MechELK：一种用于激发大型语言模型中潜在知识的机制可解释性框架
MechELK: A Mechanistic Interpretability Framework for Eliciting Latent Knowledge in Large Language Models
AI总结：
提出MechELK框架，通过定位、验证和激发三个阶段，利用稀疏自编码器特征分析和因果探测等方法，从大型语言模型中提取隐藏知识，在TruthfulQA等基准上平均激发准确率达84.7%。
摘要：
大型语言模型（LLMs）经常在其内部表示中编码事实和推理知识，但这些知识并未在其表面输出中忠实反映——这种现象被称为“潜在知识”。现有的潜在知识激发方法，如对比一致性搜索（CCS），依赖于对比激活模式，在处理复杂的多步推理任务时存在困难，而机制可解释性工具主要用于“理解”模型行为而非“提取”隐藏知识。我们提出了**MechELK**，一个统一的三个阶段框架，桥接了机制可解释性和潜在知识激发。MechELK通过以下步骤运作：（1）**定位**——使用稀疏自编码器（SAE）特征分析和激活修补来识别承载知识的表示；（2）**验证**——采用因果探测来区分真正的潜在知识与虚假相关性；（3）**激发**——应用表示工程在不修改模型权重的情况下揭示隐藏知识。在TruthfulQA、精心策划的Deceptive Alignment基准和Quirky LM数据集上评估，MechELK实现了84.7%的平均激发准确率，比CCS高出6.2%，比直接线性探测高出9.1%。关键的是，MechELK在模型表面输出不正确或回避的78.3%案例中成功识别了潜在知识，展示了其在AI安全应用（包括欺骗性对齐检测）中的实用性。
英文摘要：
Large language models (LLMs) frequently encode factual and reasoning knowledge in their internal representations that is not faithfully reflected in their surface-level outputs -- a phenomenon known as \emph{latent knowledge}. Existing approaches to eliciting latent knowledge, such as Contrastive Consistency Search (CCS), rely on contrastive activation patterns and struggle with complex multi-step reasoning tasks, while mechanistic interpretability tools have primarily been used to \emph{understand} model behavior rather than to \emph{extract} hidden knowledge. We present \textbf{MechELK}, a unified three-stage framework that bridges mechanistic interpretability and latent knowledge elicitation. MechELK operates through: (1) \textbf{Locate} -- using Sparse Autoencoder (SAE) feature analysis and activation patching to identify knowledge-bearing representations; (2) \textbf{Verify} -- employing causal probing to distinguish genuine latent knowledge from spurious correlations; and (3) \textbf{Elicit} -- applying representation engineering to surface hidden knowledge without modifying model weights. Evaluated on TruthfulQA, a curated Deceptive Alignment benchmark, and the Quirky LM dataset, MechELK achieves an average elicitation accuracy of 84.7\%, outperforming CCS by 6.2\% and direct linear probing by 9.1\%. Crucially, MechELK successfully identifies latent knowledge in 78.3\% of cases where the model's surface output is incorrect or evasive, demonstrating its utility for AI safety applications including deceptive alignment detection.
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。
