I think it just got folded under the umbrella concept of model alignment. And it moved from theoretical discussions to practical daily struggles with LLMs deleting failing unit tests