LLMs Know More Than They Show: 论大语言模型幻觉的内在表征
LLMs Know More Than They Show: On the Intrinsic Representation of LLM Hallucinations
AI总结：
本研究发现大语言模型内部表征编码的真实性信息远超此前认知，指出其真实性编码具多面性且可预测错误类型，还揭示了内部编码与外部行为的差异，为错误分析与缓解提供新视角。
摘要：
大语言模型（LLM）常会产生各类错误，包括事实性偏差、偏见以及推理失误，这些错误统称为“幻觉”。近期研究表明，LLM的内部状态编码了其输出真实性的相关信息，且该信息可用于检测错误。在本研究中，我们发现LLM内部表征所编码的真实性信息远多于此前的认知。我们首先发现真实性信息集中在特定token中，利用这一特性可显著提升错误检测性能。但我们也发现，这类错误检测器无法跨数据集泛化，这意味着与此前的论断相反，真实性编码并非通用的，而是多面性的。其次，我们证明内部表征还可用于预测模型可能出现的错误类型，为制定针对性的缓解策略提供便利。最后，我们揭示了LLM内部编码与外部行为之间的差异：模型可能编码了正确答案，却始终生成错误结果。综上，这些发现从模型内部视角深化了我们对LLM错误的理解，可为未来强化错误分析与缓解的研究提供指导。
英文摘要：
Large language models (LLMs) often produce errors, including factual inaccuracies, biases, and reasoning failures, collectively referred to as "hallucinations". Recent studies have demonstrated that LLMs' internal states encode information regarding the truthfulness of their outputs, and that this information can be utilized to detect errors. In this work, we show that the internal representations of LLMs encode much more information about truthfulness than previously recognized. We first discover that the truthfulness information is concentrated in specific tokens, and leveraging this property significantly enhances error detection performance. Yet, we show that such error detectors fail to generalize across datasets, implying that -- contrary to prior claims -- truthfulness encoding is not universal but rather multifaceted. Next, we show that internal representations can also be used for predicting the types of errors the model is likely to make, facilitating the development of tailored mitigation strategies. Lastly, we reveal a discrepancy between LLMs' internal encoding and external behavior: they may encode the correct answer, yet consistently generate an incorrect one. Taken together, these insights deepen our understanding of LLM errors from the model's internal perspective, which can guide future research on enhancing error analysis and mitigation.
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。
