Explore the complex relationship between memorization and generalization in deep learning models through this insightful conference talk. Delve into the ubiquitous nature of memorization, examining evidence from data diets, example difficulty, and pruning techniques. Investigate whether memorization is essential for generalization and consider theoretical work suggesting its elimination may not be feasible. Discover strategies to mitigate unwanted memorization, including improved data curation and efficient unlearning mechanisms. Examine the potential of pruning techniques to selectively remove memorized examples and their impact on factual recall versus in-context learning. Gain valuable insights into the dynamics of deep learning models and their implications for AI development.