(Video) SoDa Symposium Seed Grant Series: How Can Large Language Models Help Us Identify and Use Constructs That We Can Trust?

Maia Johnston - May 16, 2024

Computational linguistics experts speak on a new approach to large language models

SoDa Symposium over a yellow background.

On April 25, 2024, UMD’s Social Data Science Center (SoDa) hosted a virtual symposium as part of a series that highlights the recipients of the SoDa Seed Grant Award. SoDa is a partnership between UMD’s College of Information Studies and Joint Program in Methodology, where research in social data science, public health, economics, transportation, and education is conducted.

The panelists of this symposium were UMD Professor in the Linguistic Department and Institute for Advanced Computer Studies Phillip Resnik, UMD PhD student Alexander Hoyle, and industry expert Andrew Stavisky. They discussed a new research approach to using large language models, where AI is partnered with human expertise to identify and work with constructs in a way that is both efficient and trustworthy. Real world examples were provided to showcase how this approach can be applied to daily operations. 

“The most important thing to know about large language models is that all language models predict the next word based on context…without any reference to a model of truth” Resnik shared. 

View previous SoDa symposiums and events here

Watch the full video here.