JMIR Publications

ABSTRACT

The integration of large language models (LLMs), as seen with the Generative Pre-trained Transformers (GPT) series, into healthcare education and clinical management represents a transformative potential. While the very applications and direct employment of GPT in various domains of healthcare practices sparked great anticipation for new avenues and promising opportunities, it is of equal significance to constructively dissect its current flaws before envisaging its integration into existing systems. The current review takes a conservative epistemic stance in evaluating the utility of state-of-the-art LLMs – especially GPT3.5 and 4.0 – in healthcare education and practice, for we underscore the following limitations as areas requiring significant and urgent refinement: a) risks for plagiarism and academic dishonesty, b) dissemination of misinformation, c) limited and inconsistent delivery of knowledge, d) inequity of access, e) presence of algorithmic bias and moral instability, f) inadequacies in text-to-image capacity, and g) lack of regulatory measures to address ethical challenges. This paper then exemplifies, through the direct maneuvering of GPT-3.5, how LLMs can be utilized effectively to provide astonishing assistance and advancements to healthcare education and delivery.