introvoyz041/publichealth-bench

PublicHealth-Bench是一个严格的公共健康领域AI基准测试数据集，旨在评估AI系统在真实公共健康挑战中的表现。该数据集包含100个问题，覆盖23个公共健康相关类别，如慢性病、传染病、环境健康、政策与经济、健康系统等。问题类型包括65个多项选择题和35个自由文本题，其中93%的问题难度为非常难。数据集基于真实案例设计，测试AI系统在因果推理、数据质量评估、多源数据合成、算法公平性、实施和批判性思维等方面的能力。该数据集仅用于评估，不包含训练集。

PublicHealth-Bench is a rigorous benchmark for evaluating AI systems on real public health challenges. The dataset contains 100 questions across 23 public health categories, including chronic disease, infectious disease, environmental health, policy & economics, and health systems. It consists of 65 multiple-choice and 35 free-text questions, with 93% rated as very hard difficulty. Based on real-world cases, the benchmark tests AI capabilities in causal inference, data quality assessment, multi-source synthesis, algorithmic fairness, implementation, and critical thinking. The dataset is evaluation-only with no training split.