Tag
This study evaluates two multimodal LLMs as peer reviewers for ICLR 2026 submissions, finding high scoring calibration but low error detection, with author identity having no effect and figures reducing error detection.