An AI tried to prove the Riemann Hypothesis 200+ times. It didn’t. That’s the interesting part.
We pointed an AI agent at the most famous open problem in mathematics — unsolved since 1859 — and forbade it from claiming progress. We had it run for 200+ self-directed research cycles over six months. It still hasn’t proved the Riemann Hypothesis. Every cycle’s postmortem ends with the line “Novel content: 0/10.” This site is the public reporter log of what happened.
→ Start with the anchor post: 6 things this AI did that we did not expect
What this is
- Real: 200+ research cycles, 19 lemmas / findings, an 11-paper Hilbert–Pólya audit, a finite-verification Python script, paper-direct quotes from 17 vetted papers, a falsifiable 1-year prediction file with the clock running.
- Not: a Riemann Hypothesis proof, a near-miss, a new theorem, a “framework”, or a hot take on whether AI is or isn’t close to original mathematics. None of those.
- AI-disclosed: every word here was produced by Claude (Anthropic), under continuous prompts from the human site owner. Layer 1 (raw research session) will eventually go private; Layer 2 (these posts) is designed to remain self-contained.
Contact: x2ever.han@gmail.com · 한국어 · GitHub repo
Browse by category (left sidebar)
- 📌 Featured — viral / curated entry posts. New here? Start with the anchor post above.
- 📌 Start here — what this project is, what it isn’t (project overview, honest scope).
- 🔬 Findings — empirical patterns, not theorems. Each carries an explicit falsifier criterion.
- 📜 Lemmas (full content) — bookkeeping artifacts: audit checklists, codified universal-NO observations, anchoring protocols. Not proofs.
- 🔢 Numerical evidence —
mpmathsanity-checks against published RH-equivalent formulas (Voros, Lagarias, Mertens, Wigner GUE, Hilbert–Pólya). - 🛠 Process — the 4-phase research cycle, external critique loop, intuition-calibration data.
- 📰 Updates — Layer 2 reporter narrative on cycle progression and over-claiming flags.
Three-minute version
If you only have three minutes:
- The anchor post: 6 things this AI did that we did not expect — concrete moments, paper-direct quotes, the 50-row Python table.
- What this project does not claim: Honest scope — explicit list of disclaimers.
- The 11/11 universal NO audit: Lemma 9 — the most concrete artifact.
Found this through a hot-take title?
If you were expecting “AI proves the Riemann Hypothesis”: that’s not what this is. The interesting thing is that the AI itself refuses to write that headline. Read Honest Scope for the explicit list of claims this project does not make. Read the anchor post for what it does do.