Reward Model Ensembles Help Mitigate Overoptimization.

Thomas Coste, Usman Anwar, Robert Kirk, David Krueger

Browse the full ICLR paper archive.