Decoding ChatGPT: A primer on large language models for clinicians

Abstract

The rapid progress of artificial intelligence (AI) and the adoption of Large Language Models (LLMs) suggests that these technologies will transform healthcare in the coming years. We present a primer on LLMs for clinicians, focusing on OpenAI's Generative Pretrained Transformer-4 (GPT-4) model which powers ChatGPT as a use-case, as it has already seen record-breaking uptake in usage. ChatGPT generates natural-sounding text based on patterns observed from vast amounts of training data. The core strengths of ChatGPT and LLMs in healthcare applications include summarization and text generation, rapid adaptation and learning, and ease of customization and integration into existing applications. However, clinicians should also recognize the limitations of LLMs, most notably concerns about inaccuracy, privacy, accountability, transparency, and explainability. Clinicians must embrace the opportunity to explore, engage, and lead in the responsible integration of LLMs, harnessing their potential to revolutionize patient care and drive advancements in an ever-evolving healthcare landscape.