Author: Zhiding Liang, Jinglei Cheng, Rui Yang, Hang Ren, Zhixin Song, Di Wu, Xuehai Qian, Tongyang Li, Yiyu Shi

Original Paper

This paper discusses the potential of large language models (LLMs), specifically generative pretrained transformers (GPTs), in the field of quantum computing. The authors propose a Quantum GPT-Guided Architecture Search (QGAS) model that utilizes GPT-4 to recommend high-quality ansatz architectures for variational quantum algorithms (VQAs). The ansatz architecture is a crucial component of quantum computing and determines the efficiency and accuracy of quantum algorithms. The authors conduct experiments using a series of application benchmarks, including portfolio optimization, the MaxCut problem, the Traveling Salesman Problem (TSP), and the estimation of molecule ground state energy for Lithium Hydride (LiH) and Water (H2O). They compare the performance of the ansatz architectures generated by QGAS with existing ansatzes and state-of-the-art ansatz architecture search methods. The results show that QGAS outperforms other ansatz architectures in some benchmark applications, demonstrating the potential of LLMs in quantum architecture design. The authors highlight the importance of human feedback in guiding the performance of GPT-4. Human experts provide specific guidance and feedback to improve the search strategies and evaluate the generated ansatz architectures. The iterative feedback loop between human experts and GPT-4 leads to better performance and optimization of the quantum circuits. The paper also discusses the limitations of GPT in the field of quantum computing. GPT is not a general artificial intelligence and cannot think dynamically about quantum physics or make accurate predictions about scientific phenomena in quantum experiments. It also relies on large-scale data models, which may contain biased or misleading information about quantum computing. The authors suggest future directions for the integration of LLMs, such as GPT, in quantum computing. They propose that GPT can be used to design and optimize fault-tolerant quantum algorithms and assist in the calibration of quantum hardware. They also envision GPT playing a role in the simulation of quantum computers and providing agile validation of algorithmic innovations. In conclusion, this paper highlights the potential of LLMs, specifically GPT, in the field of quantum computing. The QGAS model demonstrates the effectiveness of using GPT-4 to generate high-performance ansatz architectures for quantum algorithms. The integration of human feedback and the power of GPT-4 provides a promising avenue for advancing quantum architecture design and optimization. However, the limitations of GPT and the challenges of applying LLMs to quantum computing should be considered. The authors suggest further research and development to leverage the capabilities of GPT and address the limitations to fully harness the potential of LLMs in quantum computing.