For decades we have been told that mathematical research is a noble activity: one sits quietly, thinks deeply, struggles with beautiful structures, proves theorems, and occasionally writes them down.

Then universities discovered bibliometric indicators. Suddenly, the quiet mathematician became a small publishing machine. Papers were no longer just the natural outcome of research; they became measurable units of academic existence. Funding? Count papers. Hiring? Count papers. Promotion? Count papers. Scientific value? Well, ideally yes - but first, count papers.

The result has been predictable. The number of publications has grown enormously, in spite of quality. Everyone publishes more, everyone reads less, and everyone feels guilty about not keeping up with a literature that is expanding faster and faster. Publishers, in their side, have discovered that academic anxiety is a renewable resource. And now come Large Language Models.

At first, people imagined that LLMs would mainly help with coding, applied mathematics, or polishing English. But this is only the beginning. They will also help formulate conjectures, explore examples, organise arguments, compare theories, and perhaps even assist in developing new mathematical frameworks. In other words: they will not merely help us write more papers. They will help us produce many more plausible reasons to write more papers.

So, if nothing changes, the future of mathematics may consist of millions of perfectly formatted papers that nobody has time to read, written with the help of machines, reviewed by exhausted humans, and evaluated by indices that were already dubious before the machines arrived.

This is precisely why the system will have to change. Bibliometric indicators depend on scarcity, slowness, and a certain correlation between number of papers and amount of research. Once the production of papers becomes cheap, fast, assisted, fragmented, and potentially exponential, that correlation collapses. Counting papers will become as informative as counting emails.

Peer review will also have to change. Editorial systems cannot survive if they are asked to process every minor observation, every technical lemma, every small but publishable result, and every AI-assisted micro-contribution as if each were a self-contained monument to human thought.

And here is the genuinely good news. Mathematics may finally be forced to return to something healthier. We may publish fewer papers, but more substantial ones: works that are significant, organic, readable, and worth preserving. Smaller results will still matter, but they may circulate differently: in notes, repositories, seminar talks, collaborative documents, curated databases, or other forms that do not pretend that every useful observation must become a journal article. Researchers, meanwhile, may be liberated from the strange duty of converting every tiny step into a separate publication merely to prove that they are alive.

In this sense, the end of publish or perish may not come because universities suddenly become wise. That would be too optimistic, even for a mathematician. It may come because the system will become impossible to maintain.

And mathematics, as usual, will develop new perspectives.

So, after all, maybe it is good news.

Comments and discussion are welcome on Mathstodon