Publishing a paper a week is a bad thing, actually
Faced with an AI slopocalypse, a key site researchers use to share their work capped submissions to 2 per month
During one of my more intense language-learning phases, I used an app called Slowly to send digital letters to language-exchange penpals around the world. And they were digital letters, not emails. That's because, on Slowly, letters take time to travel from inbox to inbox — and the further away your penpal lives, the longer it takes for the letter to arrive.
Slowly is basically just email with cute stamps and cartoon avatars. But its artificial time delay is a difference that makes all the difference. I found myself thinking quite a lot about what I wanted to say in each letter, knowing that I couldn't quickly follow-up to clarify a point or elaborate. I took care to say something worth waiting for, and I looked forward to checking my "mail." Writing to penpals became something I set aside time for and handled in one sitting, rather than a little here and a little there throughout the day. And I experienced nothing like the distracted feeling of having too many mental "open tabs" that plagues me whenever I'm stuck in a back-and-forth via email or instant messaging. Perhaps most importantly, I didn't get too many letters. Even intense correspondence was limited to something like one letter every few days.
I bring up Slowly because a key piece of scientific infrastructure is forcing researchers to slow down. Several weeks ago, an important site that scientists use to share their research papers online — the arXiv preprint server — decided to artificially limit the pace of submissions. Researchers can now submit at most two papers per month to the arXiv. Several journals and conferences now have variations on similar policies. The Center for Open Science's OSF Preprints server paused submissions indefinitely in August 2025. I assume most people reading this newsletter like scientific progress and don't want it to slow down. But scientific publication is not the same thing as progress; might putting on the brakes be a good thing?
The arXiv slopocalypse
The arXiv submission cap is, explicitly, a response to an tsunami of AI-generated research. Over the past several years, the arXiv has been absolutely swamped with low-quality submissions, including AI slop. September 2026 hit a record-breaking 40,363 submissions — compare that to the 17,271 papers submitted roughly 4 years ago in November 2022, the month ChatGPT was released. The surge in monthly submissions since ChatGPT outstrips the entire pre-ChatGPT increase since the arXiv went online in 1991. The arXiv can't deal with the deluge. It relies on about 300 human volunteer moderators to ensure that submissions meet minimum quality standards, and they're overwhelmed.
According to an arXiv blog post announcing the change, putting on the breaks is meant to be a temporary measure. The arXiv needs time to "determine what may be the new best practice for authors employing more and more advanced AI tools, and to give us time to improve our moderation tools and procedures accordingly." From the outside, it seems to me like this sweeping cap on submissions — across all disciplines, not just longtime problem child computer science — was a bit of a last resort. Over the past several years, the arXiv has introduced several increasingly tight restrictions in an attempt to close the floodgates:
This is not a paywall!
Subscribe for free to keep reading.
