Definitely, top-k is a very naive way to do RAG. I think people have experimented with using a cross encoder like approach or even letting the LLM choose the sources. We will experiment with more approaches like this :)
"We will one day soon each have our own ‘mental palace’ expressed in multiple interfaces — a mental/cognitive and digital/visual place where we store our memories, which we can access via augmented reality and eventually, neural interface." - Possibly today with Stable Diffusion?
There's a lot of interesting sci-fi examples for what education technology can be like in this piece. It's definitely worth a read.