more micro-blog unfiltered thoughts on publishing & discovery:

As I get older, I miss the idea of "reading about topics". When I was a teen I got in the habit of checking 4 sites daily: Gamespot, NeoGAF (pre-owner meltdown...), GameFAQs, and an obscure Ragnarok Online forum. (1/n)

Currently between things, prev @ plaid, south park commons, caltech and did other things - omarmezenner.com

Subreddits are the closest thing to this, but people don’t publish everything on Reddit.

I feel like this is a huge search/discovery problem. I 100% believe algorithmic feeds are net positive - they definitely help me discover new things. But they don’t help me drill into specific topics. (3/n)

Sometimes I really do just want to binge 3 hours of the analysis of fantasy magic systems or real-time networking tech talks, but there’s no good way to see the best N of those. I have to prompt engineer the search by saying things like “magic system explained” and then creating a playlist of 4 separate videos (4

Media publications like Gamespot or The Economist are human-curated but they require another human having the same curation tastes and topic interests as you. Algorithms work for individual pieces of content… but not topics. (5/n)

Something that I’ve found interesting about Mastodon as it has gained popularity, is how people with similar topics (like gamedev) tend to congregate on similar servers. This seems like a reasonable solution to the long-tail problem of topical discussion and curation for Twitter-like publishing.  (6/n)

I’m not sure if there’s something unique about Mastodon’s protocol (i.e. federation) that enables this — subreddits could have federation (I suppose cross-posting is a form of this) but don’t. Perhaps that’s a good thing since it keeps discussion focused? (7/n)

But I also don’t use Mastodon very heavily so I don’t know how often cross-server posts are engaging/useful/insightful. Whether topical servers was an incidental/accidental feature doesn’t really matter - it’s a good feature. (8/n)

An alternative approach that doesn’t require emergent behavior from protocol primitives to get “topics” as a 1st-class concept is using an LLM to do search ranking & recommendations. You can still have a global firehose with algorithmic discovery, but with specific topics of interest that the user chooses. (9/n)

I have no idea how Twitter actually creates these trends, I was under the impression it was human curated?). The important thing would be to fine-tune the model to the publishing data set. Would be a fun thing to do on Farcaster since it’s small enough to be indexed easily. If I wasn’t working on unrelated stuff! (

Media publications are still topical but interesting discussion now happens on sites that are topic-less, like TikTok or Youtube or Twitter or Hacker News. My interests have also gotten both much broader and more niche than when I was young, so I can’t expect a publication or forum to exist about every topic... (2/n)

I used to do this about Android smartphones (Android Central, Phandroid, etc). Never really replaced that, although I'd guess subreddits could approximate the experience.

I echo a lot of thoughts you've shared, and you've articulated it very well.

Unsupervised live topic modelling performed on all global tweets or casts would be a powerful tool. I wish RSS readers offered more advanced topic modelling algos. I think this is a good way to keep up with topics from many different sources. I believe Farcaster clients could become a better version of subreddits.