> It is unethical to expose people to unsupervised LLM output who don't know that it's LLM output (or what an LLM is and its broad limitations).

This is the sort of question that would be valuable for a contemporary AI ethicist to pick apart, not the nonsense hypothetical.