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

Physical Superintelligence (PSI), a startup lab, says a member of its technical staff used early access to Anthropic's unreleased Fable 5.2 model to resolve the Aaronson-Ambainis conjecture, a quantum computing question that had stood open for eighteen years. David Turturean, a recent MIT graduate who deferred a master's at MIT's computer science and AI lab to join PSI, told TechCrunch the result may end his academic publishing habit: "I'm totally accepting the idea that this might be the last paper I ever put out."

The conjecture concerns how much faster quantum computers can solve certain problems than classical machines. Scott Aaronson, who posed the problem, wrote that he does not quite understand the proof yet believes it is correct, because it was autoformalized in Lean and explained by a team of top researchers with whom PSI shared the solution. Aaronson described his reaction as satisfied, awestruck, disappointed, and fatigued.

The answer is a negative result: it forecloses one avenue quantum computers might have followed to solve certain difficult problems at high speed, while leaving others open. "It's a negative result that allows people to know, okay, we're no longer looking there," Turturean said. Aaronson also wrote that he was surprised the problem was missing from OpenAI's Mathocalypse dump a few weeks earlier, asking whether it was really so much harder than the problems that were solved — a point in favor of PSI's approach.

PSI's thesis is that research capability comes from scaffolding as much as from the model itself. "An effective kind of research system should have lots of things related to harnesses, lots related to tools, memory systems, simulation infrastructure," CEO Matt Pines said, adding that a system must go after a problem at breadth and depth. Revenue today comes from using those tools to optimize data centers — simulating heat, power, water, and sound, where a site's performance and its community impact are direct functions of its physics. The long play is converting theoretical physics breakthroughs into actual technologies.

Pines said the investor conversation has moved quickly: six to nine months ago he had to stretch to validate the claim that theoretical physics would soon be mostly the domain of machines, and while "it's not there yet... it's on its way." The limits remain concrete — AI has claimed important solutions in mathematics and computer science, where problems can be expressed in validated code, while less abstracted disciplines like the life sciences and physics have proven harder. Last week, startup Discovered Materials released a $1 million grand challenge for researchers using AI on problems such as superconductivity and heat dissipation. What to watch: whether PSI turns this proof into the commercial technology Pines promises, and whether its harness-heavy system replicates the win beyond quantum query complexity.