Relevant memory is essentially a duplicate of the conversation

[yackimoff]

⚠️ Search for existing issues first ⚠️

• [X] I have searched the existing issues, and there is no existing issue for my problem

GPT-3 or GPT-4

• [X] I am using Auto-GPT with GPT-3 (GPT-3.5)

Steps to reproduce 🕹

No response

Current behavior 😯

Look at the code here: https://github.com/Significant-Gravitas/Auto-GPT/blob/75baa11e8196d9cdb26d26bf971ce3f98ebdaee5/autogpt/chat.py#L85

To generate context for the request, autogpt currently fetches 10 most relevant "memories" to previous 10 messages without any filtering. It doesn't make sense, since these previous 10 messages are in the database, so they're essentially most relevant to themselves.

Expected behavior 🤔

1. Messages that are still part of a conversation sent to the model should not at the same time be sent as "past events" to it.
2. "Past events" should be sorted chronologically to not confuse the model.

Your prompt 📝

Irrelevant.

[jeremylaratro]

From my experience, there is no contextual memory when using the OpenAI API ( only when you're using the web interface), so I'm assuming this is the workaround the devs decided to use for this issue, but I'm not a dev on this project so I don't know for sure

[yackimoff]

Yeah, it is a workaround of context length limit. It's actually great idea. It's just not implemented thoughtfully.

[fluxism]

I added some discussion of this issue at https://github.com/Significant-Gravitas/Auto-GPT/discussions/2847 before I saẇ this issue.

[github-actions[bot]]

This issue has automatically been marked as stale because it has not had any activity in the last 50 days. You can unstale it by commenting or removing the label. Otherwise, this issue will be closed in 10 days.

[github-actions[bot]]

This issue was closed automatically because it has been stale for 10 days with no activity.