Hidden Anthropic Model Proves Aaronson-Ambainis Quantum Conjecture

A team of researchers at Physical Superintelligence says their AI solved a long-standing open problem in quantum computing theory.

They used a version of an Anthropic model, one that was never released publicly, to solve the Aaronson-Ambainis conjecture. This is a mathematical conjecture in quantum computing. It had been unsolved since 2008. The Anthropic model worked on it for three days. At the end, it produced a complete solution. A Physical Superintelligence researcher named David Turturean had framed the problem. He also checked the solution himself. After that, two separate automated proof checkers verified the result: Lean and lean4lean.

In this case, humans only framed the problem and checked the results. The model did the actual research work. This fits a pattern now. Important research is increasingly done with Anthropic's models, such as Claude, rather than with public GPT or Gemini models.