A researcher at Physical Superintelligence (PSI) used early access to Anthropic’s unreleased Fable 5.2 model to solve a long-standing problem in quantum computing, the startup announced in November 2026. The result marks one of the more significant demonstrations of AI-driven scientific discovery to date.
David Turturean, a recent MIT graduate who deferred a master’s program at MIT’s computer science and AI lab to join PSI, used the model to crack the Aaronson-Ambainis conjecture — an open problem in quantum query complexity that researchers had pursued for nearly 18 years. The conjecture concerns whether quantum computers can solve certain problems faster than classical ones. The result is a negative one, ruling out one specific avenue quantum computers might use to achieve speed advantages. “It’s a negative result that allows people to know, okay, we’re no longer looking there,” Turturean said.
Scott Aaronson, one of the two researchers who originally posed the problem, said he does not fully understand the proof but believes it is correct. The solution was autoformalized in Lean, a coding language used to verify mathematical proofs, and reviewed by a team of researchers with whom PSI shared the result ahead of publication.
PSI CEO Matt Pines said the company’s approach relies on specialized model harnesses, memory systems, and simulation infrastructure to pursue scientific problems at both breadth and depth. The startup currently generates revenue by using its tools to optimize data centers — simulating variables like heat, power, water, and sound. Its longer-term goal is converting physics breakthroughs into deployable technologies.
The result also raised questions about emerging standards in AI-assisted science. Jim Halverson, a Northeastern University professor who advises PSI, said best practices are still being worked out, and noted he will attend a summit at MIT where physicists plan to discuss how to respond to the accelerating pace of AI-driven discovery. “Transparency is a good way to go, and contacting experts is a good way to go,” he said.
Pines acknowledged that validating computational results with real-world experiments remains a key challenge, particularly in physics. “Nature gets a vote in your models of reality,” he said.
Source: TechCrunch