Submissions to preprint server keep rising despite author cap

The number of papers submitted to arXiv, the world’s largest preprint server, has continued to rise despite the introduction of a cap intended to curb a surge in artificial intelligence “slop”.

In October, arXiv received a record 42,152 research paper submissions, according to a new analysis. The figures come a month after arXiv limited individual authors to two submissions per month following an unmanageable 40,363 submissions in September.

The knowledge-sharing platform does not conduct traditional peer review but it does carry out basic checks through a small team of about 170 voluntary moderators.

The latest unpublished analysis, shared with Times Higher Education, was authored by Originality.ai founder Jonathan Gillham as part of his new “Maths Slop” tracker series.

When the cap was introduced, experts predicted that some would find a way around it by asking co-authors to submit papers on their behalf.

Gillham said his analysis appeared to show that this workaround was already happening, explaining that there appeared to have been a “shift towards sharing submission duties”.

He has also identified that there were 9,465 new mathematics papers in October, a 6.5 per cent increase compared with September.

The analysis showed that the academic listed as co-author on the highest number of papers, at 19 submitted by 17 individuals, was prominent Chinese mathematician Juncheng Wei.

Wei, professor of mathematics at the Chinese University of Hong Kong, toldTimes Higher Education that all the papers “arose from my ongoing projects that began before October”, adding that the arXiv submission date “does not indicate when the research was done or how long it took”.

He stressed that several collaborations he was part of happened to finish at the same time. “I made substantial mathematical contributions to each paper and stand by my work. I do not personally use AI tools in my mathematical research. Some of my collaborators use AI, and this is acknowledged in the relevant papers.”

Wei added that he has been active in mathematical research for three decades and has published 570 papers with more than 200 collaborators over that period.

“Being a co-author on several papers does not by itself show that someone is working around a submission cap,” he said.

The co-author who featured on the second largest number of papers, Alexander Gasnikov, rector of Innopolis University, was named on 11 arXiv submissions in October, which were posted by 8 different scholars.

Gasnikov, head of the private university in the Russian republic of Tatarstan, said the influx was because of the breakthroughs he and his research team had made as a result of the advent of new AI tools.

This summer OpenAI released its GPT Astra model and Anthropic rolled out its Claude Fable model, allowing Gasnikov and his PhD students to solve “a number of complex, unsolved problems”.

“We developed a system that leverages AI agents…to assist in writing research papers,” he explained. They also used “other agents to conduct independent peer reviews” and repeated this cycle until the review reached a “sufficiently high standard” with the published work also subjected to a “rigorous human review process”.

frances.jones@timeshighereducation.com