Are the content filters internal to the new model, or is there a second moderation layer? I thought that with the API you're straight querying the base model, but I might be mistaken.
I'm querying OpenAI API directly, and they are applying content filters. Previously (when working with davinci-003) the developers were supposed to query content filter API separately. Not anymore, with ChatGPT it's all baked in.
Which makes the ChatGPT version basically useless. Not because of the racism, I would probably filter that myself if it became an issue but because half the things you ask it to do are, “I’m sorry as an LLM I’m not able to do <thing it can very obviously do>”.
Well, people seem to be enjoying it. I wouldn't go as far as calling it useless.
Fair, "It's such a step down in experience that even for 10% of the cost and a simpler API it's not worth it."