David Turturean, a MIT graduate and member of the technical team at the startup lab Physical Superintelligence (PSI), solved an 18-year-old quantum computing challenge using Anthropic’s Fable 5.2 model.
“I fully accept the idea that this might be the last paper I ever publish,” Turturean told TechCrunch. He had originally postponed his plan to pursue a master’s degree at MIT’s renowned Computer Science and Artificial Intelligence Laboratory in favor of joining PSI. “In a few months, why would you even bother publishing a paper?”
This problem is the Aaronson-Ambainis conjecture, which concerns whether quantum computers can solve problems faster than classical computers. One of the two researchers who posed this question, Scott Aaronson, wrote that he does not fully understand the proof, but he believes it is correct because it has been automatically formalized using Lean, a mathematical coding language, and explained by a group of leading researchers who shared the solution with PSI.
This answer appears to close off one path for quantum computers to solve certain difficult problems at high speed, but not all paths. “It’s a negative result that tells us, okay, we no longer need to pursue that direction,” Turturean said.
More broadly, this also reflects the impact that AI-driven problem-solving is having on the scientific community.
“Proving the full [Aaronson-Ambainis conjecture] is one of the last few remaining unsolved problems in quantum query complexity that everyone agrees is critically important,” wrote Aaronson. He said his reaction to the news included satisfaction, awe, disappointment, and exhaustion. “To be honest, I was surprised it wasn’t included in OpenAI’s recent Mathocalypse roundup. They must have tried it—so is it really much harder than all those other problems that have already been solved?”
This also supports PSI’s model: addressing scientific problems through specialized model orchestration and supporting systems.
PSI CEO Matt Pines told TechCrunch: “An effective research system should have a lot of components related to the orchestration layer, tools, memory systems, and simulation infrastructure. You need to be able to tackle a problem thoroughly, both in breadth and depth, and consider various potential alternatives.”
Currently, the company generates revenue by using its tools to optimize data centers, simulating variables such as heat, electricity, water, and noise—“the performance of these sites and their impact on the community are direct functions of their physical characteristics.” However, in the long term, the company aims to translate breakthroughs in theoretical physics into practical technologies.
Six to nine months ago, when I spoke with investors, I still had to work hard to validate that condition, right? For example, 'No, no, no, believe me, theoretical physics will soon be primarily the domain of machines.'” Pines said. “We’re not there yet, but we’re on the way.”
However, although AI has achieved significant breakthroughs in fields like mathematics and computer science—where problems can be expressed in verifiable code—it remains more challenging to achieve similar results in disciplines with lower levels of abstraction, such as life sciences or physics. Last week, the startup Discovered Materials launched a $1 million challenge targeting researchers using AI to study issues like superconductivity and thermal dissipation.
“In the end, physics is different from mathematics because it deals with the real world… nature will vote on your model of reality,” said Pines. He noted that the startup focuses on researching how to most efficiently combine computational results with experimental validation.
“If it’s a real fact of nature, then it’s a trick you can exploit,” Pines said. “You can build a device based on it, right? AI proposes an idea; it enables you to do things in the real world that were previously impossible… So these are all the loops PSI is building.”
As "neolabs in the field of science" emerge as a potential pathway to breakthrough scientific achievements, participants are still navigating the standards of this new landscape. Since PSI must wait until Anthropic’s model is nearing release to disclose results, they are able to share findings with relevant researchers to foster deeper understanding.
“As a scholar, we want to play our part in getting things right,” said Professor Jim Halverson of Northeastern University, who advises PSI. He will travel to MIT next week to attend a summit where leading physicists will discuss how to respond to the accelerating pace of science. “Best practices are still evolving… we believe that maintaining transparency and engaging with experts are good approaches.”
“We had some time to set it aside for now… This was the first time I printed out a solution I got from AI, read through it, and thought, ‘Okay, how do I make this… more natural for humans?’” Turturean told TechCrunch. He noted that academia would expect him to explain this result. “It made me realize just how significant what AI can do really is.”