当真实表征在欺骗性指令下发生翻转时？
When Truthful Representations Flip Under Deceptive Instructions?
AI总结：
该研究探究大语言模型在欺骗性指令下内部表征的翻转机制，通过线性探针和稀疏自编码器分析两种模型在事实验证任务中的表征，发现欺骗性指令引发中早期层的显著表征偏移，为检测和缓解LLM受指令引导的不诚实行为提供新依据。
摘要：
大语言模型（LLM）往往会遵循恶意设计的指令生成具有欺骗性的回复，这带来了安全挑战。除了输出层面的分析之外，人们对欺骗性指令相较于真实指令如何改变LLM的内部表征仍知之甚少。为填补这一空白，我们研究了在欺骗性指令与真实/中性指令的对比下，这些表征何时以及如何发生“翻转”（例如从真实转向欺骗）。通过在事实验证任务上分析Llama-3.1-8B-Instruct和Gemma-2-9B-Instruct的内部表征，我们发现，在所有条件下，都可以基于内部表征通过线性探针预测模型受指令引导的真/假输出。此外，我们使用稀疏自编码器（Sparse Autoencoders, SAEs）证明，与（彼此相似的）真实/中性表征相比，欺骗性指令会引发显著的表征偏移，这种偏移集中在中早期层，即使在复杂数据集上也能被检测到。我们还识别出了对欺骗性指令高度敏感的特定SAE特征，并通过针对性可视化证实了存在截然不同的真实/欺骗表征子空间。我们的发现揭示了欺骗行为在特征层面和层间的特征信号，为检测和缓解LLM中的受指令引导的不诚实行为提供了新的见解。
英文摘要：
Large language models (LLMs) tend to follow maliciously crafted instructions to generate deceptive responses, posing safety challenges. How deceptive instructions alter the internal representations of LLM compared to truthful ones remains poorly understood beyond output analysis. To bridge this gap, we investigate when and how these representations ``flip'', such as from truthful to deceptive, under deceptive versus truthful/neutral instructions. Analyzing the internal representations of Llama-3.1-8B-Instruct and Gemma-2-9B-Instruct on a factual verification task, we find the model's instructed True/False output is predictable via linear probes across all conditions based on the internal representation. Further, we use Sparse Autoencoders (SAEs) to show that the Deceptive instructions induce significant representational shifts compared to Truthful/Neutral representations (which are similar), concentrated in early-to-mid layers and detectable even on complex datasets. We also identify specific SAE features highly sensitive to deceptive instruction and use targeted visualizations to confirm distinct truthful/deceptive representational subspaces. % Our analysis pinpoints layer-wise and feature-level correlates of instructed dishonesty, offering insights for LLM detection and control. Our findings expose feature- and layer-level signatures of deception, offering new insights for detecting and mitigating instructed dishonesty in LLMs.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
