Says Twitter's algorithm should be open source.
"Twitter algorithm should be open source"
He agreed you have to see what is in frame. He said the X feed can be a little dark at times, and he switches to a domain-specific feed such as AI to actually learn what is going on.
"Twitter algorithm should be open source"
"open source the algorithm"
"that's why i'm an advocate of having the algorithm be open source"
"If Twitter can accurately recommend tweets to you that you don't explicitly follow, it can by the same token, also show ads that are relevant to you."
"an ad that is highly relevant is content. An ad that is irrelevant is spam. So we want ads that are as relevant to people's needs as possible."
"Twitter's advertising is actually for the most part irrelevance optimizing."
"I spent two hours on TikTok, but I regret those two hours."
"the thing that actually matters is like true user minutes, not like fake user minutes."
"since the acquisition, I was like, no, guys, we need to maximize relevance."
"So I think the thing we'd want to optimize for is unregretted true human user minutes."
"The giant block lists are problematic. They mess up the recommendation system & create a DDoS vector."
"Long day at Twitter HQ with eng team Two significant problems mostly addressed: 1. Fanout service for Following feed was getting overloaded when I tweeted, resulting in up to 95% of my tweets not getting delivered at all. Following is now pulling from search (aka Earlybird). When Fanout crashed, it would also destroy anyone else’s tweets in queue. 2. Recommendation algorithm was using absolute block count, rather than percentile block count, causing accounts with many followers to be dumped, even if blocks were only 0.1% of followers. Also, it’s trivial to bot spam accounts with blocks."
"Fixing the recommendation algorithm is our current top priority. Twitter engineering has been working super hard on this. Proud of the team."
"Prepare to be disappointed at first when our algorithm is made open source next week, but it will improve rapidly!"
"Our “algorithm” is overly complex & not fully understood internally. People will discover many silly things , but we’ll patch issues as soon as they’re found! We’re developing a simplified approach to serve more compelling tweets, but it’s still a work in progress. That’ll also be open source. Providing code transparency will be incredibly embarrassing at first, but it should lead to rapid improvement in recommendation quality. Most importantly, we hope to earn your trust."
"Twitter will open source all code used to recommend tweets on March 31st"
"It will be made open source, but we’re finding so many bugs that those should be fixed first"
"Twitter will be updating its recommendation algorithm based on user suggestions every 24 to 48 hours"
"Most of the recommendation algorithm will be made open source today. The rest will follow. Acid test is that independent third parties should be able to determine, with reasonable accuracy, what will probably be shown to users. No doubt, many embarrassing issues will be discovered, but we will fix them fast!"
"This is most of the recommendation code. In the coming weeks, we will open source literally everything that contributes to showing a tweet, such that the output can be at least roughly approximated."
"Nice work by the recommendations algorithm team! As always, changes will be open source. So far, over 1M lines of code have been uploaded to our GitHub repository."
"The commitment to open source transparency and accepting a wide range of viewpoints remains unchanged"
"The algorithm is trying to maximize unregretted user time on X, which is a good goal."
"A major update to the recommendation algorithm is rolling out over the next few days. This will help surface smaller accounts and posts outside of your friend-follows network. As always, this will be made open source and undergo continuous improvement."
"my prediction actually is that any given social network that does not move in this direction will be swarmed with bots and will simply be overwhelmed."
"biasing things to where the subscribers effectively or premium users content is elevated above that of bot amplified content is I think a very big deal."
"you have to do things that dramatically increase the cost of noise in the system"
"you have to, I think, bias the network to have basically raise the cost of manipulation and raise the cost of bots and trolls."
"A change is coming to our recommendation algorithm that will ensure that all your followers see your pinned posts. This only applies to one pinned post every ~48 hours to prevent gaming of the system."
"Actually, I made the algorithm open source and neutral to all parties, but of course that *is* “election interference” by her standards 😂"
"And it will get a *lot* better as we move to a fully AI-based post recommendation system, order the replies (and remove spam/scam) with AI and enable full semantic search with AI"
"That was hilarious 😂 Unlike the other platforms, our algorithm is open source! And if you interact and argue with “right wing” accounts, which Cuban does a lot, the algorithm will assume that’s what you *want* to do and give you more of it lmao."
"Yeah, the algorithm is open source and just tries to show people what they’re most likely to find engaging"
"The new recommendation algorithm should be noticeably better. Please reply below if not."
"A much improved Grok-powered algorithm is coming. Should help a lot."
"The 𝕏 recommendation algorithm is being replaced with a lightweight version of @Grok, so will soon be dramatically better! You should notice some improvement already."
"We are working on a Grok-powered algorithm that will be great"
"If you’re interested in applying advanced AI to the 𝕏 content recommendation system, join us! Getting this right is fundamental to improving the collective intelligence of humanity."
"The biggest change that people should notice as we shift the recommendation algorithm to use more of @Grok AI is that banger posts from small accounts should start showing up in your feed"
"Reply sorting will get much better."
"Using @Grok to understand the intrinsic value of content so that bangers from small/new accounts get shown by the algorithm is a major focus"
"Am trying out the new 𝕏 algorithm powered by Grok 4 Mini. It’s definitely better. Will require ~20k GPUs to roll out to all users and there is higher latency, but the juice is worth the squeeze."
"This is just a snapshot of the current situation. All the heuristic code will get scrapped in favor of a Grok AI recommendation system."
"The algorithm will be purely AI by November, with significant progress along the way. We will open source the algorithm every two weeks or so. By November or certainly December, you will be able to adjust your feed dynamically just by asking Grok."
"The 𝕏 recommendation system is evolving very rapidly. We are aiming for deletion of all heuristics within 4 to 6 weeks. Grok will literally read every post and watch every video (100M+ per day) to match users with content they’re most likely to find interesting. This should address the new user or small account problem, where you post something great, but nobody sees it. We will also be adding the ability for you to adjust your feed temporarily or permanently just by asking Grok."
"Grok needs enough information in the post to assessment which users will most likely find the post interesting, just like a human would need"
"This is utter crap. The AI algorithm derived from Grok and developed by xAI (not Nikita) simply tries to determine whether a post will be interesting to users. Posting a link with almost no description will get weak distribution, but posting a link with an interesting description/image will get distribution."
"To get the most reach on 𝕏 with Grok, just imagine that you’re texting a smart human who you’ve never met. If someone were to text you a link with nothing else to go on, you’re probably not going to think “wow, I should immediately forward this to everyone I know!” 🤦♂️"
"You are completely missing the point. Grok will recommend content to users based on what it thinks that user wants to see, not what users on average want to see."
"By next month, Grok will literally look at and understand all ~100 million 𝕏 posts per day (including images and video), no matter how small the account, and recommend content to users based on the intrinsic quality of the content itself. This is only possible with advanced AI using a lot of GPUs and will massively improve the quality of everyone’s feed."
"Nikita deserves a lot of credit for fixing many things about 𝕏, but is not working on the algorithm. Blame for me problems with the algorithm – it still has a long way to go to be great. Transitioning from an inscrutable hairball of heuristics, which was the old Twitter algorithm, to a pure Grok-based AI recommendation system is being done by a team mostly from the @xAI side that reports directly to me. Same with making advertising relevant to users, which basically didn’t exist in legacy Twitter. The engineering teams building 𝕏 Chat, which is encrypted and has audio/video calling and file transfer, and 𝕏 Money (launching to the public soon) also report to me, although I’ve recently asked Nikita to get involved as he finishes up other work. If anyone out there wants to help make these products insanely awesome, please join us!"
"You should notice rapid improvements in the quality of your timeline almost every week, but there will be two steps forward and one step back at times. We are also switching advertising recommendations to use the same Grok/AI system as organic posts, so quality of ads will get radically better."
"Update your 𝕏 app and look at your Following timeline. Posts of people you follow are now ranked by @Grok! You can still access unfiltered chronological if you want."
"I think probably in December we'll finally have a half decent recommendation algorithm"
"We’re adding another feature to posting which is simply “Enhance” with Grok, which will suggest ways to make your posts smarter and more detailed and offer images and video illustrations"
"The 𝕏 algorithm is built by an @xAI engineering team, not Nikita twiddling knobs"
"The @xAI team is working on providing For You tabs that are specific to topics. For example, a “For You AI” that is focused only on artificial intelligence with no political rage bait. This would be like automatically generated follow lists with content ranked by quality."
"There are another 30 or so xAI engineers working on the pure AI-based recommendation system"
"I do think there's a there's too much political rage bait on X"