Title: Reasoning in Large Language Models: Insights for and from Human Cognition

Abstract: Over the past few years, large language models have achieved human-level performance on a number of domain-general natural language processing tasks as well as on a variety of domain-specific language-based benchmarks, such as in medicine and law. Despite their success, there is debate regarding whether these models are in fact capable of reasoning and world knowledge. This talk will discuss the ways in which cognitive science and neuroscience can be leveraged to evaluate, dissociate, and potentially improve different reasoning capabilities in LLMs, as well as the ways in which AI models might themselves serve as serious theories of human cognition.

Bio: Eric is the Earl B. Dickerson Fellow & Instructor in Law at University of Chicago Law School. Eric's research investigates how humans and machines reason about law. Eric's research has been published in PNAS, Cognition, Journal of Experimental Psychology: General, Georgetown Law Journal, and other venues. Prior to coming to UChicago, Eric served as a research fellow at the institute for Law & AI. Eric holds a PhD in cognitive science from MIT and a JD from Harvard Law School.