Abstract: [Purpose/Significance] Artificial intelligence generated content (AIGC)'s content creation method has brought about a new revolution to the field of library and information science (LIS). Currently, the related research is mainly based on AIGC and ChatGPT, while ERNIE bot and Bard are less studied. Comparative analysis of the advantages and disadvantages of the AIGC large language models, discussion of the operating mechanism of AIGC, and in-depth research on application solutions in the context of university libraries provide new ideas for AIGC applications in smart libraries. [Method/Process] Taking the three AIGC applications of ChatGPT, ERNIE bot and Bard as examples, starting from the Transformer model, and on the basis of in-depth analysis of the basic principles of the large language model, the comparative analysis method is used to conduct a horizontal comparison of these three applications. The research summarizes the six common features of AIGC's large language model, and points out that it can be used in improving the work efficiency of university libraries. This paper explains and identifies nine different characteristics of the AIGC large language model, and points out how to choose three applications in university libraries. According to the characteristics of each application, six scenarios-based application modes of university libraries and the advantages of AIGC applications in university libraries are pointed out. A discussion is provided on four potential risks that may be faced by libraries in using AIGC large language models, and solutions are proposed to reduce risks, providing a reference for university libraries to choose AIGC applications. [Results/Conclusions] ChatGPT focuses on natural language understanding and content generation, and has more advantages in the ability of natural language understanding, task applicability and cross-language transfer, and is more suitable for resource integration and decision-making assistance in the context of knowledge services, subject services and administrative management. ERNIE bot has hundreds of billions of super-training parameters, and it can generate multi-modal content including text, pictures and voices. It has more advantages in learning training, model expansion and Chinese comprehension, and is more suitable for optimizing services and assisting creation in the context of reader services, technical services and cultural services in university libraries. By comparison, Bard focuses on human-machine dialogue data processing, it can use natural language to communicate with people, and it is more suitable for providing 24-hour intelligent customer service, assisting subject consultation and knowledge Q&A in the context of reference consultation in university libraries. With the application of AIGC, although university libraries will face ethical risks, privacy risks, data security, and the proliferation of false knowledge, as long as artificial intelligence data governance is strengthened, in the future, university libraries will integrate the natural language understanding and generation capabilities of AIGC large language models that can expand diversified application scenarios, innovate multi-dimensional service models, optimize the business service environment, assist administrative decision-making, and improve the level of intelligent services.

Key words: ChatGPT, ERNIE bot, Bard, AIGC, smart library