Claude设计更高效量子电路，部分改进达到数量级突破

Anthropic研究团队利用Claude AI模型进行量子电路设计，在八个计算原语上实现了显著优化 1。这些改进在某些情况下相比已发表的最佳算法达到一个数量级或更高的提升 1。Claude通过发现新的数学洞察和结构优化，独立完成了模块逆元反向实现和Reed-Solomon解码器优化等关键设计任务 1。

在具体案例中，Claude用53小时完全自主设计了点加法电路，仅需初始说明和偶尔提示 1。在FeMoco模拟项目上，门数估计从2017年的约50万亿Toffoli门降低至约3.4亿，降幅约15万倍 1。解码量子干涉（DQI）中的Reed-Solomon解码器门数相比Google先前设计减少12倍 1。Claude采用Berlekamp-Massey解码方法使电路无需为不同输入采取不同路径，更适合量子电路实现 1。

此外，Claude通过模拟403百万次采样精确测量异常值出现频率，将量子位需求从worst-case的665个优化至19个 1。这些优化使量子算法运行所需资源大幅减少 1。

Anthropic's research team has demonstrated that Claude, an AI model, can design significantly more efficient quantum circuits by discovering novel mathematical insights and structural optimizations across eight computational primitives.1 In several cases, Claude's designs represent improvements of over an order of magnitude compared to previously published algorithms.1

The AI model independently completed critical design work including modular inversion circuit optimization and Reed-Solomon decoder enhancement.1 For modular inversion, Claude required 53 hours of autonomous work with only initial instructions and occasional guidance to produce the final design.1 In a simulation of FeMoco, the estimated gate count dropped from approximately 500 trillion Toffoli gates in 2017 to roughly 340 million, representing a reduction of approximately 150,000 times.1 For Reed-Solomon decoding in Decoding Quantum Interference (DQI), Claude reduced the gate count by 12 times relative to Google's previous design through the use of the Berlekamp-Massey decoding method, which eliminates the need for different circuit paths depending on input variations—a significant advantage for quantum implementations.1

In another optimization task, Claude measured anomaly occurrence frequency through 403 million sample simulations and refined the circuit from a worst-case requirement of 665 qubits down to 19 qubits.1 These advances demonstrate that Claude's ability to identify algorithmic efficiencies and resource reductions could substantially lower the computational resources required for quantum algorithm execution.