PRISM: A Multi-Dimensional Benchmark for Evaluating LLM Peer Reviewers 事件

PRISM: A Multi-Dimensional Benchmark for Evaluating LLM Peer Reviewers arXiv:2605.26730v1 Announce Type: new Abstract: The rapid growth in submissions to machine learning venues has strained the scientific peer-review system and intensified interest in LLM-based automated peer reviewers. However, how good these systems are actually, especially compared to human reviewers at catching scientific gaps, remains poorly understood. In this work, we introduce PRISM (Peer Review Intelligence via Struct