Claude Audits Quantum Gravity, Finds Errors.
This open-source project uses Claude Opus 5.5 to re-derive a published one-loop quantum gravity calculation across 18 tasks, producing 183 pages of derivations and 38 reproducible checks. It found a sign error and missing term while confirming the original late-time physical conclusion. [GitHub repository](https://github.com/eokahya/claude-audits-quantum-gravity)
The breakthrough is the reproducible audit workflow, not simply the volume of AI-generated math. Its limits matter too: this was a bounded calculation rechecked by one of its authors, not proof that AI can independently replace physicists. [MACCUS analysis](https://maccus.fr/en/claude-quantum-gravity-audit/)
- –Independent SymPy/mpmath checks make the model’s derivations inspectable and repeatable.
- –The audit classified substantive errors separately from counterterm-absorbed mistakes and typographical errors.
- –The corrected equations did not alter the original late-time physics conclusion.
- –The strongest future test is whether researchers unfamiliar with the original work can reproduce the findings.
DISCOVERED
1h ago
2026-10-27
PUBLISHED
1h ago
2026-10-27
RELEVANCE
AUTHOR
AI Revolution
