Vibeleaderboard
← All Intel
Intel / article

Navier–Stokes Lost in Translation

Source
arxiv.org
Author
nill0
Date

Context

A paper posted to arXiv on October 6 by Alexander Bastounis, Fabian Circelli and Anders Hansen argues that a computer-checked proof can fail to confirm the human-written argument it was supposed to check. The paper, Navier-Stokes lost in translation, concerns autoformalisation: using AI to translate a natural-language math argument into Lean, a formal language whose proofs a computer verifies mechanically.

The authors' point is that Lean verifies only what was written in Lean. If the translation changed a statement, a passing check says little about the original argument. They argue that guaranteeing a faithful translation is "harder than any computational problem", illustrate this with AI mistranslations, and say that for OpenAI's announced Navier-Stokes blow-up proof the Lean version does not correspond to the natural-language proof.

This is the authors' argument in a preprint, not an adjudicated result. The summary rests on the abstract, which does not show whether their specific mismatch claims hold.

Recommended reads
Comments

Checking sign-in…

Loading comments…