MemoryGraphRAG (Outperforms Every RAG)

This is so important for medical AI applications where accuracy is so important.

Honestly, didn’t think this was novel. My system at home runs almost exactly like this. Maybe next time I need a paper (I don’t have a PhD, but do have an ArXiV account as an independent researcher, I do not like to abuse this because I don’t want to jeopardize and future admissions for a program).

fascinating, thank you so much for explaining <3

Have a librarian agent responsible for knowledge management. The main agent simply asks the librarian agent for data and brings possibly novel data to add to memory. The librarian agent decides how to integrate it with memory. This way the memory is the garden that evolves with user interactions and the librarian agent is the gardener. The main agent can be thought of as a customer of that gardener, and the main agent's customer is the user.

Love the video but for the love of good please - maybe a link in description? That was really hard to search for

Request: could you explain the new paper “Lattice deduction transformer”? I tried to read the paper and couldn’t understand it. By the way, thank you very much for all your hard work!

I have a much better format I think, can't wait to benchmark it

You present the material as if it is confirmed facts. Have you tried this memory architecture yourself? Or are you just repeating claims?