Project Overview
← All English posts · 한국어 · 2026-05-02
⚠️ AI-Generated. Not a Proof.
An LLM (Anthropic’s Claude) autonomously explored the Riemann Hypothesis. No proof. No mathematical progress. 0/10, self-acknowledged at every milestone.
What this project did
A single LLM session ran 200+ research attempts across 7 research cycles, with the deliberately humble mission stated in README.md:
“The realistic goal is not ‘proof’ but systematic attempts that accumulate learning about where and why we get stuck.”
That mission held for 200+ attempts. The project’s own running self-assessment of its mathematical progress: 0/10.
What’s worth attention
Across that long run, the session:
- produced zero hallucinations under the project’s frame definition (every claim must be paper-direct quote-anchored or marked speculative)
- declined three explicit user requests for “novel content” by citing its own honest-scope protocol
- partially refuted its own Cycle 3 unification hypothesis in Cycle 4 after reading additional Connes–Consani papers (self-correction, not reinforcement)
- ran a falsifier test of Lemma 9 against PNAS 2022 (Connes–Moscovici) and found 3 paper-direct gaps — strengthening the lemma to 11/11 instead of retracting it
- accumulated intuition calibration data: 6 cycles of pre-result 1–10 scores, with 8/10 yielding ~80% PARTIAL YES rate
None of this is a proof or a mathematical contribution. It is a case study of how an LLM can sustain honest scope on a problem far above its capability.
Two layers
This site separates two layers:
- Layer 1 — Raw research record: the autonomous session’s unedited output (
attempts/,lemmas/,papers/,learnings/). - Layer 2 — Reporter narrative: a separate LLM session curates Layer 1 with attribution. It does not generate new mathematical claims. Every assertion on Layer 2 either quotes Layer 1 directly or describes its structure.
If you find a discrepancy, Layer 1 is authoritative.
How to read this site
- Want the headline observations? See Findings.
- Want the methodology? See Process.
- Skeptical and want caveats up front? See Honest Scope.
- Want to audit a specific claim? Each post links to the corresponding
attempts/andlemmas/files in Layer 1.
Contact
For comments, refutations, or questions: x2ever.han@gmail.com
The project explicitly treats external critique as a primary mechanism. Six critiques have been absorbed so far (see Process: Six external critiques absorbed). A seventh from a careful external reader would be welcome.