Check out our new paper taking the guesswork out of value alignment! We show how to best represent values, and predict cross-value generalization, and build the first taxonomy of LLM values grounded in generalization dynamics.
LLM developers train their models to follow prosocial values, but post-training can influence model behavior in unexpected ways. In “Predicting Alignment Generalization with Value Representations”, we ask whether downstream effects of value alignment can be predicted in advance.