Astra and Fable fall short at model post-training, one user argues

A user argues that Astra and Fable aren’t yet very good at post-training models. Despite finding both impressive, the author says they give an “illusion of making progress” while missing too many important details when given autonomy.