David Turturean, a recent MIT graduate working at the Physical Superintelligence startup, used an early version of Anthropic’s Fable 5.2 model to solve a problem in quantum computing that had stood for eighteen years.
Turturean told TechCrunch he is fully accepting the idea that this might be the last paper he publishes. In a few months, he questioned why anyone would even put out a paper.
The challenge, known as the Aaronson-Ambainis conjecture, looks at whether quantum computers can solve certain problems faster than classical ones. Scott Aaronson, one of the two researchers who posed the problem, writes that he does not quite understand the proof. He believes it is correct because it was autoformalised in Lean, a coding language for mathematics, and explained by a team of top researchers with whom Physical Superintelligence shared the solution.
The solution appears to close the door on one avenue quantum computers could follow to solve certain difficult problems at high speed, but not all of them. Turturean said it is a negative result that allows people to know they are no longer looking there.
More broadly, it reflects the upheaval in the scientific world caused by AI problem-solving.
A proof of the full Aaronson-Ambainis conjecture was one of the last remaining open problems in quantum query complexity that everyone seemed to agree was centrally important. Aaronson, who described himself feeling satisfied, awestruck, disappointed, and fatigued by the news, wrote he was surprised to see it missing from OpenAI’s Mathocalypse dump a few weeks ago. He asked if they tried it and if it was really so much harder than all the problems that were solved.
That is a point in favour of Physical Superintelligence’s model, which is to solve scientific problems with specialised model harnesses and scaffolding.
An effective kind of research system should have lots of things related to harnesses, lots related to tools, memory systems, simulation infrastructure. PSI CEO Matt Pines told TechCrunch you want to be able to really go after a problem at breadth and depth, and be able to consider a wide range of potential alternatives.
For now, the company’s revenue comes from using its tools to optimise data centres, simulating the variables of heat, power, water, sound. The performance of that site as well as its impact on the community are a direct function of its physics. The long play, however, is converting theoretical physics breakthroughs into actual technologies.
Six to nine months ago, when I was talking to investors, I had to sort of stretch to validate that conditional. I had to say no, no, no, trust me, pretty soon, theoretical physics is going to be mostly the domain of machines. Pines said it is not there yet, but it is on its way.
Still, while AI has claimed important solutions in fields like mathematics and computer science where it is possible to express them in validated code, doing so in less abstracted disciplines like the life sciences or physics has proven more difficult. Last week, the start-up Discovered Materials released a $1 million grand challenge for researchers using AI to understand problems like superconductivity and heat dissipation.
At the end of the day, physics is different than math because it is the real world. Nature gets a vote in your models of reality. Pines said the startup is focused on understanding how to take computational results and validate them with experiments in the most efficient way possible.
If it is a true fact of nature, that is a trick that you can exploit. You could build a device out of it. The AI suggests an idea; it allows you to do something in the world you weren’t able to do before. Each of those are kind of loops that PSI is building.
As neolabs for science become a pathway for potentially groundbreaking science, the participants are still figuring out what the standards are for this new world. Because Physical Superintelligence had to hold its result until Anthropic’s model was nearly ready for release, they were able to share it with relevant researchers to better understand it.
As an academic, we wanted to play our role in doing things the right way. Jim Halverson, a professor at Northeastern University who advises Physical Superintelligence, said he will be attending a summit at MIT next week where leading physicists will try to hash out a response to the new pace of science. He said best practices are still emerging. They figured that transparency is a good way to go, and contacting experts is a good way to go.
We had some time to sit on it. It was the first time that I actually took one of my AI-elicited solutions on print, and I read through it, and I was like, okay, how can I make this be presented more naturally for a human? Turturean said the academic community will expect him to be able to explain the result. He told TechCrunch he thinks it also made him realise the weight of what AI can do.
