Lunch at 12:30pm, talk at 1pm, in 148 Fitzpatrick

Title: Some thoughts on chain-of-thought

Abstract: I will share some thoughts to start a discussion on (1) some interesting problems that GPT cannot solve; (2) what neural network and formal language theory have to say about them; (3) what could be done to solve them. The problems all relate to GPT’s inability to maintain an internal state. Two existing solutions are recurrence and chain-of-thought prompting. I will explain how these two address different problems, and how neither is really what we want. I won’t offer any alternative solutions, but will open up the floor to discussing ideas.

Bio: David Chiang is an Associate Professor in the Department of Computer Science and Engineering at the University of Notre Dame, and leads the Natural Language Processing Group. His research is in natural language processing, the subfield of computer science that aims to enable computers to understand and produce human language. He focuses mainly on language translation, and is interested in syntactic parsing and other areas as well.