OpenAI Publishes 722 Mathematics Papers at Once, Mathematicians Worry Research Will Be Flattened by AI

[[**OpenAI has released 722 mathematics papers all at once, claiming breakthroughs on difficult problems such as the Riemann Hypothesis, Catalan's constant, and the Kakeya conjecture. Although the work has not been peer-reviewed, it has shocked mathematicians and raised fears that AI will "bulldoze" their research.**

**Key points**

**1. OpenAI makes sweeping claims of mathematical breakthroughs at unprecedented speed and scale**

OpenAI released 722 papers at once, covering number theory, geometry, and theoretical computer science. It claims to have advanced the Riemann Hypothesis, proved that Catalan's constant is irrational, and given a four-dimensional proof of the Kakeya conjecture, all without peer review.

**2. As AI overtakes them, mathematicians face invalidated results and pressure on their livelihoods**

Fields Medalist Duminil-Copin said AI had bulldozed 16 of his research problems and wept. Some doctoral students have seen their thesis topics covered, putting funding, papers, and degrees at risk, while others question whether humanity wasted 30 years on the wrong direction.

**3. Mathematics becomes a muscle-flexing tool in the AI arms race**

Anthropic claimed to have disproved the Jacobian conjecture, and OpenAI rushed to publish results on the Navier-Stokes equations. What AI companies are competing for is technological leadership, users, capital, and markets; hard mathematical problems have become billboards to attract attention.

**4. AI's impact has spilled over from mathematics to most knowledge-based work**

Go, illustrators, code, copywriting, short dramas, and AI entertainers have been hit one after another. AI has taken over thinking and creation while leaving chores and other trivial matters to humans, shaking the belief that "honing your skills buys you stability." 2026-10-25 10:44 Will Wang Hong, Wei Dongyi, and their peers also be "slain" by AI? Phoenix WEEKLY Quick Read

This article is from the WeChat public account: Phoenix WEEKLY, author: Wang Dong, editor: Octopus, cover image from Visual China.

"On Tuesday night, I cried and woke up my daughter."

On October 8, French mathematician Hugo Duminil-Copin spoke about his experience at an online meeting. He is a 2022 Fields Medalist and has long stood at the pinnacle of mathematical research. What made him cry was a large batch of mathematical research results just announced by OpenAI.

Among them, 16 problems were directly related to his research.

In a subsequent radio interview, he said AI had flattened the problems he was working on like a bulldozer. In just 0.1 seconds, regular workers empathized with this top mathematician's workplace predicament. It was equivalent to this: you spend years studying and accumulating knowledge, finally find a business opportunity that could make you rich, and suddenly someone tells you:

This entire business model was implemented by AI at zero cost half a day ago.

Hugo Duminil-Copin is certainly not the only victim.

OpenAI released 722 mathematics papers at once, involving hundreds of mathematical results across number theory, geometry, theoretical computer science, and other fields.

On the famous preprint platform arXiv, a person can upload only two papers per month. To upload all these papers, you would need at least more than 300 accounts.

Some of these problems are extremely difficult and have troubled scientists for a century.

For example, the Riemann Hypothesis, proposed in 1859, has remained unresolved for 167 years. OpenAI says it has made important progress this time and given a stronger "zero-free region" result.

Whether Catalan's constant is irrational has also troubled mathematicians for more than 100 years. OpenAI says it has successfully proved that it is indeed an irrational number.

The Kakeya conjecture, which won Wang Hong the Fields Medal this year, was proposed in 1917. A two-dimensional proof was not given until 1971, and in 2025 Wang Hong gave a three-dimensional proof. Only a year later, AI has given a four-dimensional proof.

Among the results listed this time, there are many other heavyweight breakthroughs like these.

Even more devastating is that OpenAI solves these century-old mathematical problems with astonishing speed:

The average computation consumed per result is equivalent to ChatGPT Pro thinking for three hours.

Although we still cannot understand those complex formulas, ordinary people understand all too well the feeling of "I worked so hard for so many years, and suddenly I'm left behind."

For the past decade or so, we have stayed up late for exams and worked overtime; mathematicians have stayed up late for exams and proofs. Everyone has simply hoped that their efforts would buy them a stable future.

In terms of the number, breadth, and speed of breakthroughs, this is an unprecedented event in the history of mathematics.

Of course, OpenAI does not care about human feelings; it just keeps dropping bombs on the mathematics community.

The first to sound the horn was Anthropic.

On July 19, the night of the World Cup final, a researcher at Anthropic posted a tweet on X, roughly saying that they had disproved the Jacobian conjecture.

The Jacobian conjecture is no small matter; it is an important problem that has troubled mathematicians for more than 80 years.

The problem can be described in one sentence: if the Jacobian determinant of a polynomial mapping is a nonzero constant, then the mapping must have a polynomial inverse.

This problem is simple, but the answer is rather difficult. The simple puzzle once lured countless mathematicians in, and they never climbed back out.

Over the decades, mathematicians have repeatedly announced that they had solved the problem, only to have their proofs later shown to be wrong...

Many mathematicians, including the Chinese mathematician Zhang Yitang, have been left with PTSD. After tormenting mathematicians for decades, the conjecture was thus lightly dismissed as disproved in a single tweet...

Competitor OpenAI, unwilling to be outdone, also joined this mathematics "arms race."

Within a few months, the math problems the two sides claimed to have solved included but were not limited to:

These may be the most brute-force months in the history of mathematics.

But brute force is not necessarily a good thing, and many mathematicians say they are deeply worried about it.

For example, this time OpenAI produced a batch of results in operator algebras all at once, and the conclusion of a mathematician in the relevant field after reading them was:

AI is right; humans have wasted 30 years going in the wrong direction.

Some people also face more practical livelihood problems because of this.

Some found that the direction they were working on had been blown up, and they did not know where their project funding would go next; others had written half of a paper and no longer needed to write it, and whether they could get their degree was now up in the air.

On Reddit, a doctoral student who claimed to be studying operator algebras left a comment saying that he and a classmate had already solved a special case of a certain problem and had discussed the next step with their advisor, planning to develop the more general case into a doctoral dissertation. After OpenAI announced its results, they began rushing to organize their existing work, hoping to publish it as independent research. He believed the topic still had value for further research, but the original excitement had been greatly diminished.

Grand technological progress, when it lands on specific individuals, always makes people sigh.

After creating this big news, OpenAI CEO Sam Altman finally stopped sitting there stunned and instead looked up at the stars.

Last year AI still could not tell whether 9.8 or 9.11 was larger; this year it can already challenge the Riemann Hypothesis.

What it wants to do next year, I no longer dare to imagine.

When Wang Hong and Deng Yu won their awards, some people said this might be the last Fields Medal awarded to pure humans.

At that time we did not believe it, but now we have to.

The answer is that the two are not enemies at all. On the contrary, the major model companies are all seriously recruiting mathematical talent.

Jacob Zimmerman, who just won the Fields Medal this year together with Wang Hong and Deng Yu, has now joined OpenAI and is involved in AI safety work.

Alpöge, who recently cracked the Jacobian conjecture, is his student and currently works at Anthropic.

Zimmerman once said: I do mathematics mainly because I enjoy it, and because I am good at it.

Whichever company stops to catch its breath may be eliminated outright. It is not that you fall behind if you do not advance; rather, if you advance too slowly, that is equivalent to falling behind.

Anthropic is sprinting toward an IPO

From GPT3 to GPT6, the iteration time for each major version was 3 years, 2 years, and 1 year respectively, and minor versions are updated almost weekly.

Google's Gemini, because it fell behind in one or two key versions, was stripped of its title by netizens and demoted to America's big soybean bun, and to this day it has still not been able to return to the first tier.

This is also why Altman is constantly talking about atomic bomb explosions and sitting there stunned; Anthropic CEO Dario also acts mysterious and alarmist.

Mathematics is a field that produces a wealth of difficult problems.

A few percent improvement in reasoning ability, a few percent reduction in token consumption—users find it hard to feel these intuitively, but cracking a problem that has troubled mathematicians for decades is a result anyone can understand.

If you do not believe it, go take a look at the crackpot science forums: 90% of amateur mathematics enthusiasts are working on the Goldbach conjecture.

Solving problems is also a way to flex one's muscles.

In September, when OpenAI heard that the neighboring company A was challenging the NS equations, it immediately concentrated all its available resources in that direction, organized ten thousand AI agents to work in parallel, and finally preempted company A by releasing its own results first.

The NS equations are one of the directions Wei Dongyi focuses on

Mathematics has now become a battlefield of the gods, and any stray shockwave can stun the humans in the arena.

Before mathematicians felt crushed, people in other industries had already been crushed several times over.

The first thing that made the whole world sweat was Go.

After 2016, AlphaGo defeated Lee Sedol and Ke Jie one after another. Since then, Go players have gone from a profession symbolizing human intelligence to a job of fighting AI on the board every day.

At that time, humans could still understand this defeat as the loss of one special skill—machines may be good at chess, but surely they cannot be better than humans at everything, right?

Later, the battlefield moved from the chessboard to the office desk.

Midjourney and Stable Diffusion began taking the jobs of illustrators; ten years of training may be worth less than one minute of AI output;

Progress in coding ability quickly reduced Vibe coding to the difficulty of Lego blocks (not that hard, actually);

Debuting solo was not enough. Just recently, Mango TV's AI lineup was officially announced. Although there are only five of them, gazing at their faces you can see a sea of people.

We used to believe that the more a job required knowledge, creativity, and professional training, the less easily it could be replaced.

But these skills often leave behind large amounts of text, images, and code, which happen to be exactly what models can repeatedly learn from and quickly generate.

Instead, the trivial chores we most want machines to take over have not yet disappeared from our lives so smoothly.

For example, doing housework.

We originally expected machines to take over tedious labor so that people would have more time to think and create. Now the situation is reversed:

But if so much work can be completed faster, why is the first thing we feel pressure rather than finally being able to relax a little?

Is simply living day to day humanity's only choice?

We have always been preparing to "avoid being eliminated." When we were young, we worked hard to get into good schools; when we graduated, we chose industries with prospects; after starting work, we took certifications, did projects, and accumulated experience.

Not everyone necessarily wants to become the top of their industry; more often, people just think about gaining a bit of bargaining power and irreplaceability by improving some skill of their own.

Ten years ago, a certain kind of inspirational saying was very popular: "When the era abandons you, it will not even say goodbye." The real meaning of this sentence is that before the era abandons you, it actually does say goodbye, but many people do not notice.

The conclusion is that we must keep up with the times and not be the ones abandoned.

When DeepSeek first became popular, someone compiled a teaching guide and put it on WeChat Reading, and it once rose to number one in popularity.

AI now understands human speech better than 80% of employees and 99% of leaders in this world.

Could it be that this time the era really is not planning to say goodbye to us?

On October 11, Yau Shing-Tung led the establishment of the Mathematics and Artificial Intelligence Laboratory Alliance, joining forces with eight mathematics research institutions and preparing to independently develop models dedicated to mathematical science.

No one knows where these attempts will ultimately lead, but to say that mathematics is already over is probably premature.

More than 30 years ago, someone also concluded that history had ended, and later it proved that this was only a comma in history.

In times of great change, thought is always especially active, and we do not need to rush to write the ending for ourselves.

According to incomplete statistics, this year has already seen the X-repression theory, the great flood theory, the great sleep theory, the thin-muscle theory... the Hundred Schools of Thought have reappeared in the world.

Whether it is finding fun while working or cutting costs while raising laughs, in the face of a future full of uncertainty, experts and ordinary people are about the same.

At least there are still some certainties we can grasp.

For example, starting tomorrow, care about grain and vegetables, drink plenty of water, get plenty of sleep, and be a pure day-to-day person.

This article comes from the WeChat public account: Phoenix WEEKLY, author: Wang Dong AI Venture Capital Daily Channel: Frontier Technology. This content is published with the author's authorization, and the views represent only the author and not the position of Huxiu. If you have objections or complaints about this manuscript, please contact tougao@huxiu.com. Everyone who is changing and wants to change the world is in the Huxiu APP[ ]]]