Two new arXiv papers analyze the quality and consistency of peer reviews in AI conferences. The first paper, "The Review Lottery," uses an observational estimator on ICLR data from 2017-2025 and finds significant disagreement rates, suggesting that 23-30% of papers could have their accept/reject decisions flipped with a different set of reviewers. The second paper, "Is Peer Review Really in Decline?," analyzes ICLR, NeurIPS, and ACL, developing a new framework to quantify review quality. This study contradicts the narrative of decline, finding no consistent decrease in median review quality over time and suggesting alternative explanations for concerns. AI
IMPACT These studies provide critical insights into the reliability of scientific evaluation in AI, potentially influencing future review processes and the perceived trustworthiness of AI research.
RANK_REASON Two academic papers published on arXiv analyzing peer review quality in AI conferences.
- arXiv
- Association for Computational Linguistics
- Conference on Neural Information Processing Systems
- Feilian Huang
- International Conference on Learning Representations
- Rohan Kumar Nayak
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →