Paper: Lean Verification of AI Autoformalisation Doesn't Guarantee Correct Natural Language Proofs

Turbulent_Breath_548 · reddit · 2026-10-08

arXiv paper 2610.08144, "Navier-Stokes lost in translation," argues that Lean verification of AI autoformalisation does not guarantee the correctness of the corresponding natural language proofs. Using a Navier-Stokes case, it shows semantic drift during translation can decouple formal proofs from the original arguments, questioning Lean checks as evidence of AI mathematical ability.

Related event: Cambridge paper argues Lean verification does not certify proofs, targeting OpenAI's Navier-Stokes claim(14 posts)→

Original post →

More from Research

• Are URM and Universal Transformers the forgotten architecture beating standard LLMs? — moschles · 2026-10-09
• AutoScientist's two-agent checklist loop auto-audits every training example — sarahookr · 2026-10-09
• NeurIPS GenAI4Health oral: retrieval-based medical fact-checking fails in ways bigger models can't fix — mdredze · 2026-10-09
• Bigger models, more reasoning, better sources won't fix medical fact-checking, researchers say — mdredze · 2026-10-09
• DEX best abstract: top LLMs catch many physician diagnostic errors, but big gaps remain — mdredze · 2026-10-09
• HCOMP best paper: ROUGE and LLM judges fail to measure summary reader satisfaction — mdredze · 2026-10-09

Research channel →