Because a bunch of markdown files is just RAG, and RAG is unintelligent, so the results are not great. If you want a smarter AI, it needs to have not-dumb memory. That's why this article (and the summary I posted) covers multiple kinds of memory, multiple ways of managing different memories, multiple ways of finding memories, a way to pick the best memory, and a way to manage memories long-term (and among multiple users). Now the memory isn't dumb, so the results are better. (And the article shows you why it's better)

tl;dr https://www.elastic.co/search-labs/blog/agent-memory-elastic...

> If you want a smarter AI, it needs to have not-dumb memory.

Who says? According to what metric? How would you prove that assertion?

> Now the memory isn't dumb, so the results are better. (And the article shows you why it's better)

But, it doesn't. It explains what they built, and how it behaves. It does not show why it's better than any other alternative for making models "smarter", somehow.

[flagged]