That's effectively what RLHF is; a means for LLMs to self train on their own output exclusively by using a small human curated dataset as guidance as to what a

That's effectively what RLHF is; a means for LLMs to self train on their own output exclusively by using a small human curated dataset as guidance as to what a "good" and "bad" output is.