Collapse, Not Computation · source record · webpage

An OpenAI model has disproved a central conjecture in discrete geometry

OpenAI · 2026

url

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

verification note

Announcement dated 20 May 2026. Two direct fetches of the OpenAI page timed out during this import, so the page text itself was NOT read end-to-end by this agent; the claim is instead anchored on independent verification (see SRC-SAWIN-2026, an arXiv paper by a mathematician unaffiliated with OpenAI that states the result and its attribution) and on Quanta Magazine's 3 August 2026 reporting, which was read directly. Per Sawin's paper, the OpenAI team was Lijie Chen (using an internal model) with Mark Sellke and Mehtaab Sawhney verifying correctness. The labeling here is deliberately ai_verified-with-caveat rather than verified, and the primary evidential weight is placed on the arXiv paper.

reliability notes

• A company announcement about its own unreleased internal model — the least independent link in this chain, which is why the mathematics is cited through Sawin instead.
• The result itself does not depend on OpenAI's word: the disproof has been checked by outside mathematicians, a companion paper was written by nine of them, and the bound has since been sharpened independently.
• Context that belongs on this record: an earlier round of AI Erdős-problem claims (October 2025) collapsed when the 'solutions' were found already in the literature, and a widely-publicized Christmas 2025 claim on Erdős Problem 333 was withdrawn within hours after a user identified a 1977 resolution by Erdős himself. This evidence line has a documented false-positive history.