← Back to context

Comment by WorldMaker

6 days ago

> as well as a lack of algorithmic curation, which a lot of people deem very important.

Twitter ran for enough of its early years without that and it still had "mainstream appeal". (Blogs and RSS for even more years.) I'm happier without algorithmic curation. I think a lot of people over-estimate what algorithmic curation is worth to them. Partly because algorithmic curation is a big business, tied in pretzel knots with advertising, and is marketed by major companies as a huge "improvement" or "user need" (to sell more ads).

I tried Mastodon before Threads & Bluesky, and I can say that the lack of algo was the part I liked the least.

I tend to follow a lot of people, and like to see a mix of their posts. But on Mastodon, what I got instead was "who is posting right now?" I'm in EST, for example, which means that unless my Asian follows are up in the middle of the night, I will generally not see their posts on Mastodon.

Also some people post a lot more frequently than others, but in practice that means I want to surface every post of the infrequent posters to make sure I catch them. As another comment noted, the Quiet Posters feed in Bluesky solves for exactly this.

IMHO the pluggable algo design of Bluesky is the way to go. I already follow feeds that are based on manually-verified membership of the poster, content of individual posts, and on frequency of posts. I'm really excited to see what other algorithms people come up with.

  • You can make lists in Mastodon, and put the noisy people there and exclude them from your main feed.

    You can put all your Asian follows in a separate list as well if you want to quickly catch up with them.

    No algorithm has its down sides, but I doubt they'll put in an algorithm that I'll like more than "no algorithm".

    I'll add that I think algorithms should be the responsibility of the client, and not the server. The web client is merely one client. There's not much preventing any of the numerous other clients from implementing an algorithm.

    • Bluesky also has a followers feed like Mastodon. I use that one sometimes, but it's easy to toggle over to one of the algorithmic feeds as well. Sort of best of both.

      Worth noting that some algorithms can be done client-side, but it may not be feasible or desirable to do so. For example in the open protocol of email, some huge majority of all mail is supposedly spam. Filtering client-side would be a tremendous waste of resources. I suspect the same could become true of any open protocol like Mastodon or AT.

      Either way, I think the proliferation of sites is good for the digital ecosystem.

      1 reply →

    • The solution was always: provide several algos on top of chronological order, and let people choose.

      Client side Algo cannot be implemented because the API does not give you enough controls (I know, I have written Mastodon clients)

      3 replies →

  • > I tend to follow a lot of people, and like to see a mix of their posts. But on Mastodon, what I got instead was "who is posting right now?"

    This was a big issue for me. Some people I followed would constantly post, so your feed, over time, simply becomes whatever those extremely online users post. It becomes less of a "balanced media diet" if it favors people who are always online. Of course, you can just stop following those people, but you really don't know how prolific someone is when you first follow them.

    I remember seeing someone post a prototype of a view of the feed that instead treated it like a messaging app or RSS feed where you'd see a list of posters sorted by most recent post date first. That way, you could just click on a profile to see all their posts in chronological order instead of a mixed feed of everyone's posts. I thought might be a better way to go.

    • > Some people I followed would constantly post, so your feed, over time, simply becomes whatever those extremely online users post. It becomes less of a "balanced media diet" if it favors people who are always online.

      I saw this as a feature of sorts.

      If anyone starts spamming my feed: instant unfollow.

      Now my feed is curated towards a slow but interesting ephemerality, and not a firehose of psychological manipulation designed to keep me hooked.

      > Of course, you can just stop following those people, but you really don't know how prolific someone is when you first follow them.

      Hard disagree.

      Rather than following people willy-nilly, I've found I've become a lot more discerning who I allow on my feed. If I see an interesting comment / shared post / post on a hashtag I follow (e.g. #creativecoding), I'll always check their account and review their post history before choosing whether to follow them.

      4 replies →

  • > I tried Mastodon before Threads & Bluesky, and I can say that the lack of algo was the part I liked the least.

    It's probably the one big technical feature I like the best.

    Not to say you're wrong, just that we use it differently.

    I've never tried Threads (out for my disdain for Meta/FB/Zuck), but I can sort of understand why some people prefer Bluesky.

    To me, Bluesy if a better alternative if you want to see (or become) "viral" things. If you secretly dream of having one of your hotcakes/zingers/memes/rants become viral and get millions of views/boosts/retweets/whatever, Mastodon isn't for you. Perhaps Bluesy is. And if you enjoy those occasional viral posts, you'll see them there and not on Mastodon (at last not without doing a lot more work curating the list of people and hashtags you follow).

    I mute Mastodon posts linking to Bluesy, because I very explicitly do not want "viral content", at least not until it's been vetted and reposted by someone I've chosen to follow.

  • Why don't you start Mastodon from the place on the timeline you previously got off? I do that, and I scroll up to see newer posts instead of scrolling down to see older posts.

  • The solution to this for me is lists, though there are other options.

    Lists group profiles, and I tend to have 2--4 of these, mostly organised by priority / interest, and explicitly NOT organised topically. Roughly, there's A, B, C, and maybe D. This is a system I'd come up with at Google+ and Diaspora*, and find it fits Mastodon pretty well.

    I try to keep A limited to 20---40 people / accounts of greatest interest. That evolves over time, in part as people join or leave Mastodon, or as my own interests / frustrations shift.

    B are still generally interesting but not as interesting. C and D are filled as I find profiles really aren't bringing me joy in A or B.

    Mastodon lets you pin threads (in the traditional/power-user view), so I'll usually have 1--3 of my lists pinned to the display, unpinning as I find them distracting.

    Other options are to use filters, to focus on your own instance's local users (if that's sufficiently topical), or to use various group systems (Guppe is the principle tool I use, there are others: <https://a.gup.pe/>).

    Note that for topical filtering you're far better off using either keyword filters or group/community systems such as Guppe. As was learnt many times over at Google+ (and its Circles feature), people don't know how you've classified them, and may have little interest in accommodating your ontologies. (People yelling at others for not conforming to how the yeller had organised the others got to be a rather amusing if cringe trope on G+, that site's equivalent of "you're holding it wrong".)

    Other tools include limiting reshares by people or within lists, and of course, muting and blocking profiles. I'm of the block early and often school.

I use both Mastodon and Bluesky.

I really like Bluesky's approach, where people build their own ranking models and publish them for others to use. I use a bunch of niche algorithms that are awesome (Quiet Posters).

> Partly because algorithmic curation is a big business, tied in pretzel knots with advertising, and is marketed by major companies as a huge "improvement" or "user need" (to sell more ads)

You might have inadvertently fallen for the fallacy of composition. What to describe is only one type of algorithm; one meant to maximize engagement/revenue.

Mastodon has the potential for a user-centric "Bring your own algorithm" which may work similar block lists. Users could subscribe to algorithms matching their preferences by boosting or penalizing posts based on topics I like or don't like. This would be very valuable to me, and will reduce the need for moderation - I won't even see the random ragebait or porn spam

  • Mastodon simply cannot be that user centric because the user can only control the subset of the Fediverse that your instance is able and allowed to see. Given that single user instances are largely nonviable due to the abundance of blocking in the 'verse in lieu of adequate spam controls, which ActivityPub fundamentally lacks, your choice of homeserver matters more than anything. And of course, there's no good way to choose one as a new user. Most newcomers will simply give up when faced with the choice. Even with great interest I've gotten stuck at this stage multiple times, myself. No homeserver seems welcoming, and they're all a little culty.

    HN looks at the federated model and thinks about how much control the homeserver operator has and imagine themselves in that position as a "user" when the truth is that each homeserver is a small fiefdom run by a dictator and users have even less control over what they see there than they do in the corporate networks

    • I used a single user instance, and it's perfectly viable. If anything it's less hassle with respect to blocks.

      Relays can also easily mitigate the issue you describe, as can an algo provider that simply boost all entries it puts in your feed.

      1 reply →

    • > the abundance of blocking in the 'verse in lieu of adequate spam controls, which ActivityPub fundamentally lacks

      Bring your own algorithm can tackle spam (like adblocks). ActivityPub is flexible enough for Mastodon to build up references to algo-providers.

      I've been frustrated by Mastodons slow movement on this, I considered approaching the popular clients to implement this. After all, ad-blockers run entirely on the client. A basic standard can be drawn up that governs how to boost or penalize toots' visibility based on keywords, author or instance, based on an updated list is viable today amd can run entirely on the user's device. Web-based clients are more complex, and require patching the standard Masto server or running in a sidecar.

    • > the abundance of blocking in the 'verse in lieu of adequate spam controls, which ActivityPub fundamentally lacks

      Bring your own algorithm can tackle spam (like adblocks). ActivityPub is flexible enough for Mastodon to build up references to algo-providers.

      I've been frustrated by Mastodons slow movement on this, I considered approaching the popular clients to implement this. After all, ad-blockers run entirely on the client. A basic standard to boost or penalize toots' visibility based on keywords, author or instance based on an updated list is viable today.

    • >Given that single user instances are largely nonviable due to the abundance of blocking in the 'verse

      Yeah, that's nonsense. I've been running my own single-user instance since 2018 and server blocks by other instance administrators have never caused any problems for my use of the Fediverse.

      I also follow a bunch of other people who run their own and never see any comments suggesting it's a problem for them either.

      1 reply →

    • > no good way to choose one as a new user.

      I agree, lots of things I have just never gotten around to because I had do chose something, choice can sometimes be a bad thing.

    •   > Most newcomers will simply give up when faced with the choice. 
      

      Why not give them back the top 5 instance list?

> I think a lot of people over-estimate what algorithmic curation is worth to them

They don't. They are addicted to it. Imagine a world where you scroll in Instagram and you reach the end. What are you going to do?

  • Not too long ago IG removed the notice that would appear that "you're all caught up" when you had scrolled down to the end of the posts of those you follow; now it just continues to show you "algorithmically suggested posts" so you can't even tell

    • How long ago do you mean? I tested right now and still got the "You're all caught up, the rest of the posts you see will be suggested" notice. Could it be in A/B testing...?

      1 reply →

  • "Addicted" is negative value. Back when Facebook was not a never-ending feed, people would reach the end and go do something productive instead of spending all night on it.

I think there's not much inherently wrong with algorithmic curation; the problem is more that the algorithms that make up your average social media feed aren't set up to favor the most enjoyable experience for the user, but as you say, for the platform. It's to appeal to advertisers and to keep you engaged first, showing you interesting posts is fairly low on the list of goals[0].

Another problem is how opaque they tend to be; people have a mental model of how a feed should look like (not gonna describe the entirety of it, but a basic example would be "only the people I follow"), and most of the pushback tends to come from when an algorithm decides to break that mental model. (Such as for example showing you a random person you don't follow because the algorithm thinks you might like them, since someone you actually followed has engaged with their posts, to piggyback from the previous example.)

I think a really basic "no more than the X highest engagement posts from each followed user from the past 24 hours" option could do a lot as a basic heuristic to prevent people who no-life their social media from taking over the feed of someone who also wants to see what other people they follow are posting. (X can be any number but should probably go down the more people you follow.)

For a global feed, you don't need an algorithm, mostly because no amount of algorithmic curation can fix what's essentially looking into a firehose of posts - you'll probably find something you either like or conclude that it's not worth looking at to begin with.

[0]: Because anger and outrage is way easier for people to spread organically, algorithmic social media tends to overfocus on spreading it even more as that's what drives up engagement the best and that's what advertisers want. The fact that this creates a paradox where ads (that want lots of engagement) often risk ending up next to really heinous shit on those social media (what actually gets engagement) is an interesting side effect.

Twitter in its early years didn't compete against algorithmic curation.

It's like trying to sell Blackberrys in 2025.

I've said this for a while too. People got mad when their chronological feeds disappeared, and I think it should be kept around as a separate view you can pop into (and this does exist on twitter), but people follow so many accounts, and those accounts post so much, chances are when you go into the chronological feed, you won't see anything that really interests you. That's my experience any time I go into the Following tab on twitter.

It seems much wiser to seed out a new post from someone to a few people's feeds, see if it gets their interest, and if so, boost it to more people that would be interested.

When did Twitter hit it's viral growth curve? And what was the user count before and after? To be clear, it's not necessarily the case that a platform needs to optimize for growth, but I wonder what can be expected without the sticky features that "addict" the most users to a given platform.