Cristi Isaula-Reyes
Pronouns: she/her/hers
Research Mentor(s): Renee Li
Research Mentor School/College/Department: UMSI / Information
Program:
Authors: Qiwei Li, Cristi Isaula-Reyes
Session: Session 7: 4:40 pm – 5:30 pm
Poster: 45
Abstract
Can language models be used collaboratively? We build and test the first collaborative prompting that allows multiple users to asynchronously share data for few-shot learning supported by a large language model. We take content moderation as a testbed for collaborative prompting. Content creators encounter challenge of moderating comments, grappling with the emotional toll of sifting through toxic and disruptive content while trying to curate their desired community. Existing word filters lack contextual understanding and can lead to flagging comments for having a certain keyword despite being used in a completely different context. Our work addresses this issue by creating a Chrome extension that uses language models to analyze comments in their context and flag hateful comments. Collaborative prompting enables content creators to share and learn from similar content removals during self comment moderation, fostering a collaborative community for improved customization. Collaborative prompting is not only valuable for creator content moderation but also represents a novel interaction between groups of people and a language model.
