I think a lot of sentiment online is that getting a model to do things it was instructed not to do is actually quite trivial.