Date of Award
6-9-2022
Document Type
Thesis
Publisher
Santa Clara : Santa Clara University, 2022.
Department
Computer Science and Engineering
First Advisor
Maya Ackerman
Abstract
Language has a profound impact on how we perceive the world. With GPT- 3’s rise in popularity, present in 300 applications averaging 4.5 billion words per day, it is critical for us as programmers to identify and correct biases in its generations. A variety of biases have been identified in generative language models, spanning biases based on gender, race, and religion. Our project pioneers the study of the Brilliance Bias for generative models. This implicit, yet powerful bias imposes the idea of “brilliance” being a male trait and in turn, sets back women’s achievements starting as young as 5-7 years. Our analysis reveals the presence of substantial Brilliance Bias in GPT-3 generations of stories. To address this challenge, we present Brilliance-Equalizer which can be utilized in conjunction with any generative model to counter the presence of the Brilliance Bias.
Recommended Citation
Troske, Ashley; Gonzalez, Edith; and Lawson, Nicole, "Brilliance Bias in GPT-3" (2022). Computer Science and Engineering Senior Theses. 221.
https://scholarcommons.scu.edu/cseng_senior/221