When ChatGPT is told to keep trying, it can produce a very confident fake proof
soumitrashukla9 · x · 2026-07-23
A meme-like screenshot shows someone prompting ChatGPT to “keep trying” whenever it says it can’t solve a problem.
The result is a mock-serious research-style answer: “complete finite counterexample,” followed by a long symbolic claim that looks like it escaped from a proof assistant. The joke is that relentless prompting can push the model into increasingly overconfident nonsense.
More from Fun
- Von Neumann's 1947 math quote resurfaces alongside rare 1952 IAS computer photo — pickover · 2026-10-27
- Proposal: improv talk track where you present AI-generated slides of a random paper, authors in front row — windx0303 · 2026-10-27
- Naming your agent swarm Philosopher King and SWE lead won't fix that none of them listen — teortaxesTex · 2026-10-27
- Jason Rohrer hand-codes a chess engine in C89, no libs, no AI — for fun — jasonrohrer · 2026-10-27
- €800k enterprise AI consulting deal: 18 months of workshops booked, still pre-code and pre-ROI — alex_verem · 2026-10-27
- VC Jason offers to help a $86k-in-debt dad, then sends him $69 — Sentdex · 2026-10-27