# AI cracks decades-old maths — OpenAI and DeepMind, days apart

> In May 2026 OpenAI disproved an Erdős conjecture; Google's AI solved nine more.

*Two milestones in one week. Here's the brief.*

By The FeaturedDaily Desk · FeaturedDaily
Canonical: https://featureddaily.com/news/ai-maths-erdos-brief

> **Key:** **The one-liner:** in a single week, two labs showed AI can now produce original mathematics that survives checking — narrowly, but for real.

**What happened.** On 20 May, OpenAI said a general reasoning model disproved the planar unit-distance conjecture (Erdős, 1946), using algebraic number theory rather than the long-assumed grid. A day later, Google DeepMind posted that **AlphaProof Nexus** (Gemini 3.1 Pro + Lean) had solved nine open Erdős problems and 44 conjectures.

**Why it's credible.** DeepMind's results are verified step-by-step in the Lean proof assistant — the model proposes, the checker certifies. OpenAI's was checked by mathematicians including Timothy Gowers, with formal peer review still to come (attribute, don't assert).

> **Note:** **The caveat that matters.** DeepMind's Demis Hassabis said the system is 'still not AGI'. This is AI as a verified-maths instrument for narrow problems — impressive, and a long way from general reasoning.

## Key takeaways

- OpenAI (20 May): an internal reasoning model disproved a 1946 Erdős conjecture in geometry.
- DeepMind (21 May, arXiv): AlphaProof Nexus solved nine open Erdős problems + 44 conjectures.
- Verification: DeepMind's proofs are checked in Lean; OpenAI's are human-checked, peer review pending.
- Caveat: Hassabis says it's 'still not AGI' — a tool for maths, not general intelligence.

## FAQ

### What did AI actually solve?
OpenAI: a 1946 Erdős geometry conjecture (disproved). DeepMind: nine separate open Erdős problems plus 44 integer-sequence conjectures. Different problems, different methods, both long-standing.

### Is this AGI?
No — DeepMind's CEO said so directly. It's a narrow, powerful result: AI generating mathematics that a formal checker or expert can verify, not a general-purpose thinking machine.