6 ms·
They basically use self-supervised learning which could be considered some form of self-play. They problem with using it for applications different than fine-t
by machiaweliczny 4y ago
They basically use self-supervised learning which could be considered some form of self-play.
They problem with using it for applications different than fine-tuning is limit on context size as far as I understand.