Use ChatGPT to correct essays against sample solutions

Prompting
1

Hello,

I integrated a ChatGPT assistance into a webpage and everything is working as expected. Now I am challanged with a new feature request where a essay(text) should be compared and rated to a sample solution.

As the sample solutions and the essays are longer texts I doubt if it is the best to put everything into the prompt like:

Please rate the following essay: …
A sample solution for the essay is: …

I lokked into the API docs bit found nothing for tackling this problem directly. Only “Embeddnings” sounds as if they might be used for this:
https://platform.openai.com/docs/guides/embeddings

Any other ideas? Or is putting evrything into the prompt the way to go?
(But I guess also the costs will be influenced by longer propmpts)

1 Like
2

Off the top of mt head, all you can do is try and see if it gives you expected results.

There are not ways to trick the system into dealing with lots of data for no cost, everything you need accurate processing on needs to be included with the prompt, I don’t see a way that embedding would be useful for this, other than maybe using the similarity score as some kind of benchmark between the essay and the example solution, but you are limited to 8000 tokens in that regard.