AI designs quantum circuits: Anthropic's Claude beats records on gate counts
In a 53-hour experiment, Claude Mythos 5 designed building blocks for quantum computers that need far fewer operations. They have not yet been tested on real hardware.
Archive image · Photo: QuEra Computing, Inc. / Wikimedia Commons, CC BY 4.0 (cropped and resized)
Anthropic has had its AI model Claude Mythos 5 design circuits for future quantum computers. The experiment was led by researcher Andi Gu through Anthropic's STEM Fellows Program, with researchers from Harvard University and quantum computer maker QuEra involved. Claude was tasked with improving eight computational building blocks from chemistry, physics simulation, optimisation and number theory; an automated evaluator checked each design for correctness and scored its cost.
According to the reports, Claude beat the best published designs on seven of the eight tasks. On the eighth, a circuit for elliptic-curve arithmetic, it accepted 12 percent more qubits in exchange for 17 times fewer gates. According to AI Weekly, Claude made a decoder for the Decoded Quantum Interferometry method about twelve times leaner than an earlier construction by Google.
The circuits were verified computationally but not run on a physical quantum computer; real-world costs depend on factors including error correction. According to AI Weekly, the researchers also had to push Claude to explore fundamentally different approaches in parallel.