August 5, 2026
Proofs, panic, and a PR plot?
Why the Legendary Erdős Problems Are Falling to AI
Math world stunned as AI cracks famous puzzles and the comments instantly go feral
TLDR: OpenAI says an internal AI solved and advanced several famous math problems linked to legendary problem-setter Paul Erdős, a moment many see as a major turning point. The comments split hard between awe at what comes next, suspicion that it’s slick AI marketing, and jokes that nobody can even understand the breakthrough anyway.
The big headline is wild enough: OpenAI says an internal AI system found a new counterexample for a famous 1946 Paul Erdős problem, then followed it up with 10 more math advances and three additional Erdős problem solutions. Human mathematicians quickly improved some of the results, but the real shock was that the AI brought in ideas from faraway areas of math that people hadn’t successfully used before. Translation for the rest of us: a machine didn’t just copy homework — it apparently made a clever leap.
But the real fireworks were in the community reaction. One camp was awe-struck, basically saying, “If old math puzzles are dropping now, what happens when AI starts chewing through physics, chemistry, and biology too?” That’s the full sci-fi goosebumps version. Another camp was much less starry-eyed, accusing the whole thing of smelling like a giant PR flex. One especially sharp comment mocked the article’s mention that Bloom used ChatGPT to build his website, joking, “I could not have written this comment without Illy Espresso,” which is an all-timer drag on AI product placement.
Then there was the existential confusion: some researchers reportedly looked at OpenAI’s claimed breakthroughs and admitted they were way over their heads. That led to the thread’s funniest mini-crisis — if a math breakthrough happens and hardly anyone understands it, did it even happen? Add in commenters getting distracted by Erdős himself — the silk clothes, suitcase life, amphetamines, and general lovable-weirdo legend — and the whole thing felt less like a research update and more like the internet arguing over whether genius has entered its robot era
Key Points
- •OpenAI said in May 2026 that one of its internal AI models found a counterexample to Paul Erdős’ 1946 unit distance problem.
- •The article describes that result as the first historically significant proof produced by an AI model, while noting human mathematicians improved it within weeks.
- •OpenAI later announced that an unreleased model called Astra achieved 10 more mathematical advances, including solutions to three additional Erdős problems.
- •The article says leading mathematicians see these developments as a major change in AI’s mathematical capability and in mathematical research workflows.
- •Thomas Bloom’s erdosproblems.com, launched in 2023 to catalog Erdős problems, is presented as an important factor in making those problems a focal point for AI research.