AI News

OpenAI and Anthropic, together with the Simons Foundation, invest $15 million to support mathematicians in checking AI results

Phoenix TechnologyUpdated Nov 10 13:12

AI summary

The Simons Foundation announced that it, together with OpenAI and Anthropic, has committed $15 million in funding and resources to support mathematicians in checking, understanding, and extending results produced by AI. OpenAI is contributing $5 million and providing models and compute, while Anthropic has pledged $5 million including access to Claude. Previously, on October 6, OpenAI publicly released 722 mathematical manuscripts generated by internal models on GitHub; the next day it retracted 3 and revised 14, triggered by a sign error, and about 42% of the main results came with Lean formalization materials. Terence Tao noted that mathematics is moving from a scarcity of proofs to an abundance of proofs, while human capacity for verification and understanding may not keep pace.

Why read this

AI score 60

The original text presents a concrete case of human verification failing to keep up after AI mass-produces mathematical results, as well as the response of the two labs funding mathematicians to review them.

Topics

Tags

Body · AI translation

$15 million! OpenAI teams up with Anthropic to ask mathematicians to "grade the papers" for AI. Xinzhiyuan, November 10, 2026, 11:38:36, from Beijing. Updated November 10, 2026, 12:32:36

On October 6, OpenAI publicly released 722 mathematical manuscripts generated by its internal AI models in one go.

Just one day later, 3 of them were announced as retracted.

What brought down three manuscripts together turned out to be a single sign error: one paper made the mistake, and the other two reused the same construction, so they were retracted as well.

On October 15, the Simons Foundation announced that, together with OpenAI and Anthropic, it was committing $15 million in funding and resources to support mathematicians in checking, understanding, and extending AI results.

OpenAI contributed $5 million and additionally provided models and compute, hoping to turn AI progress into deeper human understanding and new discoveries.

Anthropic's pledged $5 million in support includes funding, compute, and access to Claude. It also acknowledged frankly that using AI well in mathematics and in broader scientific research requires time, effort, and resources.

The related funding is managed by the Simons Foundation and used for workshops, conferences, and other activities.

In recent months, OpenAI has set off a series of "major shocks" in the mathematics world:

first it announced 10 research results involving long-standing open problems, then it announced a proof of the Millennium Prize Problem, the Navier–Stokes problem, and afterward it released hundreds of mathematical manuscripts in a concentrated burst.

With results pouring in like a tide, humanity is having a bit of "indigestion."

Terence Tao pointed out incisively that mathematics is moving from "proof scarcity" to "proof abundance," but humanity's capacity for verification and understanding may not keep up.

Tao proposed "the era of proof abundance" in his lecture notes.

Justin Solomon, associate dean of engineering education at MIT, admitted frankly on a podcast that, as for OpenAI's Navier–Stokes proof, he read the first few pages and could not continue.

Solomon admitted on the "Odd Lots" podcast that, faced with the Navier–Stokes proof provided by OpenAI, he could not even get through the first few pages.

722 manuscripts

From "writing it out" to "being able to understand it"

On October 6, OpenAI published this batch of results from an internal unreleased model on GitHub, and the next day it left an awkward record in its changelog of retracting 3 papers and revising 14.

October 7 changelog: 3 manuscripts retracted, another 14 revised.

The problem lay in a single sign:

according to the paper's convention it should have been -1, but in the proof the AI calculated with +1. After correction, the count that was thought to vanish was not zero, and the premises of the theorem used collapsed outright.

The other two manuscripts used the same method and were retracted as well; one of them studied the rational Hodge conjecture for products of K3 surfaces.

This batch of results came from the model's attempts at and screening of about 4,000 problems. The quantity is astonishing, but verification has not yet caught up:

about 42% of the main results are accompanied by Lean formalization materials, allowing computers to check the reasoning process. Some results not yet formalized may still have problems.

Even if the proof itself is fine, the research is not over.

Just knowing that the calculation is correct is not enough; one must also understand how it was done correctly: what is the key method? Can the applicable conditions be relaxed? Can it be used to solve other problems?

In doing mathematical research, authors must understand their own proofs and take responsibility for them.

So finding errors is only one step.

More critical is to explain the useful methods inside clearly, so that others can learn them, use them, and continue the research.

On October 7, the Association for Human Mathematics (AHM) issued a stern statement criticizing this behavior of dumping more than 700 files at once as "a display of power," and even called on mathematicians to stop cooperating with OpenAI.

The AHM guest statement published on Terence Tao's blog retains the working group's signature.

This statement, signed by the association's communications working group, was also posted by Tao as a guest article on his own blog.

Cornell University mathematician Alex Townsend's suggestion was direct:

OpenAI should first release manuscripts verified by Lean, put other materials out separately, and ask everyone to help check them; do not mix verified and unverified ones together.

MIT senior research scientist Andrew Sutherland affirmed OpenAI's attitude of quickly correcting errors, but also said that more work is needed to regain trust.

The advisory group (AGMAI) of the Institute for Advanced Study in Princeton had proposed as early as late September that AI labs should not only focus on releasing results, but also provide money and resources to support humans in understanding these results.

OpenAI and Anthropic's joint support this time precisely responds to the academic community's earlier demands.

How exactly should this $15 million be spent?

AGMAI previously also suggested holding summer schools around complex proofs, or organizing long-term collaborative research groups, so that everyone has the time and resources to take AI-written long proofs apart step by step and explain them clearly.

Simons Foundation president David Spergel also emphasized:

these discussions must be led by the mathematical community.

David Spergel emphasized in the announcement that mathematics should not be seen merely as a collection of proofs and theorems, but as collective understanding driven by human curiosity.

Terence Tao's Math 2.0 diagram, placing "human understanding" at the center.

And this discussion has already reached beyond mathematics.

The Simons announcement explicitly mentions that the tools and compute provided by OpenAI and Anthropic will also be used in basic scientific research beyond mathematics.

Who checks errors, who can understand, and who is responsible when problems arise—other disciplines also need to think about this in advance.

Recently, a reversal occurred in a theoretical physics test:

GPT-6 Astra was judged wrong, but after expert review it was found that some of the reference answers themselves were wrong. After correction, Astra answered 49 of the 50 research-level questions in TheoPhysEval correctly, reaching an accuracy of 98%.

Then explain complex derivations clearly so that others can understand and use them; only then can these results become the starting point for the next discovery.

References:

Editor: Yuanyu

Source

Phoenix Technology