From the article:

Kai Shaikh, a Ph.D. student in mathematics at the University of Toronto, also currently at SLMath, observed that theorems are easy to use as benchmarks, but that they capture little of the wonderment at the heart of math. “Just like a photograph of a mile marker can’t provide much of the value of actually going on a hike,” he said, “it would seem to me that proofs without stories, or even stories without struggle or mystery, and first glimpses of strange beasts, are not very meaningful or valuable to us humans.”

And:

Initially, the rumors about these OpenAI results had it that there were solutions to 400 problems, and now the rumors are that further dumps are arriving soon. The number 400 had made Mr. Zhang think of “The 400 Blows,” a coming-of-age film by François Truffaut.

“The film is about a boy who is evaluated, judged and shut away from society, and no one really takes the time to understand him,” Mr. Zhang said. “Towards the end, he runs to the sea, and the film ends on a freeze frame that passes no judgment on whether this is an escape or a dead end. I don’t think that’s a bad picture of where mathematics stands with these results.”

The Daily interviewed mathematician and author Steven Strogatz about When A.I. Comes for Your Purpose in Life.

High level math can feel more like art than science. My undergrad is actually a bachelor of arts in mathematics. AI completing a proof of a problem that you have been working on would feel like someone publicly releasing a novel, movie, or symphony that you have been drafting for maybe years. 

And from my amateur review of some of the papers and from commentary I have seen (sorry no links at hand) I think these papers from OpenAI will need a lot of human work to make sensible. They may be correct. But they may also read like someone fed the dailies from a movie into a LLM and asked for a finished movie. It might technically tell a story but it will likely not be a critical or popular success. But you can imagine how the cast and crew would feel betrayed and wonder what their effort was for.  

To be clear: it’s not as if any noob prompted ChatGPT saying “go solve this”. This is still heavily guided by a bunch of mathematicians guiding it and nudging it. Obviously still impressive. But it shows that it’s rather an impressive tool for rather than a replacement for mathematicians. 

It's hard to overstate how significant these results are and the impact they'll have on mathematics and the practice of mathematics.

My understanding - based on the limited information provided by OpenAI - is that relatively little human guidance was provided. And, this matches the "taking out the trash" vibe with which the release happened; it seems like OpenAI wanted to test their internal model, and newly discovered maths is a byproduct rather than the purpose of their testing.

Some responses from mathematicians are collected here: https://www.futilitycloset.com/2026/09/26/summary-advice/

For the area that I know most about, self-avoiding walks, the AI model was able to make dramatic and (to me) unexpected progress, and it just appears as #237 on the list: https://github.com/openai/math/blob/main/CONTENTS.md

OK, I don't know why the link I tried to provide got garbled (probably my fault) or why I can't edit it (possibly because it was my first comment and I'm now flagged as a bot), but here's the correct link to comments from mathematicians on the OpenAI release: https://proofsandprompts.com/2026/10/08/100-reactions-to-100-solutions/

Vi Hart's Big Navier-Stokes Post.

All that is just to explain why, when I saw OpenAI's announcement of a Navier-Stokes Lean-verified proof, I immediately knew with direct and complete surety that their result required years of work from a huge community of university mathematicians, students, and hobbyist volunteers that have been formalizing mathematics into Lean for over half a decade now.

As a former physics researcher myself, I read this and just felt sad. The quotes from mathematicians saying that the AI has scaled mountains, but there are higher mountains ahead, just feels like a desperate attempt to spin the facts. The joy of research, the second pleasure, is finding something new. The utility of the new is (often) only a secondary concern. The goal isn't to know the answer, it's the process of solving the problem. This result from openAI, at a stroke, devalues the field of research by equating it to a list of ticks. And as is increasingly common, the creativity has been obliterated by the AI, with the grind of fact checking and validation left to the humans. What is our purpose indeed....

The shift to AI parallels what 3D modeling and photoshop software did to hand drawing in landscape and architectural rendering. When first out of school, art theory classes and hand drawing workshops taught a lot about color theory, composition and sketching.

Now that 3D modeling is the basis for rendering, much of that is lost, because the repetition and observation skills gained from sketching and creating by hand is no longer a part of the process. AI is changing this again as another shift away from the human craft and a heavier reliance on technical computer based tools. Like identifying plants through observation and memory in place of an app, there is a more layered approach to thinking/designing being lost. 

It must be terribly frustrating to have worked on a problem for a good chunk of time, believing you are toiling in obscurity, but with the goal of discovering something few minds could, only to have that discovery made by a machine and announced offhandedly; but the minds who set themself to these tasks are special, and hopefully can be reset, and tasked to work on even bolder problems. Would AI have come up with relativity, or natural selection, or Feynman diagrams? Perhaps maths will evolve into something even greater when these minds are freed. 

Another interesting item is that nearly all of this (372 total) was completed with one agent and a single prompt (https://www.scientificamerican.com/article/openai-unleashes-hundreds-more-math-results-upon-a-field-already-in-shock/).  This is a considerable contrast to the Navier-Stokes solution which required thousands of agents cost millions in tokens.  Clearly they saw the splash they made there and set out to do it again.  All in the name of marketing.  Not to provide an elegant answer.  Not to help humanity.  Just marketshare.  Why do I feel I am going to be greeted with "Resistance is futile" pretty soon?