I suppose that begs the question, what if we trained an LLM only on examples of formal languages?