增加推理努力未改善LLM购物选择
AI 总结
一篇arXiv预印本在基于WebShop、Amazon SQID和Wayfair WANDS构建的三个商品选择基准上评测了10个前沿模型。论文报告称,提高推理努力未在统计意义上改善决策质量,而模型自报置信度显著上升。多数模型在各推理努力水平下均未实现良好校准。模型的错误也高度重合:一个模型在某项任务中出错时,其他多个模型往往也会出错,其中许多模型会选择同一错误商品。
Computer Science > Artificial Intelligence
arXiv:2611.22033 (cs)
[Submitted on 17 Nov 2026]
Title:Thinking Harder Doesn't Lead to Better Choices: Reasoning Effort Increases the Confidence but Not the Correctness of LLM Shopping Agents
Authors:Adi Krishnamoorthy, X.Y. Han
View a PDF of the paper titled Thinking Harder Doesn't Lead to Better Choices: Reasoning Effort Increases the Confidence but Not the Correctness of LLM Shopping Agents, by Adi Krishnamoorthy and 1 other authors