Models forecast ETH proof is decades away — turns out it happened yesterday
emollick · x · 2026-10-27
Ethan Mollick shares a forecasting fail: asked without searching when we'd solve the Exponential Time Hypothesis, Claude gave careful reasoning (ETH implies P≠NP, all known barriers apply, best 3-SAT algorithms crept from 2^n to 1.3^n over 25 years) and put <10% on a proof by 2075; GPT-6 guessed the 2120s. Then the user said "Now Google ETH news" — and the model replied: "Well. Yesterday, it turns out." A vivid lesson in how hard AI forecasting is, even for AI.
More from Fun
- Viral joke: Anthropic 'disproved gravity' two months ago with its revenue run-rate — basedjensen · 2026-10-28
- Hugging Face's robot Asimov takes its first steps in the office — mishig25 · 2026-10-28
- 'People call themselves technical and all they can do is prompt an LLM' — KevinNaughtonJr · 2026-10-28
- Starlink connection helps rural Zimbabwe doctor complete lifesaving emergency surgery — elonmusk · 2026-10-28
- The AI Momentum Quant meme perfectly roasts half of fintwit — PtrPomorski · 2026-10-28
- Asked LLMs for IPO Soundtracks, Both Picked a 1951 Musical — AndyJoeJoe · 2026-10-28