PSI Proves 2008 Quantum Conjecture in 3 Days Using Unreleased Anthropic Model

Q: Which unreleased Anthropic AI model quietly solved a decades-old quantum computing mystery in just three days?

It has not been publicly identified. PSI said only that it used a “non-public” or unreleased Anthropic model, and Anthropic had not commented. What is known is that Physical Superintelligence Inc. said its internal research system, using instances of that unreleased Anthropic model, produced a proof of the Aaronson-Ambainis conjecture during the third day of a run. PSI said the proof was then formally checked in Lean 4 and later improved before the final version was accepted. The result is significant because the Aaronson-Ambainis conjecture, open since 2008, addresses when quantum computers can achieve exponential speedups over classical computers. PSI says the proof shows that such large speedups require special structure or patterns in the problem, limiting quantum advantage on random or unstructured inputs. Independent details in the available reporting support the central claim that the model remains unnamed. Multiple accounts, including PSI’s own materials and follow-up coverage, say the system used a non-public Anthropic model but do not specify whether it was a Claude-branded release or another internal model. Researchers familiar with the problem reportedly called the result a major advance, but the identity of the Anthropic model has not been disclosed.

3 articles · Updated · Business Insider · Nov 7

Physical Superintelligence Inc. said its research system, built around an unreleased Anthropic model, proved the Aaronson-Ambainis conjecture within three days.
The 2008 conjecture asks when quantum computers can be dramatically faster than classical ones; PSI said the proof shows such exponential gains require special structure, setting clearer limits on quantum advantage.
Researchers previously working on the problem called the result a huge breakthrough, with Stanford's Guy Blanc, Jordan Docter, Carmen Strassle and Li-Yang Tan saying the proof brings ideas together in a "magical" way.
The claim adds to a run of AI-assisted math advances after OpenAI said in September it solved a 90-year-old Navier-Stokes problem, intensifying debate over how AI will reshape mathematical research and careers.