You don’t even have to “break” llm into anything. It continues your prompts, making sentences as close to something people will mistake for language as possible. If you give it paranoid request, it will continue with the same language.
The only thing that training gave it is the ability to create sequences of words that resemble sentences.