The “math vs. AI” wars are continuing to heat up. I predict things will become more intense in the coming months. OpenAI and Anthropic will likely release new proofs, possibly including solutions to remaining Millennium Prize Problems or another batch of unsolved problems. This will lead to a new round of intense debate.
In a viral article this week, Derek Thompson published “I Know Why the Mathematicians Weep“. The tweet below quoting mathematician Terence Tao also went viral. He proclaims that “we’re seeing the end of the romantic era of mathematics, the age of heroes. we’re moving into an industrial era where math becomes large-scale, done in big teams like the other sciences”.
Mathematician Terence Tao:
"we’re seeing the end of the romantic era of mathematics, the age of heroes. we’re moving into an industrial era where math becomes large-scale, done in big teams like the other sciences"
We’re entering, rather unwillingly, a new era where exciting new things are possible, but some things are being lost
— Haider. (@haider1) October 15, 2026
This agrees with my post last week, where I said:
So why sadness instead of joy? Math operates off of credit: names are inseparably attached to results more so than other areas of research. “Curing cancer” is much more of a team effort compared to proving or disproving some long-standing conjecture. Physics labs are huge collaborative processes, compared to solitary math endeavors. There are fewer opportunities for lone geniuses to stand out in other areas of research, due to costs and other factors (good luck building a home physics lab). Turning math into just another collaborative or lab-driven field means that the last holdout for lone geniuses goes away.
This is mainly about status. No one cares that much about AI automating drug discovery or coding, but it’s bad when it assists in math. Math results are attached to individuals, whereas cancer drugs or other medical discoveries are more of a team effort by a lab. AI now has the effect of turning math into something more lab-like and less about the individual.
This is the fear, but I predict that individuals will still retain credit for finding results, even with AI. Whenever there is a new result, the announcement says “so-and-so, with the help of AI, found such-and-such result,” unless the work was explicitly done by a lab, in which case OpenAI or Anthropic gets the credit. Moreover, results produced by these labs will remain a minority of overall mathematics. Math is too vast for any entity, even OpenAI or Anthropic, to have a monopoly on it. It’s also not profitable. These discoveries demonstrate the capabilities of the AI software and generate free marketing through media coverage and everyone on Twitter talking about them, in the hope that firms and individuals are impressed enough to pay up.
A common argument is that AI-assisted proofs are too hard to follow. This is an excuse for not having to read the papers or for dismissing the results. Since when were math papers involving new, difficult, abstract concepts ever easy to read? Has there ever been a time when cutting-edge math was easy to follow? For example, the proof of Fermat’s Last Theorem in the early ’90s was regarded as so difficult that only a couple dozen of the leading experts could follow it in its entirety. The “math community” didn’t just refuse to review Wiles’s paper because it was “too hard to follow”. Rather, they buckled-down and went through it, line by line.
Here’s a screenshot of a paper from 40 years ago:
Tell me how this is easier to follow than any AI proof? At least with AI, you can download the LaTeX, upload the paper to your favorite LLM, and have it explain the paper at whatever level of granularity you like. That wasn’t possible even two years ago. I recall a number theorist’s paper being hard to follow—it took days for me to read a page. This was in early 2026. Now, in just the last six months, AI has gotten good enough that I can just upload all of it and have it decipher everything. What would have taken hours of reading can now be summarized in minutes.
Math journals commonly reject papers as “out of scope” or “not of sufficient interest” without a referee actually reading them. So it’s not as if the “math community” was always checking these papers for accuracy in the first place. The surplus of AI papers will simply mean more rejections by editors before the refereeing stage. If journals and experts can’t be bothered to read the papers, why should authors bother to write them without AI? Writing a paper is a huge time investment, and and AI helps at both stages: finding novel results and writing the paper itself.
Overall, the blame goes both ways. Although mathematicians may be the victims, some also share blame for creating the broken system that exists today, in which the use of AI has become necessary for new researchers to have any hope of producing papers of “sufficient novelty”. Anyone who doesn’t use AI is at a huge disadvantage to those who do. So, I don’t “weep” for the mathematicians, as many of them share in the blame for creating a math culture that values novelty or difficulty above all. [Again, this does not apply to all mathematicians, as many are neutral or even support AI.]
It’s econ 101: people respond to incentives. Demand novelty above all, and people will find ways to do that, including using AI. So, there is no going back: AI-assisted math is here to stay. But wanting the status that comes with tenure and the fame of having discovered results and published papers in the easier, less competitive pre-AI era–while refusing to accept or acknowledge the legitimacy of AI-assisted papers–amounts to pulling up the ladder for everyone else.