Super HN

   Generalized on-policy distillation with reward extrapolation (arxiv.org) 3 points by fzliu 1 day ago