Most of these models are lazily trained indiscriminately on as much content as they can get their hands on, and then guardrails added on the output to try to limit their inappropriateness according to the context.

See recently grok etc. where insufficient guardrails are put in place, presumably due to staffing, knowing Musk. But people wrongly assumed the model was somehow trained differently.

So yes, probably the worst porn, and this is the least-worst product for the prompt, according to some analysis, so released.