arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2605.27752cs.AI
收藏

询问是不够的:LLM 置信度校准中的协议敏感性

Same Answer, Different Confidence: Protocol Sensitivity in LLM Confidence Calibration

Seoul National University(首尔国立大学)

Hankyeol Kim, Pilsung Kang

AI总结:

研究通过改变测量协议(如条件上下文、令牌读取方式)发现,LLM 的令牌概率置信度与口头置信度之间的比较高度敏感,且口头置信度不仅反映正确性还反映答案的合理性和来源。

摘要:

LLM 置信度校准通常通过比较两种信号来评估:令牌概率分数和口头置信度。这些信号有时被视为模型不确定性的直接读数,但它们的比较取决于很少明确说明的测量选择。在主要分析中,我们固定口头置信度的引出方式:一个单一的提示模板、概率尺度和输出格式。然后,我们改变定义口头与令牌比较的测量轴:哪个答案字符串获得令牌概率分数,如何从答案令牌中读取该分数,以及在哪种条件上下文中测量它。我们在三个开放 7--8B 基础/指令模型家族的四个 QA 基准上评估了这种设计,并使用更大的 Qwen2.5 变体作为同家族鲁棒性检查。结果比较对这些选择敏感:条件上下文改变了跨设置的 ECE 差距的符号或大小,令牌读取产生了更小但仍改变符号的变化,而改变 ECE 估计器影响很小。在默认的生成答案、裸上下文协议下,指令设置接近平衡,而不是显示口头置信度的大幅校准增益。在单独的提供答案分析中,表面合理的错误答案与提供的正确答案获得几乎相同的置信度,这表明口头置信度也反映了答案的合理性和来源,而不仅仅是正确性。我们认为,两种置信度信号都应被视为依赖于协议的测量行为,并提供了一个报告清单,涵盖引出来源、评分答案、令牌概率读取和条件上下文。

英文摘要:

Large language models (LLM) can state their confidence in an answer (verbalized confidence), or the confidence can be read from the probability of the answer's tokens (token confidence). Comparisons of the two ask which is better calibrated, but the token probability of an answer depends on the text that precedes it. Across five open LLMs and four question-answering datasets, we fix each answer, its verbalized confidence, and its correctness label, and vary how each score is measured. The scores themselves change with the protocol, and so does the comparison between them. Changing the context in which the token probability is read shifts the gap between the two scores by about the size of the gap itself (.287 vs. .304 in expected calibration error, ECE; .189 vs. .145 in AUROC). The 95% interval of the shift excludes zero in 19 and 17 of the 20 settings, and under AUROC the shift has the same sign in all of them, so the context changes how answers are ranked. The better score flips in 8 settings under each metric, counting only flips that the intervals support, and the shift also appears in four larger models of up to 72B parameters. Most of the shift appears with changes to the instruction, not to the position of the answer. Other measurement choices also matter, and stated confidence depends on how a supplied answer is written. We turn these results into a reporting checklist for LLM confidence and its comparisons.

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。

↑