A general plea to the OpenAI developers. Could you please take as much care to the error messages ChatGPT gets when accessing plugins or GPT actions as you would when displaying an error message to the user? I’ve had several occasions
where I was getting quite desperate trying to diagnose an unspecific error that was returned to the GPT without any description of the problem, and had to resort to wild trying out things until the error disappeared.
A really good error message should describe the problem on a human level, but often it’s also enough just give the technical exception message including all additional attributes and wrapped exceptions to the GPT as ChatGPT or the
programmer configuring the GPT can often guess the problem from it. At least that should be done when the GPT actions are called during preview mode during the GPT configuration.
An even better error message could include instructions for steps that the GPT itself or the user can take to remedy the problem, and whether just retrying the same thing makes sense.
Specifically, I had trouble with the authentication. When I’m configuring Authentication Type “API Key” with “Custom”
but do not set the custom header name, that happens. There have been other cases, too, but I cannot reproduce them right
now. But that’s just one example, my plea is more general than that.
Thanks to OpenAI for all the marvellous stuff they provide, and thank you all for the interesting discussions here!
I wish you a nice new year!