Studying Large Language Model Generalization with Influence Functions

Roger Grosse, Juhan Bae, Cem Anil +14

When trying to gain better visibility into a machine learning model in order to understand and mitigate the associated risks, a potentially valuable source of evidence is: which tr…