An Analysis of Bias Towards Women in Large Language Models

UPDATE: This event will be virtual-only this month.

An Analysis of Bias Towards Women in Large Language Models
The popularity of closed-source large language models (LLMs) built by major technology companies continues to rise, alongside growing discussions over the safety of their outputs. This presentation evaluates three leading closed-source LLMs: OpenAI's ChatGPT, Google's Gemini, and Anthropic's Claude, by using established psychological measures of gender bias to design prompts and then evaluate the models' responses. Using the Ambivalent Sexism Index, Modern Sexism Scale, and Belief in Sexism Shift assessments, it was examined how each model responded to prompts reflecting both traditional and modern forms of sexism. Both Likert-scale ordinal ratings and free-response outputs were collected to capture the models' underlying reasoning. Thematic analysis of the free-response data was also conducted to identify recurring patterns.

Attendance:
We will be broadcasting this presentation via Zoom, beginning promptly at 7:00PM PT. We will send the Zoom link out a few hours before the event starts to people who have RSVP'd.