Qortora · Search · Indexed page
blog.iroha1203.dev Fetched 2026-08-17T13:58:02Z
Uncharted Waters: 13 Rules for Not Trusting an AI's Proofs 13 field-tested rules for running AI agent loops you can't trust: an AI proved 7 theorems in 15 days in Lean 4, all past adversarial review.
Open original source · Full cached text
Uncharted Waters: 13 Rules for Not Trusting an AI's Proofs Command Palette Search for a command to run... HHiroyuki Nakahata I’m a software engineer in Japan. Outside of work, I enjoy horse racing. ORCID:0009-0008-5928-0234 Outline Command Palette Uncharted Waters: Thirteen Rules for Not Trusting an AI's Proofs TL;DR Introduction Part 1: The Thirteen Rules Rules of preparation Rule 1 — Split the state three ways Rule 2 — Never let the loop rewrite the spec Rule 3 — Design failure as a deliverable Rule 4 — One obligation per cycle Caveats Rules of verification Presented by Qortora, a product of Qortora, LLC. Content remains the property of the original publisher. This reference page supports transparent discovery within the Qortora index.