arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2603.24967cs.AI
收藏

大语言模型中不确定性的解剖结构

The Anatomy of Uncertainty in LLMs

School of Computing and Augmented Intelligence, Arizona State University(亚利桑那州立大学计算与增强智能学院);Lawrence Livermore National Laboratory(劳伦斯利弗莫尔国家实验室)

Aditya Taparia, Ransalu Senanayake, Kowshik Thopalli, Vivek Narayanaswamy

AI总结:

本文提出了一种将大语言模型不确定性分解为输入模糊性、知识缺口和解码随机性的框架,通过实验展示不同模型规模和任务中各组件的重要性,以提升模型可靠性与可信度。

摘要:

理解大语言模型(LLM)为何对响应不确定对于其可靠部署至关重要。当前方法要么提供单一不确定性评分,要么依赖经典的概率-epistemic二元对立,无法为生成模型提供改进的可行见解。近期研究表明,此类方法不足以理解LLM中的不确定性。本文倡导一种不确定性分解框架,将LLM的不确定性分解为三个不同的语义组件:(i)输入模糊性,源于模糊的提示;(ii)知识缺口,由参数证据不足引起;(iii)解码随机性,源于随机采样。通过一系列实验,我们证明这些组件的主导地位会随模型规模和任务变化。我们的框架提供了更好的理解以审计LLM可靠性并检测幻觉,为有针对性的干预和更可信的系统铺平道路。

英文摘要:

Understanding why a large language model (LLM) is uncertain about the response is important for their reliable deployment. Current approaches, which either provide a single uncertainty score or rely on the classical aleatoric-epistemic dichotomy, fail to offer actionable insights for improving the generative model. Recent studies have also shown that such methods are not enough for understanding uncertainty in LLMs. In this work, we advocate for an uncertainty decomposition framework that dissects LLM uncertainty into three distinct semantic components: (i) input ambiguity, arising from ambiguous prompts; (ii) knowledge gaps, caused by insufficient parametric evidence; and (iii) decoding randomness, stemming from stochastic sampling. Through a series of experiments we demonstrate that the dominance of these components can shift across model size and task. Our framework provides a better understanding to audit LLM reliability and detect hallucinations, paving the way for targeted interventions and more trustworthy systems.

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。

↑