Researcher admits SETH/CSETH mixup, revises odds in P vs NP debate to >80% on CSETH
geoffreyirving · x · 2026-10-28
Researcher Geoffrey Irving, discussing odds on P vs NP, admits his earlier >99% confidence in the Exponential Time Hypothesis (ETH) contained significant "model error" — he had conflated SETH with a stronger circuit conjecture he calls CSETH (n-variable, n^(1+δ)-gate circuit SAT can't be solved in (2-ε)^n time). Using a ChatGPT-elicited consensus estimate: P(ETH | P≠NP) ≈ 90%, P(ETH | P=NP) = 0%, a 10:1 likelihood ratio favoring P=NP. After correcting for what he learned about the NP/ETH gap, he now puts >80% on CSETH, conceding he "ate a ton of Bayes points."
Related event: Researcher Rethinks Bayesian Odds on P vs NP After Confidence Blunder(2 posts)→
More from Fun
- Steering Vector Recipe Turns Qwen 27B Into "Evil Qwen" — basedjensen · 2026-10-28
- Mathematician's tongue-in-cheek guide to AI-proofing your research — NC_Renic · 2026-10-28
- Mystery domain poteto.com goes live, redirects to ChatGPT after a few clicks — umesh_ai · 2026-10-28
- Need for Speed: Most Wanted (2005) now playable in your browser, iOS Safari included — CatAstro_Piyush · 2026-10-28
- Yacine: more compute means more work, not less — yacineMTB · 2026-10-28
- ChapterPal dev emails authors for hosting rights: half say yes, AI haters rage — burkov · 2026-10-28