Large language models (LLMs) are increasingly evaluated for fairness, yet most existing benchmarks audit one sensitive attribute at a time. This single-attri…
机构：中科大
来源：arXiv 2610.35069 | AI4Papers 论文推荐平台