国际学习表征会议（ICLR）2027年发布新规定，规定每位作者最多投稿20篇论文，超出配额的论文将被随机拒绝[1]。这一举措旨在应对投稿量的快速增长——ICLR 2026年收到19525篇有效投稿，相比2025年约11603篇增长了68%[1]。
Google DeepMind研究员Dan Roy以反讽方式对这一限制提出质疑[1]。他在X平台发文指出，AI已深入论文写作和评审的全流程，质疑仅限制投稿数量是否能有效解决问题[1]。Roy讽刺地表示："AI Agent已能完成研究工作，应鼓励尽可能多地提交论文，所有评审工作都交给大语言模型处理了……承认吧！"[1]
Roy的批评有数据支撑[1]。根据评审员Graham Neubig对ICLR 2026年约75800条评审意见的检测，约21%被高度怀疑为AI完全生成[1]。该年度共涉及18054名评审提供了76139份审稿意见[1]。
The International Conference on Learning Representations (ICLR) has introduced a new policy for 2027 limiting each author to a maximum of 20 paper submissions, with excess papers subject to random rejection [1]. The measure comes in response to surging submission volumes, with ICLR 2026 receiving 19,525 valid submissions, representing a 68% increase compared to approximately 11,603 submissions in 2025 [1]. The conference processed input from 18,054 reviewers who provided 76,139 review opinions [1].
However, the new rule has prompted criticism from Google DeepMind researcher Dan Roy, who challenged the policy through sarcastic commentary on platform X [1]. Roy questioned whether capping submission numbers addresses the underlying problems, noting that artificial intelligence has already permeated the entire workflow of paper writing and peer review [1]. According to analysis by Graham Neubig of approximately 75,800 review opinions, roughly 21% were highly suspected to be entirely AI-generated [1]. In his pointed remarks, Roy sarcastically suggested: "AI agents can now complete research work, so we should encourage submitting as many papers as possible, with all review work delegated to large language models... Let's admit it!" [1]
