Discovery here IS bad, and — I hate to say it — the solution to that is an (entirely optional) algorithm that surfaces potential follows and compelling content right out of the gate.
Algorithms aren’t inherently evil. Hell, “show posts in reverse chronological order” is in itself an algorithm.
What matters are 1) choice and 2) the intention with which the algorithm was designed.
#Mastodon https://mastodon.social/@mimsical/110865836818850808
I was able to bootstrap myself here relatively easily because I was both part of the deluge of Twitter users in November of 2022 as well as highly-motivated to make it work because I hated what Twitter was becoming and inherently grokked the benefit of federation.
That’s not going to be the common case.
If Mastodon is to expand far beyond low-hanging fruit like me it absolutely needs to meet users halfway.
@jeff totally agree. A lot of really good accounts came here in November 2022 but some got immediately put off and still others gave up around New Year. Many stayed of course, and the software is better now, but a lot was lost.
I understand not wanting to repeat the mistakes of corporate social media but I think Masto has been overly cautious with this, and as the default Fedi option should aim to implement all Birdsite's core features in a more consensual, better way.
@jeff would it not be enough to ask new people "type in a few hashtags you're interested in or select from this well-curated list of popular hashtags" and then auto-follow these hashtags for them?
Boom, their timeline gets filled out immediately with stuff they're interested in, no need to have any kind of "recommendation algorithm" involved.
Sidenote: this is less of a problem on smaller instances with lively and relevant local timelines, I feel.
> I dont get to see post from people I follow from other timezones
I get such posts boosted into my timeline all the time.
> there is no real way for me to find people posts or news local to me
That's a fair point.
> Making multiple algorithms available, making them open to analysis, and making them opt in/out sounds just fine to me.
I am worried about the power dynamics of this. Such tools tend towards "winner takes all" situations, I feel.
@SarraceniaWilds agreed on the large instances, I even wrote about it:
https://rys.io/en/168.html
The gatekeeping problem is wider, it's about who gets to decide the direction of the largest instance project on fedi, and why are certain features implemented and others are not.
But as I noted in a separate thread, recommendation algos can be implemented as bots, without having to wait for instance software to implement them.
@SarraceniaWilds sure.
But with recco algos you rely on chance that particular set of circumstances converge such that the given post gets picked up and promoted to you by the algo. And I am not even getting into the questions of who designed the algo, what are the edge cases, and so on.
You always necessarily rely on "chance", simply because there is too much stuff published on fedi, and physically not enough time to read it all.
@SarraceniaWilds : "there is no real way for me to find people posts or news local to me"
@rysiek : "That's a fair point."
The solution is: hashtags of geographical names.
@jeff
@xdej @SarraceniaWilds @rysiek I live in Pasadena, California.
If I want to find local posts for my area, do I search for #Pasadena?
If so, I’ll likely get news for Pasadena, Texas.
Do I search for #PasadenaCA? That might work, but some people might tag their posts with #PasadenaCalifornia and I’ll never see them.
And that’s just a local example. Will people searching for #Georgia get posts about the US state, or the country?
Hashtags are a hack.
@xdej @SarraceniaWilds @rysiek And that’s not even considering the fact that people are awful about remembering to add hashtags to their posts. And who can blame them? Adding all the hashtags needed to make stuff discoverable results in a lot of ugly, spammy-looking posts.
It’s not just finding things relating to geography, either. They are at best a stopgap for more robust metadata and full-text search functionalities.
@TedT A fellow local! 😃
On an unrelated note: I was a regular visitor to TheOneRing during an extended Tolkien kick I went in around the time the LOTR films came out (re-read the books, got into the lore, etc). Great work with that site — it was incredibly well-done for what it was.
@bauser first of all, there is a bunch of other software on fedi than just Mastodon.
Secondly, I don't think saying things like "[fedi] designers are so afraid of the world" is actually helpful as far as finding solutions to this is concerned.
It ignores — and I would even say: ridicules — the very real experience of people who got doxxed or otherwise harassed thanks to the design of other social networks, and decided to take precautions against that here.
@bauser I'm sure you'd agree that summarizing your position in this using terms like "ignorant of how certain tools can power harassment" would be a pretty low blow. Maybe let's exercise some civility and at least try to pretend we respect the (very real, often coming from personal experiences) concerns of others involved in this conversation.
@bauser I'm not convinced that calling people's lived experiences "stupid fear" is the winning strategy here, but you do you my friend. 🤷♀️
@rysiek @bauser @xdej @SarraceniaWilds Yeah, I’m not going to co-sign that framing.
I do think, though, that there’s a tendency on here to reflexively dismiss a lot of pretty basic functionality needs outright rather than constructively engage with attempts to make them work more safely.
Discovery doesn’t inherently preclude adding privacy controls, opt-in/opt-out, etc, but a lot of these people aren’t even interested in having that discussion.
@rysiek @bauser @xdej @SarraceniaWilds Also, it’s worth noting that a lot of features that are de rigueur on Mastodon now have also been used for harassment campaigns.
A hashtag, for example, served as the rallying point (and even very identity) of the ‘GamerGate’ harassment mob.
@bauser @xdej @SarraceniaWilds @rysiek It’s a general problem with hashtags, really. Though geography presents some unique challenges.
There are other examples. One I came across with a couple months ago was using hashtags to keep up with news out of Apple’s developer’s conference.
We had people using some combination #WWDC, #WWDC23, and #WWDC2023. To get everyone talking together you had to pollute your post with ALL of those tags. It was a mess.
@SarraceniaWilds an algorithm doesn't either, as I'm sure we agree.
I'm not going to die on this hill. But I do feel there is value in the direct human connection, so to speak, created by what I see in my timeline coming either directly from people I follow, or boosted by them.
I can always say: "this is exactly why I got this in my timeline, and this is the person responsible for it, for good or bad."
I find this kind of trivial transparency and humanity refreshing.
@SarraceniaWilds of course as long as I can disable or not use any kind of recommendation algo, I can have my trivial transparency and humanity of my timeline, and you can have your recommendations. Which is great.
But the point about power dynamics of such recommendation systems remains.
@SarraceniaWilds where is this coming from? Does any mention that "I like the humanity of X" implies that any and all alternatives are immediately "inhumane"?
I don't see how "I like the humanity of having a walk with a friend" would imply that walking alone or cycling is "inhumane." "Inhumane" is an extremely loaded term, and no, I am not implying recco algos are inevitably "inhumane."
Why read and interpret what I wrote above this particular (pretty hostile, it feels) way?
@SarraceniaWilds where did I say an algo "cannot be open-source"? 👀
Algorithmic transparency requires more than just openness of code, though. If the algo, for example, uses any kind of machine learning, apart from the code itself you need information on the datasets used in training, how they are labeled/categorized/etc, and so on.
That's why I wrote "trivial transparency." There is a way to have transparency with recco algos, it's just *more difficult*.
@rysiek @jeff Not really, because you're expecting newbies to know which hashtags will find good content on their specific instance/federation, which isn't a newbie-friendly approach (especially on young or otherwise under-federated instances).
I'm noticing a trend in these discussions, in that the response to any criticism of #Mastodon is effectively "New users just need to be social media experts."
@feelnotes Probably not.
But it is possible to make using discovery features entirely opt-in, as well as provide full transparency over what information is collected and how it is to be used.
Plus, it’s not as if a malicious actor couldn’t be collecting this sort of information now, if you use their server.
@jeff i dunno, it’s hard for me to imagine that kind of algorithmic feed existing without the incentives of data harvesting and advertising, but maybe that’s just a failure of imagination on my part.
i do remember early social media, before algorithmic feeds/ recommendations, and it was just fine.
@jeff When I'm in the mood for some algorithm I check out the # Explore feed.
When I need to search for something I saw before, I frown and go to ex-Twitter.
@djvdq That’s a curious one. It does work though. It feels a bit like the Reddit frontpage in 2011 or so, where it has somewhat interesting stuff, but it doesn’t feel like it’s personalised.
Think that might be what throws me off the modern web, where it’s different for everyone. I miss when there was 1 “front page” for everyone on a site
@jeff Yes! As long as it’s chosen/activated by the user, it would be a great addition to Mastodon. Some help text could even transparently explain in a couple of lines what the algo does.
It could take several form, for instance a different sorting mode for the home timeline, or a “For you” mode for the Explore tab, or its own timeline.
I had opened this issue on Github which might be partly relevant: https://github.com/mastodon/mastodon/issues/4451 (as well as this one about the Explore tab https://github.com/mastodon/mastodon/issues/23973 )
@jeff and, to add to that argument, the queries and logic around "in reverse chronological order" in Mastodon are ridiculously complex¹. Complex algorithms already decide what&who you see in your timeline.
¹spread through the codebase. But to fill your seemingly simple reverse chronological timeline, Masto must consider instance blocks&filters, personal blocks&filters, deletion, replies, visibility, delivery latency, hashtag subscriptions, deduplication of all that. And federation of all that.
bara mudita
Jeff C. 🇺🇦🇬🇱
Design by Adrian
Lê Fê is back
DuncanMSussexPol
Michał "rysiek" Woźniak · 🇺🇦
Veebs
Xdej
Blackthorn
Ted Tschopp
Michael Bauser
Jesse
John Coggeshall
Raccoon🇺🇸🏳️🌈
Kevin in Alaska
Jef Poskanzer
Lee 🌏
Stu Maschwitz
aces
Weyoun 6
EvilGinger- Queen of Naps
Dawid
Luke Sleeman
Allie
nicolas ⁂
Bèr Kessels 🐝 🚐 🏄 🌱