Wonder how many tokens would be saved if everyone just put “be brief” in their prompts.

Also wonder if there is some perverse incentive for models to be verbose to juice tokens.