Physical Superintelligence (PSI), a startup research lab, says David Turturean — a recent Massachusetts Institute of Technology graduate and member of PSI’s technical staff — used early access to Anthropic’s Fable 5.2 model to produce a solution to a quantum computing problem that has been open for 18 years. Turturean told TechCrunch he delayed plans to pursue a master’s at MIT’s computer science and AI lab in order to join PSI and commented, “I’m totally accepting the idea that this might be the last paper I ever put out.”
The Aaronson–Ambainis conjecture
The specific question addressed is the Aaronson–Ambainis conjecture, which concerns whether and how quantum computers can achieve speedups over classical machines on certain problems. Scott Aaronson, one of the conjecture’s originators, wrote that he does not fully understand the proof but believes it is correct because the solution was autoformalized in Lean (a formal proof language) and was explained to him by a group of leading researchers with whom PSI shared the result.
According to PSI, the proof closes off one route by which quantum machines might have realized certain large speedups. Turturean described the outcome as a “negative result” in the sense that it tells researchers which directions no longer merit further searching.
What this means for AI-driven research
The episode highlights the changing role of AI in scientific problem-solving. Aaronson said that a full proof of the Aaronson–Ambainis conjecture was one of the last central open problems in quantum query complexity; he described feeling a mixture of satisfaction, awe, disappointment, and fatigue at the news. He also noted surprise that the result did not appear in OpenAI’s recent “Mathocalypse” dump, suggesting that while many problems have been solved by AI systems, some tasks remain comparatively harder.
PSI’s approach relies on specialized model harnesses and other tooling to tackle scientific challenges. PSI CEO Matt Pines told TechCrunch that an effective research system needs many harnesses, tools, memory systems and simulation infrastructure to pursue a problem both broadly and deeply and to evaluate a wide range of alternatives.
Business model and practical goals
Currently, PSI generates revenue by using its tooling to optimize data centers — simulating interactions of heat, power, water and noise, where site performance and community impact are governed by physical factors. The company’s longer-term aim is to translate theoretical physics breakthroughs into practical technologies.
Pines said that six to nine months ago he still had to stretch to convince investors that theoretical physics would soon become largely a domain of machines, but he believes the field is moving in that direction even if it hasn’t arrived yet.
Validation challenges and standards
While AI has produced notable solutions in mathematics and computer science where results can be encoded and verified, translating computational discoveries into experimentally validated advances is harder in less abstract fields like the life sciences and experimental physics. Last week, Discovered Materials announced a $1 million grand challenge for researchers using AI to explore problems such as superconductivity and heat dissipation.
Pines emphasized that physics differs from pure math because “nature gets a vote” in models of reality: computational results must still be checked against experiments. PSI focuses on finding efficient ways to validate computational outputs in the lab, with the goal of turning true facts of nature into devices and applications.
Transparency and expert review
As “neolabs for science” emerge, norms for best practice are still forming. PSI delayed publicizing the result until Anthropic’s model was nearly ready for release, which gave them time to share the work with relevant experts and better understand and explain the proof.
Jim Halverson, a Northeastern University professor who advises PSI, said academics involved wanted to “play our role in doing things the right way.” He will attend a summit at MIT where leading physicists will discuss how to respond to the faster pace of AI-enabled science. Halverson and others argue that transparency and expert consultation are sensible early practices.
Turturean said this was the first time he took an AI-elicited solution into print, that he read through and edited it to make it present more naturally for human readers, and that the experience underscored for him the scale of AI’s potential impact.
