Thread
Just realizing that for a lot of you this is probably your first And only Twitter cluster, you haven't experienced the same dynamics happening in different circles due to algorithmic incentives. maybe your first online community at all
okay, i Will try to elaborate.
So each social medium has its own dynamics. TikTok has the "for you" page, an entirely algorithmic feed that doesn't care who your friends with. discord and reddt have niche channels u join. FB/insta u connect w IRL people
So each social medium has its own dynamics. TikTok has the "for you" page, an entirely algorithmic feed that doesn't care who your friends with. discord and reddt have niche channels u join. FB/insta u connect w IRL people
Twitter has a weird mix where you follow people and get to see their posts, but you also see a lot of other posts via retweet/sharing/algorithm.
This makes it more of a public forum than a deliberate nest of communities. tho unlike TikTok, you have some control over who you see
This makes it more of a public forum than a deliberate nest of communities. tho unlike TikTok, you have some control over who you see
The Home timeline uses some kind of secret spice algorithm similar to the For You page. It shows you viral tweets, things your mutuals have liked, trending topics. This algorithm changes sometimes. I think that it has changed again recently since the new CEO, and for the worse
People think you can avoid this by using Latest Tweets mode. but note that it doesn't say All tweets show up as they happen. If they did, there wouldn't be people like Visa inventing ways to get a view like "old Twitter"
The secret to understanding Latest Tweets mode is in the SimClusters paper released by Twitter a couple years ago.
This algorithm simplifies recommendation across the astronomical number of tweets and users by creating *artificial communities*
This algorithm simplifies recommendation across the astronomical number of tweets and users by creating *artificial communities*
these artificial communities — SimClusters — are found by first separating accounts into large accounts and small accounts. A rough guess from the numbers in the paper suggests the line is somewhere near 10k followers.
these are the v-nodes above
these are the v-nodes above
similarity is then calculated across the graph of which U-nodes follow which v-nodes.
If you follow similar big accounts to someone you've never interacted with, you will be clustered into the same group. these are c1 and c2 in the example diagram.
If you follow similar big accounts to someone you've never interacted with, you will be clustered into the same group. these are c1 and c2 in the example diagram.
My suspicion — I don't think there's a hard evidence for this in the paper, but it tracks with experience— is that the tweets you see in the latest tweets are based off those in your SimCluster. Not all of the people you follow.
This is why "old Twitter" looks different
This is why "old Twitter" looks different
You might think, wouldn't it be computationally easier to just have latest tweets show the people you follow? And yeah. except that the clustering calculation has already been done and cached (At what frequency I don't know)
plus Twitter has incentive to reduce variance
plus Twitter has incentive to reduce variance
they want you to use the Home timeline, not Latest Tweets.
So if you want to see all the people you follow, you might feel you have to use Home (probably won't work but the high variance might trick you into feeling like it does)
Too, Latest has more continuity w fewer people
So if you want to see all the people you follow, you might feel you have to use Home (probably won't work but the high variance might trick you into feeling like it does)
Too, Latest has more continuity w fewer people
You're going to feel like you're in a community, because of the limited amount of people whose tweets will show up.
also the reinforcement effects of talking to the same people in roughly real time, rather than experiencing "old" content
also the reinforcement effects of talking to the same people in roughly real time, rather than experiencing "old" content
and these computational dynamics prefigure the social dynamics between users.
This is why different corners have their own niche internet microcelebrities, and either hate or don't know of the microcelebrities of other niches.
different schools have different popular kids
This is why different corners have their own niche internet microcelebrities, and either hate or don't know of the microcelebrities of other niches.
different schools have different popular kids
as in any human grouping, popular kids create injokes and memes that help to reinforce "who gets it" and reproduce social norms and status hierarchies. This doesn't have to be deliberate, although it can be.
This essay mentions some of these dynamics naturalhazard.xyz/is_twitter_a_lemon_market_for_smartypants.html
This essay mentions some of these dynamics naturalhazard.xyz/is_twitter_a_lemon_market_for_smartypants.html
as groups become aware of their simclusters they develop names & mythologies.
"ingroup" was a funny one because it was self-referential and ironic. At first.
then people who just wanted to be *in the group* started using it in earnest. these things take on a life of their own
"ingroup" was a funny one because it was self-referential and ironic. At first.
then people who just wanted to be *in the group* started using it in earnest. these things take on a life of their own
sometimes these groups develop subgroups based on affinity. these are more like real communities.
cozies are a subgroup of ingroup that are highly insular, Not interacting much with outside Twitter. yet they are still subject to the same egregore dynamic
cozies are a subgroup of ingroup that are highly insular, Not interacting much with outside Twitter. yet they are still subject to the same egregore dynamic
sometimes a poster goes from U-node to v-node so rapidly that you can watch the simcluster form around them.
@tszzl ascended recently, and in short order had a crypto rug and a "cult" named after him. and his (admittedly clever) memes have now entered banal normie discourse
@tszzl ascended recently, and in short order had a crypto rug and a "cult" named after him. and his (admittedly clever) memes have now entered banal normie discourse
anyway the point of all this is not to tell you that your friends aren't your friends. it's totally possible to make real friends on Twitter. SimClusters are probably a good way of helping users find each other, for the most part.
but i find it frustrating to watch
but i find it frustrating to watch
once you have played into these dynamics enough, You can become trapped in a sort of sub-twitter. the sunken Twitter.
I have lurked and posted for years on this and other accounts, and seen it happen over and over. yet each time people think it's some new and personal phenomenon
I have lurked and posted for years on this and other accounts, and seen it happen over and over. yet each time people think it's some new and personal phenomenon
It doesn't "end".
People's egos clash, there's drama, infighting, some people leave because they're bored and some because they're ostracized, the scene calcifies around those who are left or morphs into something else around the ideas of newcomers.
People's egos clash, there's drama, infighting, some people leave because they're bored and some because they're ostracized, the scene calcifies around those who are left or morphs into something else around the ideas of newcomers.
This is why I'm going to vibecamp by the way. Not because I like all of the people who are going, (tho i like some of the people who are going and we're meeting up)
these things happen way faster and with more dimensionality in person. "Ingroup" will not be the same afterward
these things happen way faster and with more dimensionality in person. "Ingroup" will not be the same afterward
here's the link where you can download the SimClusters paper yourself btw. it's a cool piece of engineering, whatever its downstream effects might be
www.kdd.org/kdd2020/accepted-papers/view/simclusters-community-based-representations-for-heterogeneou...
www.kdd.org/kdd2020/accepted-papers/view/simclusters-community-based-representations-for-heterogeneou...
