PSI says an Anthropic AI model helped solve a long open quantum question

In short: A startup lab called Physical Superintelligence (PSI) says it used an unreleased Anthropic AI model to solve an 18-year-old research problem in quantum computing.

What happened

David Turturean, a recent MIT graduate working at PSI, said he used early access to Anthropic’s Fable 5.2 model to solve the Aaronson-Ambainis conjecture. This is a long-running question about how much faster a quantum computer can be than a normal computer for certain kinds of tasks.

Quantum computers are a different kind of computer that use the rules of very small particles, and researchers hope they can speed up some hard calculations. This conjecture was one of the key unanswered questions in a subfield that studies how many “checks” a quantum computer needs to answer a question (think of it like how many times you need to peek inside a box to be sure what is inside).

Scott Aaronson, one of the researchers who originally posed the problem, wrote that he does not fully understand the proof. However, he said he believes it is correct because it was “autoformalized” in Lean, which is a math-checking programming language (like a strict spellchecker, but for logic), and because PSI shared the work with expert researchers who explained it.

PSI’s CEO, Matt Pines, said the company is building a “research system” around AI, including tools, memory, and simulation software. PSI currently makes money by using its tools to optimize data centers by simulating heat, power, water, and noise.

Why it matters

If AI systems can reliably help prove or disprove tough scientific ideas, it could change how research gets done and how quickly dead ends get ruled out. PSI says its long-term goal is to turn theoretical results into real technologies, but Pines also noted that physics still needs experiments because nature, not software, decides what is true.

Source: TechCrunch AI