With the breakthrough of the transformer infrastructure, explosion in natural language processing development, and race to produce larger language models (LLMs), critical AI studies have increasingly scrutinized bias embedded within AI training data and output.

This project explores how such representational bias provides insight for historians working with generative AI trained on historical data. Working with over ten thousand digitized issues of the University of Florida’s century-long student-run newspaper, The Florida Alligator and The Independent Florida Alligator, this project provides the issues as context within OpenAI’s GPT-4o mini model and requests the LLM to produce one hundred profiles of a UF alum. Analyzing the generated profiles for explicit and subtle representational bias yielded insight into the gendered representations of students.