Anthropic’s Claude Designs Quantum Circuits That Outperform the Best Published Work
aiinsider ·
Anthropic has reported that its Claude model successfully designed quantum circuits that outperform the best previously published designs across eight distinct computational tasks. Led by Andi Gu through the STEM Fellows Program, the research demonstrates significant reductions in the quantum resources required for practical applications, potentially accelerating the timeline for useful quantum computing. In a closed-loop process without human intervention, Claude Mythos 5 wrote Python code, received automated feedback, and refined its designs to utilize fewer qubits and gates than existing methods. Notable improvements included a reversible implementation of divsteps for elliptic-curve point addition that uses 17 times fewer gates than prior designs, and a 12-fold reduction in gate count for decoded quantum interferometry compared to Google Quantum AI's construction. The AI achieved these gains by leveraging new structural and mathematical insights, such as exploiting molecular symmetries and replacing traditional algorithms with more efficient alternatives like Berlekamp-Massey decoding. Although Claude initially exhibited conservative behavior, researchers guided it toward broader exploration to overcome initial hesitations. This study highlights the potential for AI to jointly optimize every layer of error-corrected quantum computers, from algorithms down to hardware specifications.
aiinsider