Learning Heterogeneous Preferences

Shiwali Mohan, Matt Hong, Dule Shu +3

Learning from human feedback has become a central paradigm for training modern AI systems, where models of human utility are used as reward models in policy learning. Existing meth…