大型语言模型不确定性中的人类对齐、校准与激活模式
Human-Alignment, Calibration, and Activation Patterns in Large Language Model Uncertainty
AI总结：
研究大型语言模型的不确定性与人类不确定性的相似性，通过分析行为与内部激活模式，发现模型在多项选择和开放式事实回忆数据集上同时存在对齐与校准，并描述了指令微调的影响。
摘要：
不确定性量化是大型语言模型行为分析中一个庞大且不断发展的子领域。为了识别和对抗幻觉，该领域主要关注测量和改进校准，即不确定性判断对任务效能的准确性。在这项工作中，我们探讨了一个相对未被充分探索的问题：大型语言模型的不确定性与人类不确定性有多相似。我们研究了大型语言模型的外部行为和内部激活模式中是否存在类似人类的不确定性信号，即不确定性对齐。我们识别了模型在涵盖多项选择和开放式事实回忆的多种数据集上是否同时表现出对齐和校准的证据。并且我们描述了指令微调对这些方面的影响。
英文摘要：
Uncertainty Quantification is a large and growing subfield of large language model behavioral analysis. Primarily to recognize and combat hallucination, the field has largely focused on measuring and improving calibration, the accuracy of uncertainty judgments to task efficacy. In this work, we investigate the relatively underexplored question of how similar large language model uncertainty is to human uncertainty. We investigate the presence and strength of human-similar uncertainty signals, deemed uncertainty alignment, in large language model overt behavior and internal activation patterns. We identify whether the models show evidence of simultaneous alignment and calibration on a variety of datasets covering both multiple choice and open ended factual recall. And we characterize the effect of instruct fine-tuning on each of these facets.
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。
