Comment by XorNot

I think this hits on a genre which hasn't quite happened yet: deguard-railed, presumably pirate model LLM outputs.

Like the early days of ChatGPT when you could fairly easily jailbreak it to ease GPT4 output, I think we're likely to see a whole community of getting whichever of these models leaks to explore the weirder areas of their weights as a kind of internet performance art thing.

I suspect your system being suitably grey market or possibly stolen would be a pre-requisite of such a scene.