Unanswered

Is Accessing GPT Chat History Possible?

(0) Share

Report

Posted on by 4-more

Hello everyone,

Up until now, I have successfully created models that address problems by breaking a task into smaller segments, using the output from the previous step for each new one. However, due to token limitations, I am unable to include the output from every prior step.

Is there a method for accessing complete chat interactions, enabling the model to "learn" from earlier prompts and responses?

One such task would involve summarizing extensive documents exceeding 100 pages, which may contain cross-references (thus, simply dividing them wouldn't be very effective).

Thank you!

Categories:

AI Builder

All responses (3)

Answers (0)

Sort by

CedrickB Microsoft Employee on at

Like
a
(0)

Report
Copy link

Link copied!

We are working on this kind of feature at the moment, we are targeting to release a first version this year.

In the meantime, the alternative is to add meaningful fragments of the document as context in the prompt.
4-more 41 on at

Like
a
(0)

Report
Copy link

Link copied!

Yes, absolutely. However, can fragmenting be automated in a meaningful way? It makes no sense if a document is split in the middle of a chapter.
CedrickB Microsoft Employee on at

Like
a
(0)

Report
Copy link

Link copied!

It is not an easy problem to solve and again we are working on a builtin solutions at the moment.

What you could do is

1. Index your documents by spliting in into chunks which would fit inside a prompt, saving all of them to a table.

Using Dataverse search to retrieve the most applicable chunk depending on the request.

Pass the most applicable chunk to the prompt together with the initial request to get a final answer.

2. Wait until we provide this capability 🙂