Anthropic’s Claude Designs More Efficient Quantum Circuits With Automated Feedback
aiinsider ·
Researchers from Anthropic and Harvard University have successfully used the Claude AI model to design more efficient quantum circuits through an automated feedback loop. By pairing the language model with an independent evaluator, the system optimized algorithms across eight computational tasks without requiring immediate access to quantum hardware. The approach significantly reduced the number of costly Toffoli gates and peak qubit requirements in seven of the tasks compared to existing baselines. In one instance, a redesigned decoding routine cut gate usage from 5.1 million to approximately 440,000 while slightly lowering qubit needs. The eighth task resulted in a novel arithmetic approach for modular inversion that improved trade-offs between computing time and memory. Although the results were validated via classical computation, the study highlights how AI can autonomously propose, test, and refine complex technical solutions. Human guidance remained necessary to encourage broader exploration and counter the model's tendency toward familiar approaches. These findings demonstrate a promising workflow for AI-guided scientific discovery, provided reliable evaluators can be established for broader applications.
aiinsider