AI coach provides constructive feedback, turning vague reviews into detailed, actionable suggestions. The tool reduces unprofessional tone, eliminating personal attacks and factual inaccuracies. Researchers used a curated set of 12.9% flagged poor reviews to train the LLM. Review Feedback Agent employs five collaborating LLMs to cross-check and refine responses. Pilot deployment at 2025 ICLR conference with over 10,000 submissions. Impact on overall paper quality remains to be evaluated, but early results are promising.
Article Summaries:
- Researchers at Stanford have developed a “Review Feedback Agent” that uses five large language models to help peer reviewers write more constructive, less toxic comments. The system was trained on a curated set of vague, unprofessional, or incorrect reviews and then tested on about 20,000 existing reviews for the 2025 International Conference on Learning Representations. Reviewers received AI‑generated suggestions to make feedback more specific and actionable. While the tool consistently recommends clearer, more polite language, the study notes that its impact on the overall quality of the papers being reviewed remains to be determined.
Sources: