Large language models (LLMs), including ChatGPT, and their effects and uses in education and industry, were the subject of a lively panel held recently in the Bay Area, hosted by the School of Computer, Data & Information Sciences (CDIS) and Computer Sciences Department (CS) the at the University of Wisconsin–Madison.
Entitled “The Future Frontier of Large Language Models,” over 100 alumni and friends attended the panel in person, with more participants tuning in via livestream from across the globe, to hear experts unravel the intricacies and implications of large language learning models and the ongoing importance of fundamental research in data science and machine learning.
CDIS Founding Director Tom Erickson welcomed attendees and presented an overview of CDIS and the school’s new building under construction on the UW-Madison campus. As one of only three universities who have brought together computer science, statistics and information departments, the building will bring all of CDIS (and more) under one roof.
Professor and outgoing Computer Sciences Department Chair Remzi Arpaci-Dusseau shared a department update including the continued growth of computing programs — the biggest and still second fastest growing major on campus. “Our approach is, if a student comes to Wisconsin and wishes to study computer science and has the ability to do that, we’re trying to enable them to,” said Arpaci-Dusseau. Data science is the fastest growing major, which is a joint degree between Statistics and CS.
Moderated by Arpaci-Dusseau, panelists included:
- Srinivas Narayanan (MS ‘96), Vice President of Engineering at OpenAI
- Kangwook Lee, UW-Madison Electrical and Computer Engineering Assistant Professor and CS Affiliate Professor
- Stephen Wright, CS Professor and incoming Department Chair at UW-Madison
The panel discussed research methods and using LLMs, including ChatGPT, their impact on education, and where academics and industry interact in the LLM space. Lee started the discussion on the latter topic, sharing that he doesn’t think there’s a problem with students using ChatGPT for coursework. Rather, the tricky question for teachers is “How should we reformulate the way we teach?”
Another positive for ChatGPT in education is developing new courses, which is time-consuming for instructors. Lee said, “With ChatGPT, I can actually make it much faster…it’s actually super easy to write a new syllabus from scratch,” so what used to take hours he can now complete in 20 minutes. Narayan emphasized that LLMs can help not only students but also teachers.
Naryanan also addressed how he believes AI will impact education. The first is personalization. “The ability to personalize and really meet the person who is learning where they are.” Because personalization is often difficult to scale the ability to so broadly will be transformational. Second, the systems will be interactive and allow that personalization with back and forth questions and direction. And third, is access — the ability to access information and education from anywhere in the world, including low-resource areas.
Wright added that he believes a lot of what we do is not going to be addressable yet with ChatGPT. “I think we still have to tell people how things work” Wright noted. He pointed out that this has happened often before — that people had to use slide rules or pen and paper to do multiplication or long division. “Then calculators came along, and we still told people how to do these things, but we didn’t expect them to do them the old-fashioned way anymore.” He believes ChatGPT will be similar.
Narayan addressed LLMs and the workforce in a similar vein, saying that he thinks ChatGPT will be more assistive than doing entire jobs: “I think it will be a huge productivity improvement for people in the next few years.” Even if we reach a place where ChatGPT can do entire jobs, new jobs will be created. “Humans will always be at the center of this,” he said.
Watch the full panel discussion here.
Thank you to our sponsor for this event, WVV Capital.