Does Plugin Response Type Change Behavior?

I’m testing a ChatGPT plugin that integrates with a data warehousing library for access to proprietary reporting. The output is horribly slow on ChatGPT’s end though to the point of being kinda useless. It gets data back from the API and then outputs it like it’s slowly generating the text rather than just directly displaying the data.

I had wondered if there is anything special that needs to be done in the OpenAPI spec return type for the endpoint that would make ChatGPT a bit smarter in how it handles the output. Right now I’m returning JSON rows but the response type of the endpoint is not specified. It’s a reporting endpoint so the response columns vary with the request. Maybe that’s the issue?

1 Like

I use guiding prompts as part of my responses. Include them after and tell them how to format or adjust the NLP of the data provided.

And with that you are able to get it to just spit out what your plugin returns more quickly (a table of data in my case), or does it still take it’s time as if you are watching someone slowly type the output?

If you want something fast use an image, and if you something descriptive, use text and mark down for formatting

2 Likes