It already is depending on the model and training. I might get a mild kick out of someone calling out a 7B, or smaller like most bots are built with. Tell a 70B or 8×7B “in the style of doctorate-level dialog and vocabulary” and watch what happens.

Models are mirrors of the dataset culture and the person writing the prompt. A junk bot is like a drunk human; the good ones are indistinguishable.