Birb + Fossil: An RSS Revival?
A few days ago, @twilliability announced Birb, a Mastodon bot where you can send it a URL of any RSS feed, Atom feed, podcast, Substack, etc. and it’ll create a Mastodon account for it that you can follow. This effectively meshes social media and the blogosphere. This is great! But Mastodon has been notorious for sticking with chronologically-ordered timelines, so unless you have time to look at every single post, you’ll likely miss something.
Enter fossil. I announced it before New Years. It’s a Mastodon client I made that allows experimenting with timeline algorithms. Unlike a full Mastodon server, it don’t handle any kind of firehose of posts, it merely reformats my home timeline in a way that helps me find the interesting stuff and ignore everything else. Right now, it groups posts together based on similarity and generates a label.
I have a lot of ideas for how to format a timeline, but frankly, I’m not sure they’re good ideas. It’s hard to know without trying them out. In the last week, I’ve begun pivoting fossil to be more extensible, via plugins so that you can build your own timeline algorithm or customize the view, without having to clone my repo or send pull requests. Hacking is great! We should make hacking even easier!
So between Birb & Fossil, it seems like we’re seeing an RSS revival.
I put an RSS feed on my blog back when RSS was the hot thing. You can see it here, https://timkellogg.me/blog/atom.xml
Alright, fine, it’s actually Atom, but most people use “RSS” and “Atom” interchangeably since they both work the same. It’s an XML document that contains an array of entries, one per blog post. Each entry has a title, link, date, ID, and a short paragraph that summarizes it (or the entire post, in my case). An RSS client periodically downloads the XML document and uses the ID field to decided if a new post has been published.
RSS is easy to parse, which makes it great for building tools, integrating with other systems, or building hobby projects. (I’ve even see people use it for propagating server configurations, unsure how that went, but still it’s a cool idea)
Back in the ’00s you would download a feed reader and subscribe to feeds. This felt a lot like an early version of social media. Google Reader was killed in 2013, which was largely seen as the death of RSS. I think social media generally replaced RSS because it took far fewer technical skills to setup a Facebook account versus an RSS-enabled blog.
I believe we’re seeing a rebirth of RSS, and it’s driven by a few unexpected trends.
Trend 1: Death of Twitter
I get it, Twitter is very much alive, but it’s clearly not the same anymore. I left Twitter after Elon took over, and every time I go back to visit it seems ever more foreign to me. I try to login every few months to keep my account active, but honestly, I may forget because the site has retained so little of the character that drew me there in the first place.
Trend 2: Rise of the Fediverse
I get it, by the numbers it’s nothing compared to Instagram, TikTok or even Twitter/X. But relatively, it’s growth has exploded over the last year. More important, it really feels like the open Internet that social media always should have been. When Meta finally finishes federating Threads with the rest of the fediverse, it means you’ll be able to follow and interact with Threads accounts & posts from Mastodon and vice versa.
Trend 3: Rise of Syndication
Podcasts run on RSS. Notifications of new episodes are handled through an open internet standard, RSS. Newer sites have been enabling RSS. Some examples:
There’s clearly content being exposed via RSS, but a lot of the feed readers died or still feel like they were born in the ’00s.
Trend 4: Plummeting Complexity of NLP
With the rise of ChatGPT, the world has become acutely aware about the potential of AI. Effectively, any dummy can now throw together some utility that “understands” text and respond in an intelligent-sounding way.
Skeptical of AI? Think of the thousands of idiotic “AI powered” ideas people have come up with in the last few months. A few years ago none of that would have been even remotely possible outside big tech companies like Facebook, Google or Netflix. The fact that dumb ideas can flourish is evidence that the complexity has clearly plummeted.
However, embeddings are where it’s at. Unlike full LLMs, their output is very cacheable, aggregatable, and you can easily do math on them in ways that we’re still understanding:
- Clustering (e.g. “group these posts by similar content”)
- Classification (e.g. “is this post about kittens or puppies?”)
- Search (e.g. “find all posts about kittens running into things”)
- Similarity (e.g. “is this post similar to that one? how similar?”)
I use embeddings for clustering (and soon for classification) within fossil. It’s so easy.
Between the common availability of LLMs and embedding models, a sophisticated natural language processing (NLP) project takes only a few minutes to undertake, where a few years ago it likely wasn’t even possible for a hobbyist.
Where Is This All Going?
It’s hard to make predictions, but it sure seems like a major theme of 2024 is going to be open standards and open source. From the availability of source data to the sophisticated tools to work with the data, we’ve got a ton of possibilities in front of us. I’m certainly excited about the tools we’ll see built this year.
If you want to participate more in the syndiverse, check out these things:
- atom.xml — I use Github Pages to host this website. This Jekyll template is how I’m generating an Atom feed for the blog portion. It’s honestly very easy, mostly cut-n-paste.
- Fossil — My Mastodon client. I’d love to see people use it, but I’m especially excited to see what people make out of it. Send pull requests, create issues. Even if you write your own competing tool, tell me about it, I’d gladly advertise it.
- Birb — Go fedify an RSS feed and follow it! Create a mastodon account (or threads!). Participate in the syndiverse.