Presentation
Aligning Pedagogy with Generative AI: An Approach to Customizing Educational GPTs
DescriptionThis study presents an approach for crafting system prompts to customize Generative Pre-Trained Transformers (GPTs) for teaching purposes. As Generative AI (GenAI) becomes increasingly used as teaching tools, educators must understand GPTs limitations, operational logic, and biases to create learning experiences that are helpful, honest, and harmless. Many educators often deploy default GPT models without any form of contextual training, risking outputs that are misaligned, unreliable, or even harmful. To address this, the study provides an approach for educators to systematically train and evaluate customized GPTs for their own educational contexts. This approach translates pedagogical theories, learning objectives, and instructional strategies into structured prompts and curated knowledge bases. A case study illustrates how this approach was used to customize a GPT to teach the rhetorical goals of research writing. The customized GPT demonstrated a stronger pedagogy, persona, contextual awareness, and objective alignment compared to the default GPT model; however, it still violated key rules under adversarial queries. This study provides mitigation recommendations for structuring and refining system prompts to strengthen GPT compliance. By offering an accessible way to customize GPTs without complex technical interventions, this framework allows educators to leverage their expertise while mitigating risks and biases.
Event Type
Industry/Practitioner Content
Lecture
TimeTuesday, October 14th4:50pm - 5:10pm CDT
LocationGrand Hall M/N
Training
Similar Presentations