in early layers, peaking in middle layers, and slightly declining in later layers. Furthermore, using an iterative null space projection approach, we find multitudes of linear directions that encode deception, ranging from 20 in Qwen 3B to nearly 100 in DeepSeek 7B and Qwen 14B models.
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。