部分正确信息会诱使大模型更自信地产生错误推理。
Anchored Confabulation: Partial Evidence Non-Monotonically Amplifies Confident Hallucination in LLMs
- 用一个确定的中间事实锚定模型，使其后续推理更自信出错。
- 在多步推理中，部分证据使模型错误率上升至0.656（原为0.613）。
- 适合研究模型幻觉、RAG路由优化或提升推理可信度的人阅读。
我们发现大语言模型存在一种此前未知的校准特性：在多步推理链中提供一个已确认的中间事实，会先提高模型对错误答案的自信程度，直到完整证据出现才下降。这种现象称为‘锚定幻觉’：部分证据让模型在参数上固执于后续推理的完成。我们提出参数化幻觉置信度（PHC）进行量化，在六组证据中验证，包括因果注入实验（PHC从0.613升至0.656再降至0.595和0.536，N=160）及五类模型的能力扩展（斯皮尔曼等级相关系数rho=0.900，p=0.037）。锚定阈值定律k*(n)=floor(n/3)成功预测了随推理步数增加的PHC放大效应，有四项预测得到证实。应用于RAG路由时，基于PHC设计的LearnedRouter在四个基准上处理1800个查询，无需微调，仅用50倍少标签即达到81.1%的基线性能差距弥补（宏F1=0.426，p<1e-6）。使用认知谦逊提示可使PHC峰值下降-0.118；显式自我评分（PHC=0.684，p<0.001）优于词汇置信度作为路由信号。
原文摘要 · Abstract (English)
We identify a previously unknown calibration property of large language models: providing one confirmed intermediate fact toward a multi-step reasoning chain increases the model's confident-wrong-answer rate before full evidence eliminates it. We call this anchored confabulation: a partial anchor commits the model to confident parametric completion of remaining reasoning steps. We formalize it as Parametric Hallucination Confidence (PHC) and establish it across six lines of evidence including a causal injection experiment (PHC 0.613 to 0.656 to 0.595 to 0.536, N=160) and capability scaling across five model families (Spearman rho=0.900, p=0.037). The Anchoring Threshold Law k*(n)=floor(n/3) predicts PHC amplification by hop depth with four confirmed predictions. Applied to RAG routing, a LearnedRouter exploiting PHC closes 81.1% of the oracle performance gap (macro F1=0.426, p<1e-6) on 1,800 queries across four benchmarks with no model fine-tuning and 50x fewer labels than prior RL-based work. An epistemic humility prompt reduces the PHC spike by -0.118; explicit self-rating (PHC=0.684, p<0.001) outperforms lexical confidence as a routing signal.
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。