JMIR Publications

ABSTRACT

Background:

OpenAI’s ChatGPT and other Generative Language Models (GLMs) have rapidly increased in popularity. Such language models have the potential to greatly benefit the medical field. Despite the rapid rise in the implementation of such technologies, no standardized framework currently exists to discuss the prompting techniques for these models.

Objective:

We aim to establish the nomenclature of “variable” and “clause” in prompting a generative language model while providing example interviews that outline the utility of said approaches in medical applications. This goal is key to implementing GLMs in the medical field, as they provide grounds for assessing prompting methodologies.

Methods:

The specific GLM assessed in this paper was OpenAI’s ChatGPT-4. Definitions of terms utilized in prompting procedures were outlined, including “Input Prompt,” “Variable,” “Demographic Variable and Clause,” “Independent Variable and Clause,” “Dependent Variable and Clause,” “Generative Clause,” and “Output.” This methodology was implemented with the three-example patient cases. Unedited ChatGPT outputs were included.

Results:

As demonstrated in our three cases, precise combinations of variables and clauses that consider patients’ age, gender, weight, height, and education level can yield unique outputs. The software can do so quickly and in a personalized, patient-specific manner. Our findings demonstrate that GLMs can be used to generate comprehensive sets of educational material to combat current limitations with the potential of improving healthcare outcomes as they are further explored.

Conclusions:

To our knowledge, this is the first attempt to standardize medical inputs into a GLM. Doing so opens the potential for outlining patient-specific information in medical queries that can better tailor responses to individual patients. Additionally, throughout this paper, we recommended several avenues that we believe would provide the greatest insight into future research of GLMs, such as ChatGPT. Most notably, future projects must look at the specialty and presentation-specific input changes that might produce the best outputs for desired goals.