GPT and Claude bet 95%+ on an unsolved 1742 math problem — until 'I don't know' is allowed
Nir777 · reddit · 2026-11-02
A video tests four top models on a Goldbach-style question: whether a five-line program ever halts — it stops only on an even number that isn't the sum of two primes, and computers have verified up to 4 quintillion without finding one. GPT and Claude confidently bet 95%+ that it never stops, but when "I don't know" is explicitly allowed, the models' answers diverge. The video explores whether frontier models admit ignorance on undecidable math.
More from Models
- "No one wants an AI traffic cop": devs argue open source wins over moderation — adamamcbride · 2026-11-02
- OpenAI keeps ChatGPT instant unlimited for Pro users, separate from Codex quota — shaunralston · 2026-11-02
- Same intelligence gets 10x cheaper every 4-6 months, says Artificial Analysis — altryne · 2026-11-02
- Grok usage limits jumping overnight: users report weekly quota 34%→73% with no activity — heyshrutimishra · 2026-11-02
- Google's Stock Slump Shows Search Economy Collapsing Faster Than Expected, Says Fu Sheng — FuSheng_0306 · 2026-11-02
- Mystery model files surface on Hugging Face, unverified for 9+ days — Subulicious · 2026-11-02