OpenAI Claims AI Solves Over 100 Mathematical Problems in 24 Days

The mathematical community has truly been cornered by AI!

Just now, OpenAI suddenly released a major blog post—

An internal new model, only trained since August 28, has now solved over 100 world-class mathematical problems.

Spanning, or most areas of mathematics.

Even more astonishing, among these over 100 questions is the long-standing Millennium Prize problem known as the Navier–Stokes equations.

This rate of evolution has surprised even the mathematicians at OpenAI.

From training starting on August 28 to the official announcement on September 21, just 24 days.

An internal AI still in training has begun solving problems that humans have failed to resolve for decades, even centuries.

Today, OpenAI invited nine of the world’s top mathematicians to formally establish an independent Mathematics Advisory Group.

The list includes Fields Medalists Timothy Gowers and Martin Hairer, as well as renowned figures in theoretical physics and mathematics, Edward Witten.

This lineup is堪称 the Marvel Avengers of the mathematics world.

What they will face next is AI beginning to systematically solve complex math problems—how should the human mathematical community respond?

In the mathematical community, there is also a hierarchy of disdain among difficult problems.

To understand the true value of this progress, we need to rewind the timeline back to September 8.

OpenAI has developed a brand-new internal model that solved the NS Millennium Problem in just 88 hours.

It provides a proof of the existence and smoothness problem of the Navier-Stokes equations, along with the simultaneous public release of a 166-page paper and Lean-formalized verification code.

Further reading:

According to the company's description, this result indicates:

Under the influence of a smooth external force, an initially stationary, smooth, three-dimensional incompressible fluid can develop a singularity in finite time, corresponding to versions C and D of the official statement of the Millennium Prize Problem.

Behind this proof is a large-scale collaborative research effort involving over 10,000 concurrent agents.

They can read internet caches, execute code, and exchange information within groups, while different groups explore different versions of the problem and solution paths.

Researchers also remained actively involved: after the system achieved results related to the Euler equations, they redirected resources toward the Navier-Stokes problem, using existing solutions to guide further exploration; they then used Codex to consolidate valuable intermediate insights from various teams, enabling communication between different approaches.

On September 5, approximately 88 hours after the first agents were activated, the system arrived at a solution.

Subsequently, GPT-6 Astra spent another 17 hours completing the Lean formalization and verification.

Models, tools, parallel exploration, and human research teams together formed the foundation of this breakthrough effort.

Today, OpenAI has expanded the scope of its disclosed achievements: beyond the Navier–Stokes equations, it has addressed over a hundred long-standing open problems spanning most areas of mathematics.

OpenAI stated that this progress has sparked internal discussions on how to ensure the mathematical community is promptly informed of changes, allowing time for preparation and adaptation.

This also makes the practical question of “how to review and how to publish” follow closely behind advancements in capability.

Just three days after OpenAI announced a breakthrough on the Navier-Stokes equations, the mathematics community was stunned.

27 Fields Medalists have jointly signed a strongly worded open letter titled "The Severe Misalignment of AI in Mathematics."

Further reading: Just now, Fields Medalists such as Terence Tao and Daniel Kane jointly protested: AI companies are destroying the entire mathematics community!

The signatory is a legendary figure who spans the 48-year history of the Fields Medal.

Deng Yu, Terence Tao, June Huh, Ngô Bảo Châu, Peter Scholze, Martin Hairer.......

The recipients of the highest honors in mathematics have almost all come forward.

Their anger is not directed at AI itself. The open letter explicitly acknowledges at the outset that AI has tremendous potential to accelerate genuine mathematical research.

What truly upset the math experts was the way it was disclosed.

They criticized that the major AI giants treat public mathematical problems that humanity has spent centuries solving as benchmarks for ranking and scoring.

Solving this metric may diverge from the understanding sought by mathematical research.

Moreover, after the giants hastily announced it, there was no time to organize the ideas or thoroughly document the new methods.

This utilitarian, opaque brute-force approach may seem impressive on the surface, but it leaves mathematics in disarray.

Of course, OpenAI did not avoid it; in their latest announcement, they directly cited this article and publicly stated—

The AI giants and mathematicians need to sit down and have a serious talk.

Therefore, the "Avengers of Mathematics" is the real highlight.

The initial group of independent advisors consists of nine members.

They are respectively from institutions such as Cambridge, Oxford, Harvard, Stanford, Berkeley, and the Institute for Advanced Study at Princeton.

Just pull out a few names—they’re all scary enough.

Timothy Gowers, a Fields Medalist closely following AI advancements in mathematics. Martin Hairer, 2014 Fields Medalist.

Camillo De Lellis, a professor at the Institute for Advanced Study in Princeton, is precisely a world-leading authority on the mathematical theory of partial differential equations and fluid mechanics.

Edward Witten, needless to say, is the foremost contemporary master of theoretical physics, known as the "successor to Einstein" and the first physicist to receive the Fields Medal.

OpenAI assigned the team three main tasks:

• Help determine how significant the new mathematical breakthrough is;
• Discuss how and when to release these results;
• Ensure that mathematical research in the AI era continues to adhere to academic and professional standards.

Another interesting design is that these nine individuals receive no funding from OpenAI. They are free to publicly criticize OpenAI, offer suggestions not requested by the company, and independently decide on membership changes.

However, there is a key limitation: the advisory group is not responsible for advising OpenAI on how to control the pace of its internal mathematical research.

Nine academic luminaries have the right to speak publicly, but not the "braking power."

Everything has become extremely clear.

Less than a month after training began, over 100 historical难题 were solved. Before this astonishing speed, the technological bottleneck is no longer about whether AI can solve them, but whether humans can review them all.

What does it feel like to review AI papers? Buckmaster feels it most deeply—

The initial proof generated by the large model was the most terrifying thing he had ever read. The paper they rushed out to secure a scoop, he later dismissed as a pile of "AI slop," and publicly apologized for it.

Even after a year of refinement, human elites are still overwhelmed—now machines spit out over a hundred major problems in a single month. Worldwide, the number of people capable of reviewing a proof at the Navier-Stokes level can be counted on one hand.

So, what is the actual function of this "Nine-Person Advisory Team"?

In simple terms, OpenAI has provided this collection of machine outputs with a top-tier production line and a demining team.

Which question should be reviewed first? Which one needs to be broken down and rewritten? Who should verify it? Whose name should be credited? OpenAI dumped all these tedious tasks onto nine unpaid experts, while tightly holding the throttle of its high-intensity training engine in its own hands.

The mathematical community itself has recognized this tsunami.

Number theorist Daniel Litt from the University of Toronto recently revealed on X that, after discussions with department heads, the consensus is that incentive structures, hiring standards, and PhD evaluation methods must be completely restructured. Everyone is ready to defend mathematical culture and will no longer encourage merely producing PDFs.

Fields Medalist Figalli even predicts that the role of mathematicians is being forced to shift from "those who solve problems" to "those who decide which problems are worth solving."

Today, OpenAI unambiguously establishes the new world's unwritten rule: discovery belongs to machines, while verification and interpretation belong to humans.

Physics, chemistry, and biology will eventually reach this point as well—every discipline will have its day.

Mathematics is just there first.

References:

https://openai.com/index/advisory-group-on-mathematics-and-ai/

https://agmai.org/

https://mathandai.org/multilingual.php?lang=zh

https://openai.com/zh-Hans-CN/index/navier-stokes-solution/

This article is from the WeChat public account "New Intelligence Yuan" (ID: AI_era), authored by ASI Revelation, edited by Marco, Peach, and Moses.