Anthropic says its Claude model designed quantum circuits that beat the best published ones
The company says Claude improved on the best published designs for eight quantum circuits, in some cases by an order of magnitude or more. The results come from a single company report.
Anthropic says its model Claude Mythos 5 designed quantum circuits that use fewer qubits and fewer gates than the best previously published designs, according to a research post from the company.
The company said it gave Claude plain-language specifications for eight circuits, covering chemistry, optimization, physics simulation, number theory and numerical computing. It also gave the model an automated evaluator that checked each design and scored it. Anthropic said Claude worked in a closed loop, proposing a design, getting feedback and a score, and revising it with no human input between rounds. According to the post, some of the improvements were an order of magnitude or more.
In one task, a calculation called point addition on elliptic curves, Anthropic said Claude worked for 53 hours with minimal human input and arrived at a new approach to modular inversion, the most expensive step in that calculation. In another, the company said Claude made the costliest step of a leading quantum optimization algorithm about 12 times cheaper.
Anthropic said the work was led by Andi Gu as part of its STEM Fellows Program, with mentorship from Alek Kemeny, a quantum physicist at the company. It was done with Susanne Yelin of Harvard University and Mikhail Lukin of Harvard University and QuEra Computing. The company said a full paper is on arXiv.
The reason this matters, according to the post, is that today's quantum computers have only hundreds to thousands of qubits, and those qubits are easily disturbed by noise. More efficient circuits reduce how much hardware and computing time useful applications need. Anthropic pointed to past progress on simulating FeMoco, a cluster at the heart of an enzyme that bacteria use to make ammonia from nitrogen. It said a 2017 estimate put that job at about 50 trillion Toffoli gates and that later advances brought it down to about 340 million.
Anthropic argued that improvements like Claude's mean "practical quantum computers could arrive sooner than previously anticipated."
Some things are still unclear. This story rests on one report, and the company that wrote it also makes the model being tested. The post does not say whether the paper has been peer reviewed or whether outside researchers have checked the results. It describes designs being checked by an evaluator program, not run on a quantum computer. The copy of the post used for this story also stops partway through its explanation of the modular inversion result, so the full details of that approach are not covered here.