[Sketchiest Guy in the World Voice] Hey kid, wanna see the X algorithm? It’s right over here.
No really, Elon Musk appears to be partly making good on his promise about a week ago to open up the X recommendations algorithm for public perusal and input, theoretically making the main feed on his social media platform open source. He previously promised he would do this back in 2022, and sort of did by publishing one snapshot of the code shortly afterward, but that repository wasn’t kept sufficiently up to date to make the X platform qualify as most people’s idea of an open source product.
This release, then, is a promising step in the direction of X truly being an open source product. The next step would be to update this code repository in four weeks, as Musk promised he would do.
Even then, this release wouldn’t mean the open sourcing of X can be marked “promise kept.” In his January 10 X post promising this release, Musk said he would release “all code used to determine what organic and advertising posts are recommended to users.” From where I’m sitting, that has still not even come close to happening.
That’s because on November 26 of last year, the accounts for Musk and Grok posted that Grok is used to sort the posts on everyone’s Following feed by default, although it can be toggled from “popular” to “recent” to make it chronological. That algorithm appears to be missing. The Following and For You feeds on X also have ads, which Musk has indicated are served via an algorithm that he said he would make public. So by my count there should be at least two more releases, possibly more.
Gizmodo reached out to X for information about whether or not the advertising and Following feed code has already been released, or if it will be released at some point in the future. We will update if we hear back.
But anyway, here we are with a fresh dump of code. The first thing you should know is that it “sucks,” according to Musk.
Yes, the algorithm sucks
— Elon Musk (@elonmusk) January 20, 2026
Earlier on the same day Musk said the algorithm sucked, X head of product Nikita Bier seemed to indicate that he was proud of it, noting that in the six months from July of 2025 to this month, daily engagement time from new users has gone from less than 20 minutes to somewhere in the mid-30s. Who’s right? Is it better than ever, or does it suck?
The problem may be that Musk just can’t seem to clean out all the stubborn wokeness residue stuffed into X back when it was called Twitter. His tweet saying it sucked was a response to former video game executive Mark Kern complaining that the algorithm weights posts less heavily if they come from accounts that have been blocked a lot. Kern says he suspects that this biases the algorithm against posts from right-wing accounts like his own. That’s plausible I suppose, though it almost certainly biases the algorithm against accounts that post a lot of harassment and abuse, so make of that what you will.
Judging from what’s in the plain text readme documents in the Github dump, this latest X algorithm is what you probably expect if you use X: an update to the TikTok method of hooking users. My impression of what’s described is that, unsurprisingly, it prioritizes engagement, attempting to figure out which posts will make the user stop scrolling. It pulls from accounts you follow, but also accounts deemed to be similar to those you follow. It’s appealing to your id, not your superego. No matter what you think you’re there to see, it wants to show you whatever will make you keep staring at it.
In addition to sucking, Elon Musk also says it’s “dumb.” Replying to a complaint from blogger Robert Scoble complaining that the algorithm favors posters who hijack news events, Musk says the algorithm will improve every month—seemingly referring to the four-week expected cadence for GitHub code dumps.
We are trying to to make the algorithm less dumb.
It will improve every month.
— Elon Musk (@elonmusk) January 20, 2026
And who knows, maybe users with amazing ideas will dig not just into the readme sections, but right into the code, find the real problems, and pass along suggestions to Musk, and the algorithm will get more satisfying and profitable over time. Alternatively, maybe the needs of a company that wants to hook users in order to get them to watch ads and generate revenue for itself, and the desires of human beings who want to feel well informed and happy are two totally irreconcilable concepts, and making a recommendation algorithm open source in order to try and serve both those types of need is utterly futile. I guess we’ll see which of these maybes is actually true.
Read the full article here
