Prompts don't matter when you've heavily trained the model for verbosity (because that's what gets you the best benchmark scores).