A Platform for the Biomedical Application of Large Language Models

May, 2023

Abstract

The wealth of knowledge we have amassed in the context of biomedical science has grown exponentially in the last decades. Consequently, understanding and contextualising scientific results has become increasingly difficult for any single individual. In contrast, current Large Language Models (LLMs) can remember an enormous amount of information, but have notable shortcomings, such as a lack of generalised awareness, logical deficits, and a propensity to hallucinate. To improve biomedical analyses, we propose to combine human ingenuity and machine memory by means of an open and modular conversational platform, ChatGSE (https://chat.biocypher.org). We safeguard against common LLM shortcomings using general and biomedicine-specific measures and allow automated integration of popular bioinformatics methods. Ultimately, we aim to improve the AI-readiness of biomedicine and make LLMs more useful and trustworthy in research applications.