And acid comments are often wrong.

I still use both GPT and what I call “LibSource” (the name of the code I put in Unreal, Web or apps).

And you are correct, GPT has a strict policy, and that is why I never give up the in house development. IF you have a developer mind, you know that GPT does not know what it is talking about, so it cant trigger filters unless the specific words are said. So if my system uses the power of GPT to translate natural language to formats like JSON, I can simply use code language to say whatever I want to say, and use a library to simple create from that the kind of talk I Want it to talk without using GPT to trigger any guidelines violation. However the main reason I removed the direct link of GPT from the Mod wasnt the content worries or the effectivness of anything, it was the simple fact that external processing could hinder other aspects of the mod. So I changed to use a “pre rendered” set of code which I do ask GPT to make. (You can even see the trained version now in http://sensha.esav.app.br/, which is a version of my discord bot trained in a GPT ).
The discord bot itself shares with the website the processing system I use in the mod. And I am now in the process of expanding the use of a database driven dialogue system testing on the prototype of the mod I have locally here.
While I do like “Law and Order”, I dont really think the moderation style of GPT serves its own ends. For example, if you ask dalle to make a picture of a woman, it will in general make a big titty gal. But you cant ask it to “go easy on the boobs” because then you will be “violating the code of conduct”. While I think it is a wonderful AI, I also dont really think the human factor in there is pretty smart. But I know they have a community to think about, and that community is not easy to handle when it comes to ass and ■■■■■■■, so I just do my thing and use what I have the way I can.