💯
prompting for feeds is going to be game changing.
View quoted note →
Login to reply
Replies (14)
Wouldn’t you need some crazy indexing to enable this?
Indexing could be done on the client and sorted on there?
Massive data kill unless you use some centralized indexer
Wouldn’t need more data than you currently download? (Assuming this is about sorting people you follow rather than suggestions for content you din’t follow).
Oh I’m thinking for content discovery - you’d need to pull all events from multiple relays and probably the reaction counts - that would kill data I think. And battery
Suppose you could pull only new events and already have a local cache of seen stuff
Even just to “discover” content in your own existing feed (like stuff that happened whilst you were asleep but missed) would be cool. For content suggestion for people you don’t even follow would need more consideration.
I would recommend relays to dump to parquet files on cloudflare R2 with 0 egress charges and clients use DuckDB to query.
Doesn’t this introduce centralization that we seek to avoid in the first place?
Each relay will have their own dump. And we can query all of them together, just like how we do websocket requests
Yeah but Cloudflare can just decide what to take down
Makes sense.
But what about parquet on ipfs or arweave? I don't think DuckDB integrates with them though! 🤔
Indexes plus an LLM that was built on realtime data like Grok is.
not just feeds, also ad hoc super specialized nostr apps that can be shared easily