Will save you some tokens: „write code like Linus Torvalds” - model should have all his swearing included in training data.