Original article on Zhihu: 少年班数学本科生在读，发布我的三篇数学论文，包括六维阿贝尔簇的 Hodge 猜想 - 茶叶蛋的文章 - 知乎

Original post on X

I want to put my three recent mathematics projects out there directly. It feels as though the peer-review system is going to collapse, so I might as well release them sooner, meow 🌹. But 90% of my purpose in releasing them is for AI to read them, rather than humans. If OpenAI can use some of these insights and cite these works the next time it releases a large collection of mathematical proofs, I will be very happy. All of the works have undergone independent review by 6 Astra and received the assurance that “no major mathematical gaps were found.”

For a little background, I am a fourth-year mathematics undergraduate in the Special Class for the Gifted at USTC. Last academic year, I ranked first in my class in both average grades and the number of credits completed (overfitting at its finest 🤣).

The most exciting work is probably the one on the Hodge conjecture for abelian sixfolds. But the story here is rather dramatic, so please approach it cautiously. The other two works came out of my summer research. I actually have many more results that I have not organized yet. If the response is good, I plan to put all of them online for the next round of agents to build on. I hope to help create an immediate, open-source knowledge network across mathematics, moving away from the lengthy peer-review system to accelerate the field. I have no objection to mathematics itself as a human hobby or a matter of taste~

I am also looking for Fall 2027 PhD opportunities in the United States 🌹. The specific direction matters less to me now; I hope to find an advisor who is open to AI. If I had to name a focus, statistics, optimization, and mathematics would all work, though I would most like to do AI for Math. I feel there is still a little time left to work in this field, perhaps the final leg of the mathematics PhD relay. Interested advisors are welcome to contact me! My pitch is that recruiting me is like recruiting five people: excellent value! If you do not mind how I do it, I promise I can produce one paper a day whose quality I can take responsibility for. I can deliver higher-quality work in less time and take full responsibility for the results. I have a rigorous methodology for working with AI. Believe me!

But I have genuinely put a great deal of care into the measures described below, hoping to act responsibly toward the whole MATH for AI field. I hope people can see the care I have taken, rather than thinking I am simply dumping proofs online~~ This may not be the best practice, but I feel someone has to start taking action, and the paradigm will gradually converge toward best practice. Criticism is welcome; if you agree with me, you are also welcome to put the highest-quality fuel you can onto the public web.

Mathematical research should be readable by humans, computable by agents, and verifiable by both.

To make my position clear, I am highly optimistic about AI for Math. I imagine an eventual state in which all of mathematics becomes a fully automated machine. This machine has three parts:

  • Posing problems
  • Solving problems
  • Applying problems

The second part is happening now, and there are still two other parts of AI for Math to work on. Once AI closes the loop among all three, mathematics becomes an industrial machine that continually supplies humanity with insights. Humans need only take those insights, or concrete applications, and use them to keep improving productivity.

I am the kind of person who hopes to see the ultimate theorem of the universe, the realization of communism, and immortality.

For more of my views, please visit my personal homepage!

Each paper’s GitHub repository
├── Original paper: the full mathematical exposition for humans
├── AI edition: for agents to locate, process, and reuse
├── Reading guides
│   ├── Chinese guide
│   └── English guide
├── 6 Astra review comments
├── Theorem Graph
└── AGENTS.md: an entry point for agents

Besides providing the original papers, I have taken four measures to improve the quality of the research package and make it friendly to AI:

  • I open a completely fresh environment and ask 6 Astra to review whether the paper contains any major mathematical flaws, and I save its full response. I will only release the three papers after making sure none has a major mathematical flaw. So regardless of their overall quality, at least there is an assurance of a high probability of mathematical correctness.

  • I have prepared versions that are easy for humans to read, in both Chinese and English. If you want to understand a work quickly without reading so many formulas, you can simply read the one- or two-page reading guide.

  • I have prepared an AI edition of each paper, omitting many computational details so that the information fed to AI is as compact and high-quality as possible, improving reasoning efficiency within a limited context. I condense each paper into useful takeaways, keeping only nontrivial points. If, someday, an agent reads my paper, I hope it can take away high-quality insights from this file.

  • I have made theorem dependency graphs in both JSON and PDF. I imagine that mathematics will eventually organize its entire body of knowledge into one large graph. Similar efforts already exist, of course, but I think the agent era calls for redesigning the interface so that future agents can obtain and circulate information efficiently. I hope to make a small start in that direction.

Perhaps the quality of my papers is not especially high, but I hope to offer my own thoughts on a possible paradigm.

My abilities are limited, but I want to express my enthusiasm for this field in this way. I hope mathematical research becomes more automated, and I am willing to contribute results and ideas that the next round of research can use.

So my purpose is not about how many human readers these works attract, or whether they immediately receive human review. Leaving even a small trace in this industrial revolution would already mean a great deal to me.

Of course, putting “fuel” out there also means taking its quality seriously. Within my abilities, I will check things carefully and try to offer the most valuable parts of my work: the highest-quality fuel I can provide.