Raw LLM Responses

Depends what you intend on using it for, and how much you care about it being identifiable as AI generated by a small portion of those reading it. If you're a news site and want decent quality articles that don't appear to be AI generated and are factually correct then yeah, you need a lot of oversight and fine tuning. But if you want to shape public opinion on a particular website, and have a ton of bot accounts ready, you can just feed it the text of the existing discussion and have some of the bots reply to it with an AI generated response pushing the viewpoint you want. Sure it'll occasionally mess up in really obvious ways, eg: "As an AI Language Model", and the really savvy users who think critically might be able to identify it as AI even when it doesn't mess up quite that badly, but the users you're targetting and trying to influence won't identify it for what it is as easily. Once you've got the bot accounts and the script for the website ready, tailoring it to support additional topics or completely different view points is as easy as changing a couple of words. You can use it for years with virtually no oversight needed. Just occasionally feed it new bot accounts