PSI aims to turn AI-led scientific discoveries into business

Turturean, a recent MIT graduate who postponed plans for a master’s degree to join Physical Superintelligence (PSI), used early access to Anthropic’s Fable 5.2 model to tackle the Aaronson-Ambainis conjecture. The problem concerns whether quantum computers can solve certain problems faster than classical computers. The result appears to rule out one possible route to such speedups, though not every route. “It’s a negative result that allows people to know, okay, we’re no longer looking there,” Turturean told TechCrunch.

Scott Aaronson, one of the researchers who posed the conjecture, wrote that he did not fully understand the proof but believed it was correct. He cited its autoformalization in Lean, a language used to verify mathematical proofs, and explanations from researchers with whom PSI shared the work. Aaronson called the problem one of the last open questions in quantum query complexity that researchers broadly considered central.

PSI’s approach is to pair models with specialized tools, memory and simulation systems. CEO Matt Pines told TechCrunch that an effective research system should be able to examine problems “at breadth and depth” and weigh many alternatives. The company currently earns revenue by using its tools to optimize data centers, modeling factors including heat, power, water and sound. Its longer-term ambition is to turn theoretical physics breakthroughs into technologies.

That remains a difficult step. AI has produced notable results in mathematics and computer science, where claims can be checked through formal code, but validating discoveries in fields such as physics and life sciences is harder. Pines said PSI is working on ways to test computational results through experiments. The startup also shared its proof with relevant researchers; adviser Jim Halverson said standards for this faster pace of science are still emerging and that transparency and expert review were important.