Startup Physical Superintelligence Uses Advanced AI to Solve Physics Puzzle
A recent MIT graduate at startup Physical Superintelligence used early access to Anthropic's Fable 5.2 model to solve an 18-year-old quantum computing problem.
A scientific startup lab named Physical Superintelligence (PSI) has demonstrated how automated systems can tackle long-standing theoretical problems. David Turturean, a recent MIT graduate and PSI technical staff member, utilized early access to Anthropic’s Fable 5.2 model to crack the Aaronson-Ambainis conjecture, an 18-year-old problem in quantum query complexity.
Turturean opted to defer his master's studies at MIT’s computer science and AI lab to join PSI. Reflecting on the speed of automated research, he remarked that this might be the final paper he ever publishes, questioning the necessity of traditional academic publishing in the near future. The conjecture itself addresses whether quantum computers can always outperform classical ones for specific problems. Scott Aaronson, one of the original researchers who formulated the challenge, admitted he does not completely grasp the proof himself. However, he trusts its validity because it was autoformalized in Lean—a mathematical coding language—and reviewed by independent experts consulted by PSI.
The resulting proof offers a negative answer for a specific pathway in quantum computing, indicating researchers should look elsewhere. Aaronson described experiencing mixed feelings of satisfaction, awe, disappointment, and fatigue upon hearing the news, noting his surprise that the problem was apparently missed by OpenAI's recent math release.
Building Systems for Scientific Discovery
PSI approaches scientific problem-solving by deploying specialized model harnesses and scaffolding systems. According to PSI CEO Matt Pines, an effective research environment requires tools, memory systems, simulation infrastructure, and harnesses that allow an AI to examine problems across both breadth and depth.
At present, PSI generates revenue by applying its technology to optimize data centers, simulating variables like heat, power, water, and acoustics. However, the company's broader ambition is translating theoretical physics insights into tangible technologies. Pines noted that while convincing investors six to nine months ago that theoretical physics would soon enter the domain of machines required persuasion, that transition is now underway.
Applying AI to the physical world presents distinct challenges compared to pure mathematics, where solutions can be cleanly expressed in verified code. Unlike abstract math, physical sciences must account for reality. Pines emphasized that PSI is actively studying how to connect computational outcomes with efficient real-world experimental validation, creating feedback loops that could eventually yield novel hardware devices.
What it means for developers
The emergence of specialized AI harnesses and automated reasoning points to a shift in how complex technical domains will be approached. For software engineers and technical builders, integrating advanced models into structured workflows or simulation pipelines is becoming a viable path for tackling difficult problem spaces. Developers looking to experiment with these capabilities can try top AI models cheaply through one API at https://apixoai.online.
As these methods evolve, the scientific community is actively working to establish new standards and best practices. Because PSI held its findings until the Anthropic model neared public release, the team could consult external experts like Northeastern University professor Jim Halverson to ensure proper peer transparency. Meanwhile, researchers continue to adapt to a reality where AI-generated proofs require new approaches to human comprehension and academic validation.
Source: PSI is trying to make a business out of AI-driven scientific discovery — TechCrunch AI. Written by the Apixo team from that report.
One key for Claude, GPT, GLM, DeepSeek and more. Pay per token with crypto.
Get your API key
