I don’t think LLMs work towards anything. It just picks from a list of likely “next tokens” at random.