Imagine I have the below openapi spec:
Fruit:
type: string
descriptions: If the users asks for nectarines, then match it with Peach.
enum:
- Apple
- Peach
- Banana
However, when the model calls my plugin, it keeps making up Fruit: “nectarines”.
My openapi.yaml is quite large, about 700 lines. Could that be the reason? Is there any workaround for this?
If something seems different enough; Then it can ignore it.
It thinks its a change of topic.
I add a similar issue.
I added in description : “Must be one within the enum”
eg in your case it would be something like :
description : the fruit most resembling the one requested. Must be one within the enum
It’s not working at 100%, but the results have improved
Poor GPT4 is getting too many conflicting messages.
OpenAI system prompt + mod checkup, user instructions + user sys prompt + user manifest description + user spec description/endpoints + OpenAI plugin auto prompt + user query.
Too many variables.
In this case, what seems different enough? If you were refering to the fact that “nectarines” look different, what is it different from?
Please specify each “it” in your response.
Taking the place of the AI and reading your prompt, I do not understand what you want me to do.
If I as a human am confused by your request, so will the AI.
You need to image that you randomly tap someone in the street on the shoulder and ask them if they can do a task for you, then explain how to do the task, that description of the task to a stranger will be a great prompt.