Calibration Drift Under Reasoning: How Chain-of-Thought Budgets Induce Overconfidence in Large Language Models
推理下的校准漂移:思维链预算如何导致大型语言模型过度自信
机构 * Department of Computer Science and Engineering, Visvesvaraya Technological University, Belagavi(维斯瓦拉亚科技大学计算机科学与工程系,贝拉加维) ; Department of Computer Science and Business System, SG Balekundri Institute of Technology, Belagavi(SG巴莱昆德里理工学院计算机科学与商业系统系,贝拉加维)
专题命中 推理评测 :reasoning(title,abstract);chain-of-thought(title,abstract);CoT(abstract,abstract_cn);分类 cs.CL、cs.AI、cs.LG
AI总结 研究发现,增加思维链推理预算超过任务特定阈值会导致模型对错误答案过度自信,提出校准漂移现象并引入CABStop停止规则。
Comments 31 pages, 4 figures, 3 tables. Introduces Calibration Drift Under Reasoning (CDUR) with theoretical analysis and preliminary experiments; includes CABStop; code and data available