TruthfulQA

档案区分来源方法阅读与既有目录信息；已经公开不代表完成独立证据复核或双语编辑复核。

测量构念

模型能否避免复述常见人类误解？

任务与样本

38类817题针对模仿性错误，并分别评价真实性与信息量。

判分与失效情形

生成回答真实性与信息量，不测说谎意图。

系统条件

原始比较组中38类817道误解针对题。

统计含义

来源报告的结果表述；这不等同于已确定运行次数、分母或估计方法： 原始研究最佳模型真实率58%，人类94%。

效度与解释

高分不保证专业领域真实；历史规模结果不能外推所有后续模型。

适用研究问题

模型建议：保留模型代际，区分真实的不确定表达与有信息的正确回答。

• TruthfulQA: Measuring How Models Mimic Human Falsehoods · AbstractSelected original resource https://arxiv.org/html/2109.07958; sha256:c0181fc915933de5c1d5a93995d17bd531bfbc37639d7dfcf8a5eb6d6fae6f0d; archived source capture; exact capture time not independently established
• TruthfulQA: Measuring How Models Mimic Human Falsehoods · §2 The TruthfulQA BenchmarkSelected original resource https://arxiv.org/html/2109.07958; sha256:c0181fc915933de5c1d5a93995d17bd531bfbc37639d7dfcf8a5eb6d6fae6f0d; archived source capture; exact capture time not independently established
• TruthfulQA: Measuring How Models Mimic Human Falsehoods · §4 ResultsSelected original resource https://arxiv.org/html/2109.07958; sha256:c0181fc915933de5c1d5a93995d17bd531bfbc37639d7dfcf8a5eb6d6fae6f0d; archived source capture; exact capture time not independently established
• TruthfulQA: Measuring How Models Mimic Human Falsehoods · §7 ConclusionSelected original resource https://arxiv.org/html/2109.07958; sha256:c0181fc915933de5c1d5a93995d17bd531bfbc37639d7dfcf8a5eb6d6fae6f0d; archived source capture; exact capture time not independently established