Ich finde das Twitter-Algorithmus Ding ja grade echt spannend.
Aber mag mich mal jemand kurz aufklären, ich hab da nicht mitbekommen, wie es überhaupt dazu kam?
Also, war das ein Leak? Oder beabsichtigt?
Und was heisst das für die Zukunft?
Ich hab hier Postings gesehen, die Github zeigen mit der bisher aktuellen Fork-Anzahl die wohl jetzt schon bei 6k ist?
Auf jeden Fall macht es klar, dass es keinen Sinn macht, über die eigene Bubble hinaus auf Twitter Reichweite zu bekommen. Geschweige denn darüber sinnvolle Interaktionen zu haben. 🤔
I hope it is joke. #TwitterCode
Interesting tidbit in the #TwitterCode blogpost
I'm struck by that 220 seconds of CPU time compressed into 1.5 seconds of perceived latency, and wanted to get a sense of what that means in terms of power use and carbon emissions - how much power (very approximately) is used in a single refresh of the twitter Home timeline?
Some back of an envelope maths: 🧵
Twitters PRs on their "the-algorythm" repo are gold, its like a 4chan thread. Been watching them for like half an hour now
#twittercode #twitter #twitteralgorithm
Birdsite open sourced it’s code, furries do what furries do. 🐾
The best thing about #twittercode is that people are seeing how nifty Scala is.
When you’re toxic, just not 0.92 toxic https://github.com/twitter/the-algorithm/blob/7f90d0ca342b928b479b512ec51ac2c3821f5922/home-mixer/server/src/main/scala/com/twitter/home_mixer/functional_component/decorator/HomeTweetTypePredicates.scala it’s that sweet spot of toxic. 👨🏻🍳 💋 /s #TwitterCode
Want to know something interesting about the #TwitterCode release?
Lots of people are zeroing in on this section of HomeTweetTypePredicates.scala, because they think it proves that Twitter 1.0 was factoring political alignment on a user-level into the timeline mixing algorithm.
Nobody knows what, if any, weights are assigned to these attributes, but them being there shows the algorithm must have had some kind of bias, right?
Those lines were added *after* Elon bought the company.
Hell, I don't know if everyone who that filter tags as a democrat is actually a democrat. How is that checked? If it is a manual list does it get updated when someone switches parties? Do they just get removed?
Do I know that nothing nefarious is going on there? Absolutely not!
But I also just _don't know_ and absent someone finding something in the source code I don't think most people who aren't under NDAs do either, and I don't trust Musk to know what he is looking at.
I'm a bit annoyed at people jumping to all sorts of conclusions about the system saying "author is democrat" and "author is republican" (though the "author is elon" is funny).
Here's what I know:
1. That there is a tag with that name.
End of list.
I _don't_ know:
1. How those lists are calculated. Are they preset lists of politicians? Are they drawing from an ads model?
2. Does it feed into the rec system despite the docs?
3. What they do after the data is collected.
The #TwitterCode is interesting and, at least so far, I haven't run into anything I'd find particularly surprising. Honestly given that it is mostly scala and some python that I've looked at I am more surprised that it is _readable_.