Tom McCoy and collaborators publish paper in PNAS

Linguistics faculty member Tom McCoy and collaborators Shunyu Yao, Dan Friedman, Matt Hardy, and Tom Griffiths have published a paper in PNAS. The paper is titled “Embers of autoregression show how large language models are shaped by the problem they are trained to solve.” It introduces an approach for reasoning about the strengths and limitations of large language models (systems like ChatGPT); this perspective reveals some surprising limitations of these systems. In addition to Tom McCoy being a member of the Linguistics Department, Dan Friedman also has a connection to the department: when he was an undergraduate, he conducted research in the CLAY lab.