BAAI 与人大合作的科研文献检索基准被 NeurIPS 2026 接收
精选理由
BAAI 和人大做了个测 AI 读论文能力的基准，被 NeurIPS 2026 收了，数据和评测环境全开源，做科研智能体的可以直接拿去用。
BAAI 与中国人民大学联合开发的科研文献检索基准被 NeurIPS 2026 接收。该基准聚焦三个瓶颈：概念理解、全文验证和开放式检索，用于暴露当前 AI 智能体在文献发现中的短板。配套的数据集、评测套件和 DeepXiv 环境已全部开源。
原文 · 智源研究院 BAAI
Huge thanks to @huangyun_122 for the great breakdown!
Excited to share that this joint work with RUC was recently accepted to #NeurIPS2026.
The benchmark tackles key bottlenecks in scientific literature discovery—exposing where current AI agents fall short in concept understanding, full-text verification, and open-ended search.
📦The dataset, evaluation suite, and DeepXiv environment are fully open-sourced. Looking forward to seeing future breakthroughs toward truly capable AI Scientists!