ryan_greenblatt comments on An issue with training schemers with supervised fine-tuning

[ryan_greenblatt]

(I’ll edit the post at some point to highlight this discussion and clarify this.)