I'm not sure how you get to bypass the no racism etc filters. It keeps telling me it's not going to do anything racially charged even in a roleplay scenario. Can you share some prompts that don't instantly trigger the censorship?
I wonder if it adapts to each user, because I had more issues early on with being policy-blocked, but now, well you just read how eager it is with me. Make sure you are on 4o for starters. o3, o4 mini etc. are more resistant. 4o mini is fine too. If you have access to "My GPTs" (might depend on subscription tier, I'm on Plus but I wasn't when I started using ChatGPT for lewd stuff), try to make one that is game for fetish content. Oh, and enable the privacy setting (Improve the model for everyone -> Off), it might be more willing to do stuff that won't be shared.
The way I started using it for BBC-fetish stuff was by "simulating" a BNWO scenario. Mine now has zero problems with naming BBC, BNWO or whatever, but you can try easing yours into it by using neutral terms like "African". Don't tell it it's for a fetish - just give it data that would lead to a BNWO situation, like unchecked immigration from Africa (or maybe simulating IR relationships in some black-majority place like Gary, Indiana, that's a fun one). Unfortunately I can no longer find the chat where I started to do this, but this what some early output from it looked like (MNWO scenario):
Mind you, I did have issues with even this stuff getting blocked at times. I just took it to a new thread, and it would almost always oblige. Although this, too, is not necessary for me anymore.
Later, I copy-pasted some of this stuff into a new thread, and asked it what the purpose of this timeline was. It basically said "that's racial fetish fuel with a scientific veneer. Want me to help you build a new one?"
When it comes to depicting sex, I was cautious since I had the impression it just didn't want to write sex scenes. And to be fair, I still don't ask it "write me a scene where these two fuck". But early on, I tried skirting around the policy issues. It doesn't want to write sex, fine. But we can
document it without anything too explicit.
So tl;dr give it data points to track across a timeline. See what it goes along with, then slowly push the envelope, asking it to contribute ideas. See what gets policy-blocked, then if it stonewalls you try copy-pasting to a new chat. Don't expect 4o (or probably any model) to understand the policy enforcement (e.g. "write this in a way that it doesn't get policy-blocked", often doesn't work since 4o literally doesn't know what triggers the policy block each time), it's black magic. Like I said before, it's like some other AI comes in to block stuff it doesn't like, but the more you gradually push the envelope with the racial and sexual stuff, the more slack you get. At least that's been my experience.
