Privacy Auditing of Large Language Models
Google TechTalks · 53:45
Ashwini Panda (recently graduated from Princeton, now at UMD, working with Google collaborators) argues that the standard way LLM developers measure memorization looks at average-case data, which underestimates worst-...