区分大模型评判的两种不确定性，精准定位需人工标注的样本。
Decomposing LLM-Judge Uncertainty to Target Expert Labels
- 用贝叶斯模型分离判断中的认知不确定性和随机不确定性。
- 相同标注成本下，新方法可减少83%的错误，优于传统全量标注。
- 适合大规模模型评估中优化人工标注资源分配的场景。
大语言模型（LLM）用于大规模输出评估，但其判断存在两种不确定性：一种是专家间真实分歧（随机不确定性），无法通过更多标注消除；另一种是模型自身知识不足（认知不确定性），可通过标注改善。本文提出一个小型贝叶斯模型，仅基于已有标注数据进行回归，即可估计黑箱判别器的可信度。两个分量均可通过简单公式计算，无需额外采样或调用模型。在真实专家分歧数据集ChaosNLI上，基于认知不确定性的排序比总不确定性减少83%的误差，且效果优于直接按置信度升序标记。结果表明，模型置信度不能反映实际错误率。我们证明可识别模型真正无知的区域而非专家分歧处，并建议据此指导人工标注。
原文摘要 · Abstract (English)
An LLM judge evaluates outputs at scale. Experts should label only where it is least sure. Its natural escalation signal conflates two uncertainties: aleatoric, real disagreement in the expert pool, which labels cannot reduce, and epistemic, the judge's ignorance, which labels do reduce. A small Bayesian model separates them: a regression on labels already collected learns how far to trust a black-box judge's prediction. Both components follow as simple formulas, with no sampling or further judge calls. The components isolate on a real LLM judge against exactly known truth, and stated confidence is no guide to its actual error. On real human disagreement (ChaosNLI) the epistemic ranking removes 83% more error than total uncertainty for the same expert labels, though simply escalating the least-labelled items does as well there. We demonstrate we can estimate where a judge is ignorant rather than where experts genuinely disagree, and propose using this to direct expert labelling.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。