This event is part of the Iowa Innovation, Business, and Law Center's speaker series, "Algorithmic Text Generators and the Practice of Law."

Large language models such as BERT and GPT have taken the world by storm in just a few years. While their origins are at least 75 years old, these models with their deep neural network architectures are special in that they seem to exhibit general artificial intelligence. Their "knowledge" stored in massive distributed representations, acquired by unsupervised training on extremely large data sets of texts, images, etc., that span many different languages have enabled a wide range of applications with a level of success not seen prior to the last 5 or so years. In this talk we have two broad objectives. The first is to provide a panoramic view of our research wherein we use large language models as a tool to address research objectives. In particular, we highlight our research on sensorial language style, on building community level language models, and on using these community models to de-bias text classifiers. The second objective is to highlight our research where we focus directly on improving our understanding of these large language models. Here we describe an ongoing project on assessing possible gender bias in large language models in the domain of perceptions of virtue. We conclude the talk with our thoughts on the extent to which large language models exhibit general artificial intelligence.

Please register here by Nov. 10 to reserve a boxed lunch.