← back
arXivZhenran Wang, Zhonghan Bian, Jinsong Li, Zhangyang QiTue, Aug 4, 2026, 10:59 AM PDT
score 17.1

New benchmark tests AI's social world forecasting, not just task completion

Original: SocietyBench: Forecasting Counterfactual Social-World Evolution

Source: arxiv.org

Writing ELI5 summary…