Abstract:  There’s been a lot of talk about memorization in generative AI models. The word memorization means many different things colloquially. In this talk, we’ll discuss these different definitions of memorization, how they have been studied in the literature thus far, and why carefully delineating these definitions can be useful to our understanding of model generalization and privacy risks in model training and deployment. 

Katherine Lee, Google DeepMind

Submit your question for the speaker at Princeton AI Alignment and Safety Seminar (PASS)! We will moderate the questions and ask the speaker during the discussion period: PASS Question Submission