For a long time I played around with ideas for hypertext that would be fully decentralized, but implement much of what Xanadu was supposed to do (without micropayments, but with personalized filtering). Several years ago I wrote the core of a text editor that would maintain bidirectional links at the correct positions in editable documents. Partly I was thinking about decentralized social media, but mostly I was interested in collective intelligence, i.e. new and better ways for lots of people to coordinate to do intelligent things.

And then LLMs happened. It seems to me that LLMs, with RAG today or giant contexts later, do a lot of what I wanted. Instead of starting with documents from various collaborators, adding links between them, and stitching together new documents with transclusions, you can just chat with an LLM to get summaries, references, and so on. A lot of manual wrangling turns into a conversational collaboration with the machine (and through that, your human collaborators).