Large Language Models in Rheumatology: Prospects and Challenges for the Clinician

Artificial intelligence (AI) is rapidly evolving and reshaping everyday life for everyone, including healthcare. The general-purpose models, such as the GPT series, Claude, Gemini, Grok, and others, have demonstrated remarkable capabilities in the solution of medical problems and presenting opportunities and challenges for medical research and clinical practice [1]. Rheumatology, characterized by complex chronic diseases, benefits a lot from rapid advances in LLMs, but must also carefully manage the associated risks. Include applications in Table 1. Despite rapid and extensive resource investment in AI research, significant limitations persist. First, a primary concern for general use remains the validation of AI accuracy. AI hallucination has yet to be fully resolved within current training paradigms, although numerous approaches to solve this issue have already been proposed. Second, the various training datasets lead to data bias, potentially exacerbating health inequities. Third, cloud-based deployment raises significant concerns regarding patient privacy and data security. Conversely, running AI locally is often constrained by limited computing power on personal devices, thereby further restricting its practical application. Fourth, even with current reasoning models, the internal cognitive mechanisms remain inadequately understood. Therefore, these concerns must be carefully considered when applying AI to clinical scenarios. LLMs offer significant promise as tools for rheumatology, with potential improvement of clinical practice, medical research, and personal life, but several challenges remain. To effectively utilize these technologies, their differences in ability, accuracy, adherence to the highest ethical standards, and flexibility to continuous learning need to be carefully evaluated. Po-Cheng Shih wrote the manuscript. James Cheng Chung Wei supervise and edit the manuscript. All authors are responsible for the final version of the manuscript. Use of AI in revising draft. The views expressed in this article represent those of the authors and do not necessarily reflect the official position of the institutions with which they are affiliated. Dr. James Wei is the Editor-in-Chief of IJRD and a co-author of this article. They were excluded from editorial decision-making related to the acceptance and publication of this article. Editorial decision-making was handled independently by other editors to minimize bias. The other authors declare no conflicts of interest. The data that support the findings of this study are available from the corresponding author upon reasonable request.