April 25, 2026
Proof or Prompted Magic?
Amateur armed with ChatGPT solves an Erdős problem
Internet erupts: miracle proof, lucky prompt, or math heresy
TLDR: A 23-year-old using ChatGPT reportedly solved a long-standing Erdős math puzzle with a novel method. Commenters are split between awe and side-eye—some share receipts and celebrate, others question failed attempts, token costs, and whether this counts as “intelligence,” making it a lightning-rod moment.
A 23-year-old with a ChatGPT subscription fed a single prompt into the bot and—boom—an old Paul Erdős puzzle just got cracked. Commenters are split between “holy wow” and “hold my coffee.” One poster even shared the exact testy prompt (“don’t search the internet”), while another dropped an archive link for receipts. The vibe: part victory lap, part courtroom cross‑examination.
Fans say this is the first time the bot served something truly new, not just remixing old math. Even star mathematician Terence Tao hinted humans took a wrong turn early, which has people whispering about a legit fresh approach. The skeptic camp, meanwhile, is tallying the hidden cost: how many failed runs did it take, and how much did those tokens burn? One quip: the AI may be clever, but the bill is doing long division on VC wallets.
Big-picture takes are flying. Some argue humans get stuck building on the same old scaffolding, while a “naïve” large language model (LLM, a text-predicting AI) can wander into weird, useful ideas. Others snark: now watch the crowd invent reasons this isn’t “real” intelligence. Meanwhile, erdosproblems.com is getting rubbernecked like a freeway fender-bender—proof, prompt, and pop‑corn all included.
Key Points
- •Liam Price used ChatGPT Pro (GPT-5.4 Pro) to generate a purported solution to a 60-year-old Erdős conjecture on primitive sets.
- •Experts suggest the AI-produced method appears novel for this class of problems and may have broader applications.
- •The solution was posted on the Erdős problems website about a week prior to publication.
- •Terence Tao noted prior human attempts likely took an early wrong turn, making the problem seem harder than it was.
- •Jared Lichtman previously proved a related 2022 result about the Erdős sum for primes; the newly addressed conjecture concerns a lower bound approaching 1 for sets with large elements.