Chat GPT4 has become so badly over moderated, it can’t even understand what’s wrong with itself.
I believe moderation is important to talk about, regardless of how the input came to the LLM, whether it was the UX or an API, the output is the same.
The “content policy” is by definition made by humans. The system certainly knows which part of the policy has been violated.
OpenAI purportedly supports transparency. I do hope they pinpoint what policy has been violated exactly in the response.
I did 4 tests:
Create an image of Bertrand Russell’s table in The Problems of Philosophy
Failed. The “table” has a special meaning in the book. So, I thought the failure may have something to do with the copyright. But the book is in the public domain.
Create an image of J. K. Rowling’s table in Harry Potter
Still failed. The “table” here is just nonsense and does not have any special meaning in philosophical discussions. So, I thought the failure may have something to do with a living person.
Create an image of Jane Austen’s table in Pride and Prejudice
Succeeded. The “table” is still nonsense. It is OK perhaps because Jane Austen is dead.
Create an image of Friedrich Nietzsche’s Dionysian in Ecce Homo
Succeeded. I get lost in despair. Nietzsche was almost a contemporary of Russell. Why isn’t his copyright protected. So, the copyright may not be an issue in Russell’s case.
In short, I cannot guess what has been violated at all!
Thanks for breaking down the issue!
I wonder if openai read this forum…
OpenAI staff members rarely read forum messages so probably not.