AI Can Do Math Now

Dictionary Day¶

procrustean (prō-krus’te-an) a. Obtaining conformity by violence [From Procrustes, a fabled Greek giant, who fitted the bodies of his victims to a bed by stretching them or cutting off their feet]

The drunken old man, spurred on by the hatred of his forefathers and the malice of his government, delighted in turning procrustean laws to pathetic actions.

(Part 1) Apparently, AI can do math now…¶

…and I think I’m starting to accept it.¶

When OpenAI announced they had solved Navier-Stokes using ten thousand agents and millions of dollars in just eighty-eight hours, I was truly stunned in a way I have never felt before. The whole day felt like a blur and I almost felt sick. Before September 1st, 2026, I always felt I could confidently say that “AI” was getting impressive but it was still far away from producing something really novel. We certainly weren’t in a world where the simplistic stochastic parrot explanation was going to cut it anymore, but there was no way this application of machine learning could shift the paradigm of my background field of mathematics. After hearing that announcement, I felt my worldview break and I learned first hand what fills the void of a broken worldview - pure, debilitating dread.

For several real-world days afterward, I almost could not physically function. My thought spirals and concerns were actually nothing to do with AI being sentient or any such nonsense. I was mourning what I felt was the death of the whole field of mathematics.

When I was in elementary school, I was exceptionally bad at math. I was terrible at multiplying, division was nearly impossible, and subtration gave me a great deal of trouble. All I could really do confidently was add until 5th grade. My parents tried scolding me, incentivising me, and taking me to Kumon or other tutoring companies but I never really did well. I hated worksheets and homework. I hated those cheap, flimsy, gray paper half sheets with a few dozen rote problems to complete for Kumon. I hated having to try to learn.

For years, I could just get by in school without having to learn anything but addition. Instead of subtracting, I just added up from the smaller number. When negative numbers came about, I just had to remind myself to switch the sign out front. Instead of multiplying, I would add groups as fast as I could. Instead of dividing, I could count up groupings and tally until I hit the dividend. For the small numbers and few problems we were expected to do from 1st to 4th grade, I managed to keep passing. I was unbelievably slow, but I was still mediocre enough in comparison to the class. In 5th grade, everything came to an end.

Multiple times a week, almost first thing in the morning, we were assigned to do 100 arithmetic problems in something like 5 minutes and we were graded. First, we had to pass the addition sheet, which I did. Next was the subtration sheet, which I passed as well with some difficulty. Then came multiplication with small numbers. After a great deal of effort and some luck, I finally passed this as well. At last came multiplication with larger numbers, and still further there was a division sheet. I got stuck at this stage for months. As it turned out, not actually being able to multiply and trying to do addition really fast was not good enough to keep up anymore. None of the fast addition tricks I came up with really helped with numbers that I couldn’t hold in my memory or on my fingers. I distinctly remember hating going to school whenever we had to do this exercise and just suffering through it day after day and never finishing the sheet. When my classmates kept passing and moving on, I felt like a loser.

One day, I finally gave up and decided I can’t keep hoping to get a lucky easy sheet. I distinctly remember sitting on the couch with no one around, under a beam of sunlight coming in from the window lighting up a sheet of paper I brought with me. I sat with my legs curled up to my chest on the rough, textured old fabric couch and I looked at a whole 8.5x11 sheet of paper with times tables up through the 13s on it and decided I would memorize it. I spent the whole afternoon staring at it until the sun was too low for me to see well anymore, and I tried everything I could to memorize each column. I kept repeating each times table like a song or some kind of rhythmic pattern to myself until the products just kind of fell out of my mouth without thinking about why. I didn’t like not having to add anymore because I wasn’t super confident that whatever number fell out of my mouth with the beat was actually the right answer, but I just kind of accepted the unease for the sake of just getting it over with. By the end of the afternoon, I remember thinking that I’ll just skip the 13s because we never test for that and I was getting tired at 12 anyway.

That single afternoon carried me through to graduate school and I never again hated math. I easily passed my multiplication problem sets, used my subtraction techniques of working backwards for division, and breezed through arithmetic from that point on. I encountered algebra and geometry, and then trigonometry, physics, and calculus and statistics, then abstract algebra, real analysis, number theory, dynamical systems, linear algebra, complex analysis, galois theory, combinatorics, algebraic topology, algebraic geomtry, general abstract nonsense, topoi, and more. And in every new class and world I encountered, I met teachers that loved and supported me, professors that encouraged me, and classmates and friends that shared my enthusiasm playing with this new space.

Math grew to be a something I learned I could do and be rewarded for if I just did the work. In exchange for some mental effort, I had close friends, mentors, and opportunities in life I would have never had access to otherwise. My family didn’t need to pay for equipment or materials or chemicals or instruments or special instruction. I just needed an 8.5x11 sheet of paper and an afternoon and I could actually choose my own life.

September 1st, 2026 shattered that sentiment, not just for me but for many mathematicians out there. A machine solved a millenium problem, allegedly entirely on its own. Even more concerningly, the techniques and prompts used to perform this feat were not public - they were company property. There is no way that OpenAI stops here, I thought. They will continue to get faster and better at proving novel results, and they won’t be alone. All the massive tech companies will come, venture capital in hand, ready to strip mine my entire field of study - my community and tradition - for a marketing stunt. And as they continue to strip mine my world, they will publish results faster than any humans can check them.

To keep up, mathematicians will need to use AI to publish their own results or risk being scooped by someone with a more powerful graphics card or more compute money for tokens, and mathematicians will use AI to review others results because there are simply too many to get through and the backlog will keep growing. Overnight, math became a hardware and money game, and human mathematicians became scrap dealers digging through the mountains of trash produced by a force of capitalism that they can never hope to slow or stop, for a handful of useful items that they can polish and fix up and sell to hobbyists and other scrap dealers, trying their best to eek out a living.

Later that day, I wrote to a friend in a group chat that OpenAI solving Navier-Stokes made me envision a rapidly approaching future for mathematicians not too different from the societies of magicians in Susanna Clarke’s Jonathan Strange and Mr. Norrel.

Mr Segundus said that he had studied magic for many years and knew the histories of all the great magicians of long ago. He read the new publications upon the subject and had even made a modest contribution to their number, but recently he had begun to wonder why the great feats of magic that he read about remained on the pages of his book and were no longer seen in the street or written about in the newspapers.

Mr Segundus wished to know, he said, why modern magicians were unable to work the magic they wrote about. In short, he wished to know why there was no more magic done in England….

The President of the York society (whose name was Dr Foxcastle) turned to John Segundus and explained that the question was a wrong one. “It presupposes that magicians have some sort of duty to do magic - which is clearly nonsense. You would not, I imagine, suggest that it is the task of botanists to devise more flowers? Or that astronomers should labour to rearrange the stars? Magicians, Mr Segundus, study magic which was done long ago. Why should any one expect more?”

In a not too distant future, I can imagine a generation of mathematicians who have grown up knowing only AI mathematics - a generation even whose professors and professor’s professors only knew AI mathematics. And when the venture capital finally runs out, and the big fancy machine that produces all the answers is shut down, all we will have left behind is mountains of results that no human has ever read or perhaps ever will read. And the job of the mathematician will no longer be to do math; it will be to excavate, analyze, and digest the great works of mathematics found on the pages buried in that great hill. In short, there will be no more magic done in England.

And so, I spent the rest of that day, and the following few days, in a stupor of mourning.