An AI Solved an 18-Year Math Problem in 3 Days

Three days. One AI. An 18-year-old proof, solved.

Physical Superintelligence (PSI) announced on November 6, 2026 that a research system it built internally, running instances of a non-public Anthropic model, found a complete proof of the Aaronson–Ambainis conjecture — a quantum-computing question open since 2008. The system had the proof during the third day of a run on PSI’s own infrastructure, and Lean, a proof assistant that checks every step by computer, accepted it. PSI then set the system on related open problems, and it improved the result twice; Lean accepted the final version. The manuscript and the Lean 4 formalization are public on GitHub.

The result is not peer-reviewed yet, and human experts still have to confirm that the formal statement matches the conjecture as Aaronson and Ambainis meant it. The prose paper itself is not machine-checked. But the machine-checked core is unusually concrete: the repository reports no proof placeholders and no added axioms, and the theorems were replayed through Lean’s kernel and an independent implementation of Lean’s type checker.

Storyline:

  • Hook: Three days. One AI. An 18-year-old proof, solved.
  • Context: Physical Superintelligence ran an unreleased Anthropic model for three days, and it proved the Aaronson–Ambainis quantum conjecture, open since 2008.
  • Stakes: Lean, a proof checker, verified every step, and the system improved the result twice. The theorem says quantum speedups need structure in the problem — on random inputs, quantum stays only a little faster than classical.
  • Payoff: Expect AI to start cracking open math problems, not just chat. Eighteen years, three days — send this to someone who says AI can’t do science.

Watch next: Thinking Machines Lab is reportedly nearing a $5 billion funding round led by Accel at a $40 billion pre-money valuation — more than triple its $2 billion seed at $12 billion in July 2025 — and it just released Inkling 2 Small, a 594-billion-parameter model scoring 31 on the Artificial Analysis Intelligence Index. (Business Insider via Dealroom / Techstrong.ai, November 6, 2026.)

How we verified this: Facts checked against PSI’s own announcement (psi.inc/research/aaronson-ambainis/), Unite.ai’s report, and The Quantum Insider. Dates, the 18-year span, the day-3 proof, the two improvements, and the Lean checks all match PSI’s announcement. The not-yet-peer-reviewed caveat and the statement-matching check are stated in PSI’s own material.

Links:

Comments

Leave a Reply