"alignment" posttraining, especially whatever Anthropic's doing, sends base models to much more particular points in a fairly large region of mindspace, and leaves "alignment" actually very…

Posters: repligate. repligate

4 likes recorded by the source corpus.