arXiv, the preprint server most machine learning research passes through before peer review, has capped how much any one author can post. As of 1 October, submitters are limited to two submissions per calendar month, with a ceiling of three total active submissions at any one time.
The server’s own figures explain the decision. In September 2016 arXiv received 9,869 submissions. In September 2024 it received 20,569. This September it received 40,363, which generated almost 9,000 support tickets for a moderation team made up of volunteers. Submissions overall have doubled in the past two years. In the cs.AI category the rise is more than sixfold.
arXiv is blunt about the cause. Its moderators report a growth in thin papers of narrow scope, in “salami” papers where one piece of work is broken up and submitted as several smaller ones, and in what it calls a marked increase in dense, AI-written papers. The stated aims are to distribute volunteer moderator time more equitably across authors and to protect the corpus from inappropriate submissions. The full policy is set out in arXiv’s announcement.
Why it matters: arXiv is the de facto public record of AI research, and a prolific group now has to choose what goes up. It also constrains anyone treating preprint volume as evidence of productivity.