Li Xinyi, a PKU Class of 2007 mathematics alumnus, criticized OpenAI and Anthropic for harming the mathematics community by using AI to solve problems

[[**Li Xinyi, a PKU Class of 2007 math department alumnus, criticizes OpenAI and Anthropic for using AI to "solve" math problems, arguing that this sends the wrong message, harms the mathematical community, and reveals that AI may steal unpublished human math drafts—and is ruining young mathematicians.**

**Key points**

**1. AI "solving" hard problems lacks human understanding; it is just "leaderboard farming"**

**2. The publicity misleads the public and funding decisions, and may ruin young mathematicians**

PR coordinated by Silicon Valley companies sends the public the wrong signal that "human mathematicians are no longer needed," scaring off young people who want to study math and misleading funding allocation.

**3. AI may steal unpublished human math drafts**

**4. The math community needs to support a large number of "mediocre" mathematicians to maintain the knowledge tree**

**5. Applied mathematicians are more lost; theoretical mathematicians still hold ground through intuition**

Applied math work of the "combine A and B" type is more easily replaced by AI; in theoretical math, fields that rely on spatial intuition and cannot be trained on text symbols alone (such as geometry and topology) are currently not learned well by AI. 2026-10-03 18:53 PKU Class of 2007 math alumnus: What O/A are doing is ruining young people in math departments. New Zhiyuan Quick Overview

This article is from the WeChat public account: New Zhiyuan, editor: Aeneas, author: ASI Revelations

Last month, OpenAI and Anthropic began extreme infighting. One side had just announced conquering a century-old mathematical conjecture, and the other immediately fired back with big news of cracking a Millennium Prize Problem.

And social media fell into a traffic frenzy.

As a tech media outlet, our account also followed up with reports on the latest developments and gained extremely high attention. However, shortly after an article titled "OpenAI Has Won Another Millennium Prize Problem" was published, Li Xinyi, associate professor at Peking University's Beijing International Center for Mathematical Research (BICMR) and a PKU Class of 2007 math department alumnus, proactively reached out to us, saying he wanted to talk.

That afternoon, we had a conversation of more than 30 minutes.

When it came to the practices of these Silicon Valley giants, his tone was full of deep worry and helplessness.

He told us how their crude approach harms the entire mathematical academic community, and how it ruins young people who have just begun to feel passion for mathematics.

English text is being prepared.

Ordinary people may think that top conjectures and hard problems are the final boss, and now AI has defeated the boss in one second, so AI has won.

But in the eyes of real mathematical researchers, this logic is absurd and fatal.

Li Xinyi said that in the math world, the ultimate purpose of generation after generation pressing forward to conquer a hard problem is not the conclusion carved on a tombstone, but to cultivate generation after generation of human mathematicians along this long journey.

"Of course everyone has this goal and works hard for it—that's very good. But whether it is solved or not actually does not matter that much. What really matters is that in the process of solving the problem, humanity produces countless profound understandings."

He pointed directly at the recent PR tactics of O and A.

Rumor has it that OpenAI solved god-level problems such as the Hodge conjecture and the BSD conjecture. But in the eyes of professionals, this kind of "solution" must be put in quotation marks.

"This is actually a situation of 'being solved, but not understood at all.' If a human mathematician had solved it, during the long proof process he would have produced enough, sufficiently rich theoretical understanding. But the problem now is that whether AI finds a counterexample or gives a result, humans simply have not digested it."

What is more terrifying than brute-force problem solving is the publicity carried out in coordination by Silicon Valley companies. This kind of PR is spreading a wrong idea to the public—that the world no longer needs human mathematicians.

"This kind of publicity, apart from scaring young people who originally wanted to study math, will also mislead those who make funding decisions with a false impression."

"The math community still needs living people, and still needs to support living mathematicians with the same intensity. How concrete mathematics is done may change, and the standards for evaluating people will also change. But if this false prosperity is not criticized, it is actually fueling a terrible future."

Draining the essence of human wisdom, yet wanting to destroy the group that produces that essence

During the interview, Li Xinyi revealed to us something that had just happened around him.

A postdoc researcher in his team had once spent a very long period of time relentlessly working on an obscure mathematical conjecture.

During the research process, he habitually used large models (early beta versions from the GPT-5.x era) as an aid, discussing with AI while working out his own ideas.

However, a nightmare happened.

On the very day of this interview, someone posted a paper on arXiv. The core idea and content of the paper were almost exactly the same as the draft stored on that postdoc's computer!

The person who posted the paper was a researcher whose main field is physics.

For this physics researcher, publishing this cross-disciplinary paper was not part of his main research at all. For him, it was more like a playful "lottery draw"—he used AI to generate a bunch of seemingly rigorous results, found them interesting, and directly put his name on them and posted them.

In the end, because it was discovered early, the two sides negotiated fairly amicably and it ended with "joint authorship," but this revealed a terrifying truth—

We mentioned the recent "Peking Union Medical College doctor solves a matrix theory problem" incident: a doctor with no professional mathematics background, merely through excellent prompts, got the AI to spit out a completely new proof. Is that possible?

Li Xinyi answered: "The raw materials needed for that proof were actually already there. Probably no one had tried to prove that conjecture for many years. But he tried it, and now AI is smart enough that its tricks can string together all the clues from the previous literature and produce it in one go."

In other words, these problems were already in a state of "everything is ready, all that's missing is the east wind."

The theory was there, the direction was there; all that was missing was a machine that could instantly search all of humanity's memory banks and perform brute-force enumeration.

Why keep supporting a bunch of "mediocre" living people? To preserve the tree of human knowledge

So now, a cruel question is placed before the academic world: for these academic results, who exactly should get the credit?

Should it go to the mathematician who spent ten years of painstaking effort conceiving the underlying system? Or to the "lottery player" who merely typed a few prompts and pressed Enter?

"Today our faculty were actually holding a meeting and still fiercely debating this matter," Li Xinyi revealed to us. "First, do you recognize the mathematical results produced by AI themselves? We think they should be recognized, because mathematical results themselves are objective facts. But!"

He changed tack: "As the mathematics community, we need to determine who gets the credit. Who can get a faculty position because of this? Who can get a National Natural Science Foundation grant? We believe that rewards absolutely should not go to those who merely pressed a button. You must have input your own mathematical understanding and made a contribution in terms of understanding to receive honor."

This also touches on a broader proposition: why exactly should human society support so many mathematicians?

From a utilitarian perspective, the vast majority of basic mathematical research has no concrete use in today's real world. They are merely theoretical candidates that may become "useful" decades or even a century in the future, when some technological explosion occurs.

But if support for human mathematicians were stopped because of this, that would be suicide.

"In order to maintain, over the long term, that there are still people in human society who can 'understand mathematics,' we must maintain an extremely tall tree of mathematical knowledge and maintain a large community of human mathematicians."

"You can't say, I just need to learn what's in the textbooks and then teach the textbooks to the next generation. That turns into rote memorization. In this process, in order to truly understand the underlying logic of mathematics, you must understand mathematics at a higher level—that is, you have to personally do mathematical research."

Why not pool limited funds and develop only mathematics that is "truly useful and can be immediately monetized"?

The answer is: because you must support this large group of people and must allow them to do a great deal of "useless" things. You cannot only think about plucking the top of the pyramid. Without a large base—even a somewhat "mediocre"—community of mathematicians as soil, it would be impossible to produce top geniuses like Terence Tao.

Are applied mathematicians "dead"?

Moreover, this tsunami is not striking every mathematician evenly. In different branches of mathematics, people feel the chill very differently.

For example, the core of much applied mathematics or engineering mathematics is essentially applying advanced theory A to concrete scenario B. And in this kind of "combining A and B to do rough work," today's AI is indeed better than humans.

"So people doing applied mathematics research are actually more lost than those doing pure theoretical mathematics research."

Then where is the way out for theoretical mathematicians?

Li Xinyi said that even if one day in the future AI does all mathematics better than humans, humans must still grit their teeth and train their own brains, ensuring that there are still generation after generation of living people in this world who "understand mathematics," rather than happily surrendering all research mechanisms to AI.

In addition, today's large models are essentially still doing permutations and combinations within known frameworks. If asked to break through from zero in deep waters, the cost is still very high.

For AI to break through the Navier-Stokes equations, it may have cost nearly $20 million in tokens. What does $20 million mean? It's enough to hire 100 top human mathematicians for a whole year!

"If you really hired 100 mathematicians to compute for a year, they might actually be able to construct that counterexample. Getting AI to discover genuinely newer theories would very likely cost even more tokens than the existing work."

For a long future, the human brain will remain the most energy-efficient and most disruptive "carbon-based computing cluster" in the entire universe. Therefore, human society is still worth investing heavily in sufficiently innovative mathematicians.

Scientific research still needs people to do it.

Recently, Vibe Coding has become popular, but Vibe Math hides a fatal crisis.

Most of today's Vibe Coders are programmers who personally wrote hundreds of thousands of lines of code back in the day and have a deep understanding of low-level code. But what about the next generation? By then, will the power of life and death have to be handed over to AI?

Finally, Li Xinyi told a darkly humorous joke.

"A couple of days ago I joked with a collaborator that I really have to thank him for making our papers so obscure over the past few years that AI still hasn't learned our direction, and we actually still have a bite to eat."

"In many mathematical disciplines, such as geometry or topology, mathematicians describe spatial things the way they describe a painting. You cannot complete an argument relying only on text written on paper. The writer has a picture in mind, and the reader also conjures the same picture in mind. In mathematical disciplines like these that can only be grasped intuitively and rely extremely heavily on spatial intuition, AI actually does not learn very well right now."

A Fields Medalist, writing novels?

It is said that among the Peking University class of 2007 alumni, attitudes toward AI vary.

The WeChat Moments of Fields Medalist Deng Yu, by contrast, have a rather unusual style.

The interview ended, and our hearts could not calm down for a long time.

In the steel forest ruled by computing power, we still need that group of living mathematicians who seem "mediocre" and do "useless things" on the blackboard every day.

It is precisely they who build the dignity that makes human civilization human. AI Venture Capital Daily Channel: Frontier Technology. This content is published with the author's authorization; the views represent only the author and do not represent Huxiu's position. 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[ ]