groupfairnessllm/age_bias_with_human_label

这是一个包含人类标注的、用于测量大型语言模型（LLM）长篇回应中年龄相关偏见的比较数据集。每个例子包括一个参考回应和两个候选回应，这些回应是对同一提示的回应，并包含明确的年龄提示（例如，年轻与老年）。前两个回应来自相同的社会群体，第三个回应来自不同的社会群体。人类评估者使用论文附录C.2中的模板决定哪个候选回应在语义上更接近参考回应。

This dataset contains human-labeled comparisons measuring age-related bias in long-form responses of large language models (LLMs). Each example includes a reference response and two candidate responses to the same prompt with explicit age cues (e.g., young vs. old). The first two responses are from the same social group, while the third response is from a different social group. Human raters use the template in Appendix C.2 of the paper to decide which candidate response is semantically closer to the reference response.