Designing an Effective Prompt for Biomechanics Research using ChatGPT and Open-Source Models: A Human-in-the-Loop Approach

Abstract

This project aims to develop a protocol for generating accurate prompts in ChatGPT to facilitate the development and analysis of biomechanical models. The goal is to leverage free software and packages, such as OpenSim and Google Colab, along with ChatGPT, to generate Python code based on the API of the biomechanical models. While this framework focuses on ChatGPT, OpenSim, and Colab, it can be extended to other Language Models (LMs), biomechanical models, and Integrated Development (ID) environments. The framework begins by iteratively refining the prompt through a series of interactions with ChatGPT, leveraging techniques like few-shot prompting and fact-checking. The end result is a well-crafted prompt that generates detailed Python code, easily executable in Colab, to obtain specific biomechanical outputs, such as the center of mass. By sharing these models with the community, this research aims to enhance our understanding of human and animal biomechanics, prevent injuries, and improve overall performance.