arXiv might not survive AI slop onslaught, warn mathematicians
Mathematicians warn that pre-print service arXiv is so flooded with AI slop papers that it has become virtually unusable – but the service has struck back at "low-value" submissions with a new limit for uploaders
The preprint database, arXiv, is grappling with an overwhelming influx of AI-generated papers, threatening its relevance and survival. Academics are voicing concerns that the surge of AI-generated papers is clogging the system and rendering it useless. ArXiv, a platform for academics to host pre-publication papers in various fields, such as physics, astronomy, and economics, is facing unprecedented challenges.
The number of submissions has skyrocketed, from 20,569 in September 2024 to 40,363 in September 2026. This dramatic increase in submissions has led to difficulties for moderators who are struggling to keep up with the volume. Kevin Buzzard, a mathematician at Imperial College London, has had to rely on AI to help him identify papers worth reviewing.
The quality of AI-generated papers is often criticized for lacking clear explanations, which hinders human understanding. While non-professional mathematicians can now solve complex problems and automatically generate papers using AI, many of these papers are considered low-value research. In response to the issue, arXiv has implemented stricter limits on submissions, allowing only two papers per calendar month and a maximum of three active submissions at any given time.
Despite permitting the use of AI, arXiv has witnessed an increase in thin papers with narrow scope, forcing the platform to find new measures to maintain quality. Alternative solutions, such as the Hexagon project, are being developed to accommodate the new reality of AI-enabled research.
Written by urgent.news from New Scientist's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
