2% of ICML papers desk rejected because the authors used LLM in their reviews

Editor's take

The International Conference on Machine Learning (ICML) has desk-rejected a small but notable percentage of submitted papers due to authors using large language models (LLMs) to generate their reviews. This action highlights a growing tension between the rapid advancement of generative AI tools and the established academic integrity processes within AI research.

This development matters because it signals the challenges faced by peer-review systems in adapting to AI-driven content generation. The concern isn't just about academic dishonesty, but also about potentially diluting the quality and rigor of research evaluation, impacting the credibility of top-tier AI conferences like NeurIPS and ICLR, which are facing similar pressures.

Future watchpoints include whether other major AI conferences adopt similar strict policies, and how the community will develop robust methods for detecting and addressing LLM-generated content in reviews, beyond simple detection tools. The long-term impact on the perceived fairness and accuracy of AI research peer review remains a critical question.