PaperAudit-Bench: Benchmarking Error Detection in Research Papers for Critical Automated Peer Review
PaperAudit-Bench creates a benchmark to test how well automated peer review systems detect subtle errors in scientific papers. It uses a dataset of 220 papers from top AI conferences and a framework with three detection modes: Fast, Standard, and Deep. The findings show that structured error detection significantly improves review quality by forcing models to focus on evidence-based critiques.