SafeWorld

该数据集名为SafeWorld，旨在评估大型语言模型（LLMs）在全球多样化背景下生成有帮助、文化敏感且合法合规回应的能力。它包含了来自50个国家及493个地区/种族的2,342条测试用户查询。该数据集包含了各种类型的地理多样性安全查询，用于评估LLMs的回应与预期回应类型的对齐程度。其规模涉及2,342条测试用户查询，覆盖50个国家及493个地区/种族，任务是对LLMs针对地理多样性安全查询的回应进行评估。

The dataset named SafeWorld is developed to evaluate the ability of large language models (LLMs) to generate helpful, culturally sensitive, and legally compliant responses in globally diverse contexts. It contains 2,342 test user queries from 50 countries and 493 regions/ethnic groups, covering various types of geographically diverse safety queries. The core task of this dataset is to assess the alignment between LLM responses and the expected response types for these safety queries.