Workshop Highlight: Cutting Large Language Models Down to Size

Written by Paige Kupas, MDI Journalism Intern

This Fall, Massive Data Institute’s Workshop Series focused on using text as data. For the last installment of the series, MDI Data Science Fellow Dr. Nathan Wycoff led participants through interactive lessons on ChatGPT over the course of two days.

Wycoff’s workshop, “Cutting Large Language Models Down to Size,” gave Georgetown students, fellows, staff, and faculty in attendance the opportunity to learn how ChatGPT operates, work on their own code alongside Wycoff, and discuss the societal and philosophical implications of ChatGPT. 

“I hope that the series was able to provide attendees with a high-level understanding of the various approaches to working with text data, from tried and true methods to more recent developments like my workshop,” Wycoff said. “I also hope students enjoyed the ride from the precise technical details of how [ChatGPT] functions to high-level discussions about the future of society.”

Valeria Vera Lagos, a graduate student in data science, attended all three workshops hosted by MDI this semester because she is interested in large language models (LLMs), a type of artificial intelligence (AI) that tries to predict every word in the sequence given previous words as a way to construct answers to different questions. She found Wycoff’s workshop to be valuable for understanding the math behind LLMs and continued to attend each workshop throughout the series to learn how to fine-tune the models. 

Another attendee, MDI postdoctoral fellow Rupayan Mallick, came to the workshop because it aligns with his academic focus on generative AI and because he wanted to dive more deeply into ChatGPT, a popular technology that uses LLMs.

“LLMs are a big point of interest to me. The biggest takeaway from the lecture was how ChatGPT can be used and how ChatGPT can be misused,” Mallick said. 

Wycoff said that he and the participants all learned from each other, too. For instance, the group brought up interesting questions about the evolution of AI and how its accuracy has improved since its initial development. 

Overall, Wycoff hopes that if there is one takeaway from his workshop, it is that “ChatGPT is powerful but not magical.”

Stay tuned for information about the upcoming workshops in Spring 2024!

MDI Workshops
MDI Workshops Fall 2023