Compared to What? Baselines and Metrics for Counterfactual Prompting
与什么相比?反事实提示的基线和度量标准
机构 * Northeastern University(东北大学) ; Bar-Ilan University(巴伊兰大学) ; Allen Institute for AI(人工智能研究所)
专题命中 推理评测 :CoT(abstract,abstract_cn);分类 cs.CL、cs.LG
AI总结 本文探讨了反事实提示中基线和度量标准的重要性,指出单纯改变文本因素无法准确评估模型敏感性,提出通过统计检验比较干预差异与改写影响的框架,发现模型对患者性别等的敏感性在考虑一般模型敏感性后显著降低。
Comments Published as a conference paper at COLM 2026. 33 pages, 10 figures, 18 tables. Code: this https URL (https://github.com/redagavin/counterfactual-prompting-baselines); Python package (cfprompt): this https URL (https://github.com/redagavin/cfprompt)