Amateur armed with ChatGPT solves an Erdős problem

· ai · Source ↗

TLDR

  • ChatGPT proved an open Erdős conjecture in number theory using a method no human expert had previously considered, in what researchers are calling a ‘vibe-mathing’ advance.

Key Takeaways

  • An amateur used ChatGPT to produce a valid proof of an Erdős conjecture, a class of problem that has stumped professional mathematicians for decades.
  • The proof method was entirely novel: no human had previously attempted or even conceived of the approach the model used.
  • Experts were surprised and believe the technique may generalize to other open problems beyond this specific conjecture.
  • The term ‘vibe-mathing’ signals a recognized pattern: AI reaching correct results via unconventional reasoning paths that bypass established mathematical tradition.

Hacker News Comment Review

  • Commenters split on interpretation: some argue LLMs naturally escape expert blind spots because they are not locked into a single research tradition the way specialists are, making genuinely novel approaches more likely.
  • Survivorship bias concern surfaces: many people are currently running AI on hard unsolved math; this publicized success may be one hit among many silent failures rather than evidence of reliable capability.
  • No consensus on whether this is a repeatable research method or a striking outlier, but the ‘additive thinking’ framing – humans build on prior foundations and get stuck, LLMs do not – gets traction as a partial explanation.

Notable Comments

  • @adamgordonbell: Published the full prompt used – no web search allowed, explicit instruction for “non-trivial, novel and creative proofs” in number theory and primitive sets; model thought for 80 minutes 17 seconds.

Original | Discuss on HN