definitely have no doubt there are various ways to do RL/generation-discrimination/synthetic data/self-play-esque training on top of teacher-forcing that makes the models smarter, but especially…

Posters: repligate. repligate

3 likes recorded by the source corpus.