July 20, 2026
Proof? More like proof of chaos
Human mathematicians are being outcounterexampled
AI just crashed math’s party, and the comments are split between awe and eye-rolls
TLDR: An AI-assisted result reportedly knocked down a famous math idea, then other AI tools rapidly converted the argument into machine-checked proof code. Commenters are split between “this is old news with more compute” and “this changes everything,” with others arguing faster debunking could save researchers years of wasted work.
Math world just got a reality-TV moment: a chatbot reportedly helped knock down a famous old math claim, and then other AI systems rushed in to turn that result into machine-checked proof code at breakneck speed. To fans, this looked like a jaw-dropping “the future is here” moment. To skeptics, not so fast — the comments quickly turned into a cage match over whether this is a revolution or just a flashier version of what computers have already been doing for years.
The biggest divide? One camp basically said, calm down, computers have been finding counterexamples forever. Their take was that the real novelty is not magical intelligence, but bigger computing power and better timing: AI is trendy, so now everyone is paying attention. Another camp argued this is still a huge shift, because older tools needed humans to narrow the hunt first, while now AI is helping with that “where should we even look?” step. In plain English: the machine isn’t just checking boxes — it’s helping pick the boxes.
Then came the darker, juicier subplot: commenters pointed out how wrong math can wreck real lives. One brought up Yitang Zhang, saying a mistaken result in work around another famous problem derailed years of his career. That made the pro-AI crowd sound almost moral: if bots can find false ideas faster, maybe they save humans from wasting years chasing ghosts. And yes, there was plenty of meme energy too: if AI can topple one giant math belief now, commenters are already half-joking, half-panicking about which sacred cow gets gored next.
Key Points
- •The article says ChatGPT produced a counterexample to Erdős’ Unit Distance conjecture on 20 May 2026 using a theorem by Golod and Shafarevich.
- •Logical Intelligence, via Mike Freedman, is said to have autoformalized the ChatGPT-generated paper in Lean within a week of the announcement.
- •The author highlights that the underlying mathematics depends on difficult global class field theory, an area whose formalization was still incomplete in 2025-2026.
- •Boris Alexeev announced on 26 June 2026 that OpenAI’s model Sol had produced a complete Lean formalization of the Erdős counterexample from the axioms of mathematics.
- •The article reports that Sol generated 1.2 million lines of Lean code in three weeks and that the author verified in a sandbox that it proved nontrivial theorems about cohomology of number fields.