The Twitter algorithm is now officially public.

Someone's been doing some code analysis on Twitter and it's been pretty informative. Screenshots_2023-03-31-20-09-55.png
Thread archive


I'm leaving these two out of the spoiler because of their importance:
Screenshots_2023-03-31-20-19-19.png
Source

Screenshots_2023-03-31-20-20-27.png
Source one | Source two
 
The pull requests are full of shitposts. This one imports an ethereum miner and then divides by zero, this one embeds a Japanese Spider-Man poster into the readme (addressing this issue raised by the same user about lack of superheroes in the readme), and this one adds an eggplant emoji as a potential predictor for nsfw material.
1680323261335.png
1680323310606.png 1680323382757.png
1680323510255.png
 
Last edited:
Besides the established userbase and the pre-existing relationship with advertisers, isn't the algorithm the entire business model? I'm not that tech literate but this seems like a bad business move, no?

And the back end. Twitter's back end runs on custom network hardware; even cisco gear can't route fast enough.

This is like if you were given a copy of the exact manufacturing diagram of a Toyota civic. Yes, if you had a CNC machine and metal press and blocks of metal you could build your own Civic. But you'd need to have that, and also good luck trying to scale it. And it doesn't include the code of the onboard computer, you'd need to develop that yourself.
 
View attachment 4937061
View attachment 4937057

This was found rather quickly.

Claimed to just be for metrics, and Elon says it's getting removed ASAP in a space.
Honestly inclined to believe this is for advertising metrics. If it was for censorship purposes I doubt it would be labelled that clearly. It would be some bullshit hate speech score or something to that effect. Part of the reason these people are able to pull off alot of the insane shit they do is because they've brainwashed themselves into believing they're operating on some sort of objective scale.
 
The pull requests are full of shitposts. This one imports an ethereum miner and then divides by zero, this one embeds a Japanese Spider-Man poster into the readme (addressing this issue raised by the same user about lack of superheroes in the readme), and this one adds an eggplant emoji as a potential predictor for nsfw material.
That eggplant one’s actually not a bad idea if peaches are already getting censored.
 
Found another thread that details how it ranks interactions for recommendations.

-Likes are worth 30x. Retweets 20x. Replies only 1X.

-images and videos are 2x

-putting in a link hurts your score unless you get good engagement (which will lead to more clickbait)

-if someone mutes, blocks, unfollows, or reports you for spam or abuse you get deboosted. (This means that mass flagging campaigns and block bots that blacklist certain individuals can lead to deboosting)

-paying 8 bucks a month gets you a 4x in-network and 2x out-of-network score

-anything labeled "misinformation" gets you a huge downboost, including specifically stuff about Ukraine.

-your profile is clustered in a group. It uses yur group to try and find users in groups similar to yours

-if your post is 'outside your group' then you get a massive pentaly

-misspelling words is also a massive penalty


-the three big data points it uses to determine what to do with your posts are:
engagement (likes, RTs, etc.),
mutes, unfollows reports
Who follows you
 
-putting in a link hurts your score unless you get good engagement (which will lead to more clickbait)

-if someone mutes, blocks, unfollows, or reports you for spam or abuse you get deboosted. (This means that mass flagging campaigns and block bots that blacklist certain individuals can lead to deboosting)

-anything labeled "misinformation" gets you a huge downboost, including specifically stuff about Ukraine.

-if your post is 'outside your group' then you get a massive pentaly

-misspelling words is also a massive penalty
twitterpaddlin.png

It's insane how much Twitter goes out of its way to actively hurt its userbase just by simply using it.
 
The news surrounding this source code is pretty interesting, but it also speaks volumes about their tech policies the more that they drum down this Twitter Blue for $8 and seeing random celebrities/entertainers get overtly sensitive when they have to follow the rules and regulations like other users that are not “tech savvy”.

This kind of stuff needs to be done often, but I can’t see their “like/retweet/favorite” system working much longer. I still remember the days where there was talk about programming a way to get dislikes onto the site, much like Facebook.
 
Im a tech retard, so is the code going open source means outsiders can tinker and adjust the code?
 
Open source means that anyone can view it in full, modify it, or reproduce it, but it doesn't mean that everyone now has control over Twitter. You can take Twitter's algorithm off GitHub and replace every line in there with whatever you want but it's not going to impact the actual Twitter website. Stuff is made open source for the purposes of transparency and for allowing people to freely use that code, not to let anyone adjust particular applications of that code.
 
Hmmm...
Knowing pathological neuroticism of lefties, it could have been their mass flagging and reporting of tweets that burried non-democrat users in the algorithm pit.
that and the blocking deboost would mean those massive automatic blocklists from left wing whackjobs would actually have an effect beyond making them look stupid.
 
Writing a simple Twitter clone takes a single intermediate programmer and a weekend at most. The value is not in the code.
Actual Twitter has no similarities with the clone twitters you can make for a tutorial project. Systems get exponentially more complex the more users you have. For example, twitter clones would have all their data in one postgres instance and run queries for everyone's feed. Actual twitter keeps active user's feed in redis caches and update them real time when someone in their network posts, only using the pull model on popular accounts with lots of followers. This is both more complicated and expensive than using SQL for everything, but it is the only way Twitter can work.
 
Back