2025 ASEE Annual Conference & Exposition

Generative artificial intelligence (GenAI) has become ubiquitous. Convincing language complemented by constant modifications and upgrades have made GenAI models, such as OpenAI’s ChatGPT and Google’s Gemini, an appealing tool to address complex problems. According to a survey by Intelligent.com nearly a third of college students in AY 2022-2023 used ChatGPT for schoolwork and 77.4% of them were likely to recommend using it to study to another student. Despite their appeal, these models have proven flawed in answering technical prompts. Their convincing language may entice the user to trust the responses without verifying them. For example, the authors failed to retrieve accurate thermodynamics properties of some common substances from three publicly available models (OpenAI’s ChatGPT, Anthropic’s Claude, and Google’s Gemini). Notwithstanding the inaccuracy of the responses, the conversation suggests some promising capabilities. For instance, with instructive prompts by the user, the model was able to reduce its error percentage significantly.
Thermodynamics is one of the early core courses that students in Mechanical Engineering, Chemical Engineering, and Aerospace Engineering, among others take. This study aims to develop a GPT-based model focused on thermodynamics using publicly available resources, such as substance properties. Once proven successful, the model can be adopted by other institutions and adapted to similar courses.