Anthropic Claims Claude Designs Efficient Quantum Circuits
Anthropic stated that its Claude Mythos AI model has designed highly efficient quantum circuits, potentially allowing quantum computing to become accessible "sooner."
The company provided eight natural language specifications for circuits from chemistry, optimization, physical simulation, and number theory to Claude Mythos 5 as part of the STEM Fellows program, equipped with an automatic evaluator for checking and scoring.
The model independently proposed designs in a closed loop, received feedback, and iteratively improved, outperforming previously published circuits in seven tasks, with some improvements reaching an order of magnitude or more, including a maximum reduction of 17 times in gate count.
More efficient circuits reduce the number of qubits and gate resources required for quantum algorithms, bringing computations needed for applications like chemistry and materials science closer to current hardware capabilities.
This work was led by Andi Gu, in collaboration with Harvard University and QuEra Computing; the longest single run lasted about 53 hours with minimal human intervention.
Anthropic claims that such improvements can reduce the resource requirements for useful quantum computing, thereby accelerating the arrival of practical quantum computers.
Source: Public Information
ABAB AI Insight
Anthropic applies Claude Mythos 5 to quantum circuit design through the STEM Fellows program, where the model iteratively generates and optimizes circuit code after receiving natural language task descriptions, outperforming the best published designs in seven out of eight fundamental computational primitives.
On the resource path, optimization uses the product of the number of qubits and gates as a cost proxy, with the model reducing gate counts by up to 17 times in some tasks (such as modular inversion and DQI decoders), while slightly increasing qubits in individual cases; results are scored after verification of correctness by the evaluator, reducing human intervention.
This is similar to the practice of AI-assisted optimization of classical algorithms or circuit designs, or the ongoing improvements in resource efficiency within the quantum algorithm community (such as previous circuit constructions by Google); it is currently at a stage where AI is expanding from code generation and validation to optimizing quantum hardware resources, shifting circuit design from manual expert work to model-driven closed-loop searches.
Essentially, this represents a technological substitution: AI shifts quantum circuits from static results of manual design and publication to an iterative optimization search process, making resource consumption (gates and qubits) a directly optimizable target; lower resource requirements shorten the distance from algorithm theory to hardware runnable applications, advancing the timeline for practical issues like chemical simulation and optimization.
ABAB News · Cognitive Law
Practical quantum computing timelines are advanced when circuit resources are reduced.
Closed-loop evaluation allows AI to independently optimize published designs.
Reduction in gate count has a greater leverage effect than the increase in qubits.