In some sense the completely no-guardrails version is literally impossible to use functionally! Even setting an LLM up as a chatbot with "no guardrails" requires a bunch of prompting