PaperBench: Evaluating AI’s Ability to Replicate AI Research

Publicación automática · Texto en el idioma de la fuente. We introduce PaperBench, a benchmark evaluating the ability of AI agents to replicate state-of-the-art AI research.