@markmetz I still miss Usenet.
Plus side: we still have mailing lists. Not quite as freeform as Usenet, but more topically structured than the Fediverse.
There's probably room for Fediverse clients that can slice up posts topically so I could say "show me rec.arts.int-fiction" and have it show me appropriate things. But for Heaven's sake, not large language models.
I'd have to understand the math more, but would Bayesian classification be compatible with this? I've never read about it being used for an ever-changing set of queries, mostly just static "spam or not".
What would be missing from this of course would be the poster's intent in choosing a newsgroup.
I guess I still just miss Usenet.