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

[ryan_greenblatt]

(Also, to be clear, thanks for the comment. I strong upvoted it.)