I wonder though if models are now "benchmaxxing" against these kinds of prompts. I haven't needed to use them so I can't say, but would be interesting.