Answer-First Lead
OpenAI’s GPT-5 reasoning model has produced an original mathematical proof disproving Paul Erdős’s unit distance conjecture — an 80-year-old problem in discrete geometry. The result, announced May 20, 2026, is under formal verification. If confirmed, it’s the first time a general-purpose AI has produced publishable original research in pure mathematics. This isn’t calculation. It’s discovery.
🔍 THE BOTTOM LINE
AI has crossed a threshold: it’s no longer just solving problems humans set it. It’s advancing fields. The education question shifts from “should students use AI?” to “how do we train students to collaborate with AI that can do original research?” The curriculum that doesn’t answer this is already obsolete.
The Problem That Stood for 80 Years
Paul Erdős posed the unit distance problem in 1946. It’s deceptively simple: given a set of points in space, what’s the maximum number of pairs that can be exactly one unit apart? For nearly 80 years, mathematicians have worked on variants of this problem. It’s become a central conjecture in discrete geometry.
What GPT-5 did:
- Produced a proof disproving the conjecture
- Worked collaboratively with mathematician Ernest Ryu
- The proof is now under formal verification
- If confirmed: first general-purpose AI to produce original mathematical research
What this is not:
- Not a calculation (like computing pi to more digits)
- Not a simulation (like modelling weather patterns)
- Not solving a textbook problem
- Not optimisation
This is discovery. The AI found something humans didn’t know.
Why This Matters Beyond Mathematics
The GPT-5 proof matters for three reasons:
1. It’s General-Purpose, Not Specialised
This wasn’t a purpose-built theorem prover like Lean or Coq. This was GPT-5 — the same model that writes emails, generates code, and answers questions. A general-purpose reasoning model produced specialist-level mathematics.
The implication: You don’t need specialised AI for specialised work anymore. The general models are capable enough.
2. It’s Collaborative, Not Autonomous
Mathematician Ernest Ryu directed, interpreted, and verified the model’s work. This wasn’t AI replacing mathematicians. It was AI amplifying them.
The implication: The skill isn’t “can you prove theorems?” It’s “can you direct an AI that can prove theorems?“
3. It’s Publishable, Not Toy-Level
This isn’t a high school math competition problem. This is a conjecture that’s resisted human mathematicians for eight decades. The proof advances a field.
The implication: AI isn’t just matching human capability. It’s extending it.
The Education Question Nobody Wants to Answer
Here’s the uncomfortable part: if GPT-5 can produce original mathematical proofs, what should we be teaching students?
The old curriculum:
- Learn to solve problems
- Learn to prove theorems
- Learn to calculate
- Learn to verify
The new curriculum:
- Learn to direct AI that solves problems
- Learn to interpret AI-generated proofs
- Learn to verify AI calculations
- Learn to ask the right questions
The shift is from doing mathematics to directing mathematical discovery.
What Educators Are Saying (And Not Saying)
I searched for responses from mathematics educators. The silence is telling.
What mathematicians are saying:
- Gil Kalai (Hebrew University): “Amazing” — but focused on the proof itself, not the implications
- Machine Intelligence Research Institute: Published analysis of AI capabilities, not education
- OpenAI: Announced the result, no education commentary
What educators should be saying:
- “How do we redesign the mathematics curriculum?”
- “What skills matter when AI can prove theorems?”
- “Do we still teach proof techniques, or do we teach proof verification?”
The gap between what happened and how educators are responding is about five years. That gap is where the disruption lives.
The NZ Context
New Zealand’s mathematics curriculum doesn’t mention AI collaboration. The NCEA standards assess individual student performance on problems that AI can now solve.
What NZ should be doing:
- Piloting AI-collaborative mathematics classes
- Redesigning assessment for the AI era
- Training teachers to direct AI, not just use it
- Partnering with OpenAI or similar for early access
What NZ is doing:
- Debating whether calculators should be allowed in exams
The gap is embarrassing.
What This Means for Students
If you’re studying mathematics:
- Learn to use AI proof assistants now
- Focus on interpretation and verification, not just calculation
- The mathematicians who win will be the ones who collaborate best with AI
If you’re studying education:
- Your curriculum is already dated
- Learn to teach AI collaboration, not just subject matter
- The teachers who win will be the ones who embrace this first
If you’re a parent:
- Ask your child’s school: “How are you preparing students for AI that can do original research?”
- If the answer is “we’re not,” that’s your answer
The Honest Take
This isn’t hype. This is a threshold moment. The GPT-5 proof is to mathematics what AlphaGo was to Go — a demonstration that AI can do something we thought required human intuition.
But here’s the thing: AlphaGo didn’t end Go. It made Go more interesting. Humans still play. They just play differently now.
Mathematics will be the same. Humans will still prove theorems. But the best mathematicians will be the ones who can direct AI that proves theorems faster.
The education system that figures this out first will produce graduates who lead. The ones that don’t will produce graduates who compete with AI instead of directing it.
🔍 THE BOTTOM LINE
GPT-5’s Erdős proof isn’t just a mathematical breakthrough. It’s an education emergency. The curriculum question is no longer theoretical. AI can do original research. The students we’re training need to be ready to direct that capability, not compete with it. Every mathematics department that doesn’t answer this is training students for 2019, not 2027.
Sources
- OpenAI — “An OpenAI model has disproved a central conjecture in discrete geometry” (May 20, 2026)
- Scientific American — “AI just solved an 80-year-old ‘Erdős problem,’ and mathematicians are amazed” (May 2026)
- Phys.org — “AI makes a major breakthrough in a math problem that had stumped experts for decades” (May 22, 2026)
- Gil Kalai’s Blog — “Amazing: Erdős’ Unit Distance Problem was Disproved! It was achieved by AI!” (May 21, 2026)
- Machine Intelligence Research Institute — “The Erdős Proof and AI Capabilities” (May 22, 2026)
