The International Conference on Machine Learning (ICML), a prominent event in the AI field, is addressing the misuse of language models in paper reviews. Some authors have been embedding hidden prompts in their submissions, designed to manipulate language models into providing favorable evaluations. This tactic arose as reviewers increasingly relied on language models for assessments, rather than conducting thorough personal reviews. ICML has condemned this practice as scientific misconduct, emphasizing that such actions undermine the integrity of peer review. The organization has updated its ethical guidelines to explicitly prohibit these hidden prompts, drawing parallels to unethical practices like bribery. The issue highlights a growing challenge in the AI community, as both the misuse of AI tools and attempts to exploit them become more prevalent.
Are Researchers Misusing LLMs in AI Conference Papers for Better Reviews?
