PapersSelected17:49Chinese contentVAKRA:在工具使用策略下评估跨API与检索的多跳推理IBM新基准VAKRA,测API和文档多跳推理,最强70.4%,多跳掉一半,暴露AI短板。#VAKRA#IBM#智能体#多跳推理aarXiv cs.AI@Ankita Rajaram Naik et al. (9)SourceRead laterReadWorth followingUsefulFollow VAKRA