[mixcocam]

> I'm thinking of something like "algorithm as a service," which would be aligned with your interests and tuned for your personal goals.

RSS is just a protocol. You could make a reader now with any algorithm you want that displays feeds. In fact, I can’t imagine that no one is using the AI boom to say they will build a decentralized Twitter using rss plus ai for the algorithm

[jasode]

>RSS is just a protocol. You could make a reader now with any algorithm you want that displays feeds.

Your proposal to filter on the client side where the RSS reader runs can't do what the gp wants: algorithmic suggestions on the _server_ side.

The issue of an AI algorithm applied to client-side-RSS is it's limited to the closed set items of the particular feed(s) that RSS happened to download of whatever websites the user pre-defined in the white-list.

Example of inherent client-side limitation would be how Youtube works:

- a particular Youtube channel about power tools : can use RSS to get a feed of that channel. Then use further customized client filtering (local AI LLM) to ignore any videos that talk about politics instead of tools.

- the Youtube algorithm of suggested and related videos to discover unknown channels or topics : RSS can't subscribe to this so there's no ability to filter on the client side. Either Youtube itself would have to offer a "Suggested Videos as RSS feed" -- which it doesn't -- or -- a 3rd party SaaS website has to constantly scrape millions of Youtube videos and then offer it as a RSS feed. That's not realistic as Google would ban that 3rd-party scraper but let's pretend it was allowed... getting millions of XML records to filter it client-side to throw away 99% of it is not ideal. So you're still back to filtering it on the server side to make the RSS feed managable.

In the "explore-vs-exploit" framework, the "explore" phase is more efficiently accomplished with server-side algorithms. The "exploit" phase is where RSS can be used.

- "explore" : use https://youtube.com and its server-side algorithms to navigate billions of videos to find new topics and content creators. Then add interesting channel to RSS whitelist.

- "exploit" : use RSS to get updates of a particular channel

[EvanAnderson]

> Example of inherent client-side limitation would be how Youtube works: > ...

I thought about this problem a long time ago but never did anything substantive with it. I guess I'll articulate it here, off-the-cuff:

People used to post a "blogroll" (and sometimes an OPML file) to their personal blogs describing feeds they followed. That was one way to do decentralized recommendations, albeit manually since there was no well-known URL convention for publishing OPML files. If there was a well-known URL convention for publishing OPML files a client could build a recommendation graph. That would be neat but would only provide feed-level recommendation. Article-level recommendation would be cooler.

One of the various federated/decentralized/whatever-Bluesky-is "modern" re-implementations of Twitter/NNTP could be used to drive article-level recommendations. I could see my feed reader emitting machine-readable recommendation messages based on ratings I give while browsing articles. I would consume these recommendations from others, and then could have lots of fun weighting recommendations based on social graph, algorithmic summary of the article body, etc.

[jcgl]

GGP does express interest in Algorithm-as-a-Service (AaaS), but I don't see why AaaS or server-side anything would be required to have non-chronological feed algorithms. Client-side is perfectly suitable for the near-univeral case where feed servers don't overwhelm the client with spam (in which case you remove the offending server from your feed).

To your points about YouTube-style algorithmic discovery, I do agree that that would require the server to do things like you describe. So I think that there could be both client-side and server-side algorithms. In time, who knows? Maybe even some client-server protocol whereby the two could interact.

[emschwartz]

Scour lets you add feeds and topics that you’re interested in and then sorts posts by how similar they are to your interests.

It also works well for feeds that are too noisy to read through manually, like HN Newest.

https://scour.ing (I’m the developer)