arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2509.04464cs.CLcs.AI
收藏

大语言模型的多次响应能否揭示其不确定性的来源?

Can Multiple Responses from an LLM Reveal the Sources of Its Uncertainty?

University of Arizona(亚利桑那大学);Michigan State University(密歇根州立大学)

Yang Nan, Pengfei He, Ravi Tandon, Han Xu

AI总结:

本研究针对大语言模型不确定性来源诊断不足的问题,提出利用LLM多次响应的分歧模式,通过辅助LLM分析推断不确定性来源(如问题歧义、知识缺失等),并在多个基准上验证了框架的通用性,为提升LLM性能与可靠性提供支持。

摘要:

大语言模型(LLM)已在多个领域取得重大突破,但仍可能生成不可靠或具有误导性的输出,给实际应用带来严峻挑战。尽管近期许多研究聚焦于量化模型不确定性,但针对“诊断不确定性来源”的工作相对较少。本研究表明,当LLM存在不确定性时,其生成的多个响应之间的分歧模式包含了关于不确定性根本原因的丰富线索。为阐明这一点,我们从目标LLM收集多个响应,并使用辅助LLM分析其分歧模式。辅助模型的任务是推断不确定性的可能来源,例如其是否源于输入问题的歧义、相关知识的缺失,或两者兼有。在涉及知识缺口的情况下,辅助模型还会识别导致不确定性的具体缺失事实或概念。在实验中,我们在AmbigQA、OpenBookQA和MMLU-Pro上验证了我们的框架,证实了其在诊断不同不确定性来源方面的通用性。此类诊断显示出开展相关人工干预以提升LLM性能和可靠性的潜力。

英文摘要:

Large language models (LLMs) have delivered significant breakthroughs across diverse domains but can still produce unreliable or misleading outputs, posing critical challenges for real-world applications. While many recent studies focus on quantifying model uncertainty, relatively little work has been devoted to \textit{diagnosing the source of uncertainty}. In this study, we show that, when an LLM is uncertain, the patterns of disagreement among its multiple generated responses contain rich clues about the underlying cause of uncertainty. To illustrate this point, we collect multiple responses from a target LLM and employ an auxiliary LLM to analyze their patterns of disagreement. The auxiliary model is tasked to reason about the likely source of uncertainty, such as whether it stems from ambiguity in the input question, a lack of relevant knowledge, or both. In cases involving knowledge gaps, the auxiliary model also identifies the specific missing facts or concepts contributing to the uncertainty. In our experiment, we validate our framework on AmbigQA, OpenBookQA, and MMLU-Pro, confirming its generality in diagnosing distinct uncertainty sources. Such diagnosis shows the potential for relevant manual interventions that improve LLM performance and reliability.

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。

↑