IQN (NewsDesk): Researchers from Harvard University, Anthropic and QuEra Computing say they have used a large language model to discover quantum circuits that need far fewer resources than the best designs written by hand. The work was released on Saturday through a technical paper on arXiv and a separate release from Anthropic.
The team built an evolutionary framework in which the AI model proposes new circuit code and an independent classical evaluator checks each proposal. According to the study, this pairing cut the space and time cost of core computational building blocks by between 4.1 times and more than 9,100 times compared with hand crafted baselines.
In some cases the AI designed circuits that beat the best published versions by an order of magnitude or more. The team also reported that one discovery made the costliest step of a leading quantum optimization algorithm roughly twelve times cheaper. The model used, Claude Mythos 5, was reported to refine designs on its own by spotting underlying mathematical structure.
The findings address a central problem in the field, the trade off between the number of qubits a circuit uses and the time it takes to run. Lower resource costs could bring useful applications such as quantum chemistry closer to the hardware that exists or is expected soon. The results are from a preprint and have not yet been through formal peer review.
