An OpenAI Model Disproved an 80-Year-Old Math Conjecture Autonomously. The Service Business: Help Organizations Reformulate Their Stuck Domain Problems Into AI-Solvable Tasks.

An OpenAI general-purpose reasoning model just disproved a math conjecture that stood for nearly 80 years.

Paul Erdős posed the unit distance conjecture in 1946.

For nearly 80 years, mathematicians believed the best solutions to the planar unit distance problem looked roughly like square grids.

The AI found an entirely new family of constructions that performs better.

Three prominent mathematicians — Noga Alon, Melanie Wood, and Thomas Bloom — publicly endorsed the proof.

Thomas Bloom put it plainly: "AI is helping us to more fully explore the cathedral of mathematics we have built over the centuries. What other unseen wonders are waiting in the wings?"

What actually happened here

This is not AI autocompleting a proof someone sketched.

OpenAI's model autonomously solved an open problem.

It connected ideas from algebraic number theory — a field that studies factorization in extensions of the integers — to a geometry problem that researchers had attacked for decades with conventional tools.

The model did something human researchers had not: it reached across domain boundaries and found a connection no one had looked for.

Greg Brockman called it "the first time AI has autonomously solved a prominent open problem central to a field of mathematics."

OpenAI says the result demonstrates "AI systems are now more capable of holding together long, difficult chains of reasoning and connecting ideas across fields."

The same problem exists in every industry

Every company has its Erdős conjecture.

A supply chain optimization problem that nobody has cracked in 12 years.

A drug interaction prediction model that keeps failing validation for reasons the team cannot isolate.

A scheduling problem that has been "good enough" for a decade because nobody found a better construction.

Most of these problems are not unsolvable.

They are unstated.

No one has ever reformulated them in a way that an AI reasoning model could attack.

That is the gap — and it is a service business.

The service: AI Problem Reformulation

The offering is an AI Problem Reformulation Audit.

It is not an AI implementation project.

It is not a data science engagement.

It is a structured process for taking a domain expert's stuck problem and translating it into the kind of precise, bounded, well-stated task that a frontier reasoning model can make progress on.

The work looks like this:

Step 1 — Problem excavation: Spend 2–3 days with the domain experts. Not talking about AI. Talking about the problem. What is the actual question? What does "better" look like? What data exists? What has been tried? What have been the failure modes?

Step 2 — Reformulation: Write a formal problem statement in a structure the model can work with. Not a prompt — a specification. What is the input space? What is the output? What are the constraints? What counts as a valid solution?

Step 3 — Pilot run: Take the specification to a frontier reasoning model. Run it. Document what the model produces — including where it fails, where it surprises you, and where it finds something worth investigating.

Step 4 — Report: Deliver a structured brief for each problem: what was found, what the model returned, recommended next steps, and what a full research engagement would cost.

Who to sell this to

Mid-size pharma companies with long-standing drug discovery or interaction modeling problems.

Logistics and supply chain teams with optimization problems that have been "solved enough" for years.

Materials science and engineering R&D departments.

Financial institutions with fraud detection or risk modeling problems that keep producing unexplained edge cases.

The common thread: a domain expert who can articulate a stuck problem, access to historical data or prior research, and institutional memory of why previous approaches failed.

Pricing

$5,000–$12,000 for the full audit and pilot across 2–3 problems.

The frame for the client: this is not an AI consulting fee.

It is the cost of finding out in one week whether a decade-long problem has a tractable AI angle — versus another year of the research team working on it without new tools.

Monthly retainer ($1,500–$3,000) to run new problem formulations as reasoning models improve each quarter.

The models are getting meaningfully better every quarter.

Problems that produced nothing useful six months ago may now yield something.

Why now

OpenAI's proof is not just a math story.

It is a capability signal.

If an AI can autonomously connect algebraic number theory to a geometry problem that has been open for 80 years, it can connect domains that industrial researchers have not tried connecting.

The organizations that recognize this and act on it first will have the early advantage.

Most of them are still treating AI as a text tool.

Sources:

https://openai.com/index/model-disproves-discrete-geometry-conjecture/

https://techcrunch.com/2026/05/20/openai-claims-it-solved-an-80-year-old-math-problem-for-real-this-time/

https://news.ycombinator.com/item?id=48212493