@PalePimp @JoeBravo77 I've learned recently that LLMs work the way they do because they're essentially built around "completion." Newer models have some logical faculties tacked on and can use websearch, but it's all essentially "the quick brown fox" → "jumps over the lazy dog." So when they act like chatbots, or you tell them to write slash-fics about Samwise raping Gollum, or ask for legal briefs, it's all the same: they're just spitting out things that sounds like they would follow your prompt. I mean, that's how it is in cases like the retard in the OP, who probably got an uncensored model to run locally because the models with guardrails "wouldn't do what I wanted."