Opus 5.5 spots a missing factor of 2 in a 2015 paper's bound during math checking

xennygrimmato_ · x · 2026-11-23

ptrschmdtnlsn shares that Claude Opus 5.5, while checking a mathematical bound, claimed a 2015 paper actually dropped the factor of 2 from the bound, so testing <= log(n)^2 would suffice. He notes the number threshold (3000) caveat but says for <64-bit integers you'd just use the 10 bases that work deterministically. The author flags he hasn't double-checked the claim — a live observation of LLM-assisted math verification.

Original post →

More from Models

Models channel →