AI Lab PSI Solves Decades-Old Quantum Computing Conjecture, Aims to Commercialize Science

Physical Superintelligence (PSI), a startup lab that builds AI tools for scientific research, announced that it has resolved the Aaronson-Ambainis conjecture, a problem in quantum query complexity that has been open for 18 years. The breakthrough was achieved with early access to Anthropic’s Fable 5.2 model and marks one of the first instances where an AI system has produced a proof that settles a long-standing question in quantum computing.

The proof was generated by David Turturean, a recent MIT graduate who deferred a master’s program at the university’s computer science and AI lab to join PSI as technical staff. Using the Fable 5.2 model, Turturean produced a solution that was auto-formalized in the Lean proof language and then reviewed by a team of leading researchers. He told TechCrunch that the result may be the last academic paper he ever publishes, reflecting his belief that AI will soon replace traditional human-driven proof work.

The conjecture concerns whether quantum computers can achieve provable speedups over classical machines for certain query problems. According to its co-originator, Scott Aaronson, the new proof “closes the door on one avenue quantum computers could follow to solve certain difficult problems at high speed, but not all of them.” Aaronson expressed a mix of satisfaction and fatigue, noting surprise that the problem had not appeared in OpenAI’s recent “Mathocalypse” release, which he assumed would have captured it if it were easier than other solved problems.

PSI’s business model currently relies on applying its AI-enhanced simulation tools to data-center optimization, where it models heat, power, water and sound to improve site performance and community impact. The company views these contracts as a near-term revenue source while it builds a longer-term pipeline that converts theoretical physics breakthroughs into marketable technologies. CEO Matt Pines said investors were initially skeptical, forcing him to “stretch to validate” the claim that theoretical physics would soon become “mostly the domain of machines.” He believes that a robust research system must combine specialized model harnesses, memory systems and simulation infrastructure to explore problems with both breadth and depth.

The success in quantum computing does not automatically translate to other scientific fields. Pines acknowledged that physics and the life sciences pose additional hurdles because experimental validation must contend with the “vote” of nature. Last week, competitor startup Discovered Materials announced a $1 million grand challenge to spur AI research on superconductivity and heat dissipation, underscoring the broader industry effort to bridge computational insight and real-world experiments.

Standards for AI-driven science are still emerging. PSI delayed public disclosure of its proof until Anthropic’s model was near release, then shared the result with experts to ensure proper peer review. Jim Halverson, a Northeastern University professor who advises PSI, will attend an MIT summit where physicists will discuss how to respond to the accelerating pace of AI-generated research. He emphasized transparency and collaboration as provisional best practices.

Looking ahead, PSI aims to turn AI-elicited facts of nature into exploitable technologies, creating a feedback loop where machine-suggested discoveries lead to new devices. As the company refines its “neolab” approach, founders and investors will watch whether the model can scale beyond abstract mathematics into the messier realms of experimental physics and beyond.