Startup PSI Uses Anthropic Model to Prove a Long-Open Quantum Computing Conjecture

Scott Aaronson, the University of Texas computer scientist who posed the conjecture with Andris Ambainis, confirmed the result on his blog on November 8 under the title "Aaronson-Ambainis Conjecture proved! (By AI, of course)". He wrote that the proof was obtained with PSI's proprietary research harness running on an internal Claude model, that it comes with a machine-checked certificate in the Lean proof language, and that human experts have reviewed and understood it. David Turturean, a recent MIT graduate on PSI's technical staff, told TechCrunch he used early access to Anthropic's Fable 5.2 to reach the solution.

The result matters less for what it does to quantum computing than for what it says about who is now solving open problems. Aaronson wrote that a full proof was "one of the last remaining open problems in quantum query complexity that everyone seemed to agree was centrally important." It shows that one class of quantum speed-ups, which most researchers did not expect to exist, is ruled out. Aaronson stressed that it has no effect on Shor's algorithm and criticised claims that it ends hopes of quantum advantage. "It's a negative result that allows people to know, okay, we're no longer looking there," Turturean said.

Aaronson described feeling satisfied, awestruck, disappointed that humans did not get to the conjecture first, and fatigued by the volume of open problems now falling to AI systems. He told readers he was surprised the problem was missing from what he called "OpenAI's Mathocalypse dump" a few weeks earlier.

PSI's pitch is that specialised scaffolding around frontier models, rather than the models alone, is what cracks such problems. "An effective kind of research system should have lots of things related to harnesses, lots related to tools, memory systems, simulation infrastructure," chief executive Matt Pines told TechCrunch.

The company currently earns revenue by using its tools to optimise data centre sites, simulating heat, power, water and noise. Its longer-term aim is to turn breakthroughs in theoretical physics into technologies. Pines acknowledged that physics is harder than mathematics for machines because results must be checked against experiment. "At the end of the day, physics is different than math because it's the real world … nature gets a vote in your models of reality," he said, adding that PSI is working on the most efficient ways to validate computational results in the lab.

The episode also exposed how unsettled the norms are for publishing AI-assisted science. PSI had to hold the result until Anthropic's model was close to release, which gave it time to share the proof with outside experts before announcing it. Jim Halverson, a Northeastern University professor who advises PSI and whom Aaronson credited alongside Turturean, said transparency and early contact with specialists seemed the right approach while "best practices are still emerging."

Turturean, who deferred a master's place at MIT's computer science and AI laboratory to join PSI, said he accepts the paper may be his last. "In a few months, like, why would you even put out a paper?" he said.

Halverson is due to attend a summit at MIT next week at which leading physicists will discuss how to respond to the faster pace of AI-driven research.