AI
AI startup PSI solves long-standing quantum computing problem, signaling shift in scientific research
A newly formed company is commercializing AI-assisted scientific discovery, starting with settling an 18-year-old conjecture in quantum computing theory.
TechCrunch — AI
Physical Superintelligence (PSI), a computational research startup, has leveraged advanced language models to resolve the Aaronson-Ambainis conjecture—a foundational open question in quantum computing that had resisted solution for nearly two decades. The breakthrough was achieved by David Turturean, an MIT graduate now working at PSI, who used early access to Anthropic's Claude model to crack the problem and then had the solution formally verified using Lean, a formal verification language.
The Problem and Its Resolution#
The Aaronson-Ambainis conjecture explored whether quantum computers possess inherent computational advantages for solving specific problem classes. Rather than demonstrating new quantum capabilities, the proof establishes that a particular avenue for quantum speedup does not exist—what researchers call a negative result. This kind of finding, while not opening new possibilities, narrows the landscape for where researchers should focus their efforts. Scott Aaronson, one of the conjecture's original authors, confirmed that he finds the proof sound despite acknowledging he cannot fully follow every step, citing the formal verification in code as lending credibility to the result.
A Business Model for AI-Assisted Science#
PSI's approach differs from other AI labs by building what CEO Matt Pines describes as specialized "harnesses and scaffolding" around general-purpose models. Rather than relying solely on raw language model capabilities, the company constructs domain-specific tools, memory systems, and simulation infrastructure to tackle scientific problems at scale.
Currently, PSI derives revenue from optimizing data center operations—managing variables like thermal management, power consumption, and water usage where physics directly governs facility performance. The company's longer-term ambition, however, is to convert theoretical breakthroughs into practical technologies and products.
Emerging Standards for AI-Driven Science#
The rapid pace of machine-assisted discovery is outstripping existing norms around validation and publication. PSI coordinated with leading researchers before public disclosure of their solution, a decision motivated by academic transparency principles. Jim Halverson, a physicist at Northeastern University advising PSI, notes that the field is still establishing best practices for this new paradigm and plans to discuss standards at an upcoming MIT summit of leading physicists.

