Damn, the #ChatGPT 4 usage cap also applies to testing and implementing #CustomGPTs. And I swear there's some usage multiplier on top, like there's internal requests (like RAG or Actions) happening that count against the quota.
This does not feel like a very serious development environment!
This Custom GPT Tester is _not_ running fresh sessions when you change the prompt!
I actually would like if I could continue an old transcript with an edited prompt. But it's neither here nor there... everytime I edit at all I seem to get a new session, only GPT also seems to know something about the old prompt.
(This might be superstition... but very specific formatting instructions seem to be preserved)
The document retrieval is entirely opaque, including to the person developing the Custom GPT. This is absurd! I have no way to know why the GPT is responding the way it does; is it reading my documents but ignoring them, is it reading them but interpreting them differently than I think? Is it even reading the right documents? If I change the prompt will it read them differently? I'd would hardly be able to tell if it did!
This is not a system for serious prompt developers.
Ian Bicking
stilescrisis