The reason I'm working on this is precisely that LLMs are fantastic bullshit generators. (This is not criticism, LLMs are useful, just not in the way the Show HN/Product Hunt headlines describe it.)
One of the problems I've had with content generation in the project is that GPT-4 tends to do its job a bit too well.
GPT-3 would produce content that sounded much more funny because of how crappy and simplistic the resulting "poetry" was.
Then, with GPT-4 some of my editors started writing poetry in (relatively correct) Latin. It's a bit like comparing someone in a renaissance fair, or a LARP to reading a quote from a medieval history book, or a manuscript.
You can see a draft of the project here: https://bard-ruddy.vercel.app, it's quite old and different from the final "product", but the general idea is the same. I'll push it to HN on Wednesday.