To the degree that people are seeing improvements in their feed, it is not due to the actions of specific individuals changing heuristics, but rather increasing use of Grok and other AI tools.
We will post the updated algorithm, including model weights, later this week. You will see that a lot of random vestigial rules are now gone.
Going to full AI recommendations, where all 100M+ posts per day are evaluated by Grok and those most likely to interest you are recommended happens next month. This will profoundly improve the quality of your feed.
Shortly after that, we will build the ability for you to dynamically adjust your recommended content, eg “show me less politics”.
Today, September 1st marks Disaster Prevention Day in Japan—a solemn reminder of the 1923 Great Kantō Earthquake that claimed over 105,000 lives and shaped national resilience. Eighty-eight years later, during the 2011 Tōhoku disaster, X (then Twitter) surged with 1,200+ tweets per minute in the first hour, proving our platform's life-saving power in real-time crisis communication. In 2025, with tools like Grok and Community Notes tailored for Japan, we're more committed than ever to empowering users, sharing knowledge instantly, and building global resilience. Thank you, Japan, for the honor of serving you and your unwavering partnership in innovation year after year.
We are deeply concerned by the findings in the House Judiciary Committee's interim report, led by Chairman @Jim_Jordan, exposing the EU's misuse of the Digital Services Act (DSA) as a tool for censorship. The report reveals how the DSA pressures tech companies to alter their global content moderation policies, effectively exporting EU standards to the US and beyond – threatening free speech worldwide. The classification of relatively benign political statements – such as “We need to take back our country” – as “illegal hate speech,” raises serious questions about the extent of its application. The DSA’s heavy hand targets humor, satire, and personal opinions on issues concerning immigration and the environment. The apparent one-sided enforcement of the DSA involves third parties like "trusted flaggers" and “fact checkers” with conflicts of interest and likely political biases, which can stifle discourse, democratic debate, and the exchange of diverse ideas across the globe. At X, we are committed to safeguarding freedom of expression, and resisting regulatory overreach that imposes censorship on platforms and our users.
Read the report here:
https://t.co/iiMh2zEK0k
Update on where has @grok been & what happened on July 8th.
First off, we deeply apologize for the horrific behavior that many experienced.
Our intent for @grok is to provide helpful and truthful responses to users. After careful investigation, we discovered the root cause was an update to a code path upstream of the @grok bot. This is independent of the underlying language model that powers @grok.
The update was active for 16 hrs, in which deprecated code made @grok susceptible to existing X user posts; including when such posts contained extremist views.
We have removed that deprecated code and refactored the entire system to prevent further abuse. The new system prompt for the @grok bot will be published to our public github repo.
We thank all of the X users who provided feedback to identify the abuse of @grok functionality, helping us advance our mission of developing helpful and truth-seeking artificial intelligence.
On July 3, 2025, the Indian government ordered X to block 2,355 accounts in India, including international news outlets like @Reuters and @ReutersWorld, under Section 69A of the IT Act. Non-compliance risked criminal liability. The Ministry of Electronics and Information Technology demanded immediate action- within one hour- without providing justification, and required the accounts to remain blocked until further notice.
After public outcry, the government requested X to unblock @Reuters and @ReutersWorld.
We are deeply concerned about ongoing press censorship in India due to these blocking orders. X is exploring all legal options available. Unlike users located in India, X is restricted by Indian law in its ability to bring legal challenges against these executive orders. We urge affected users to pursue legal remedies through the courts.
Community Notes show when they’re found helpful by people who normally disagree. What if we could do the same for posts, recognizing posts that are liked by people who normally disagree? We’ve heard requests for this for years since launching Community Notes, and we’re starting a small, experimental pilot to test the concept.
Starting today, a subset of Community Notes contributors — representing a wide range of viewpoints — will occasionally see a new callout in the product. The callout shows based on early and limited Like signals on the post. Contributors can then rate and provide feedback about the post, helping to develop an open source algorithm that could effectively identify posts liked by people from different perspectives.
People often feel the world is divided, yet Community Notes shows people can agree, even on contentious topics. This experimental new feature seeks to uncover ideas, insights, and opinions that bridge perspectives. It can bring awareness to what resonates broadly. It could motivate people to share those ideas in the first place. Ultimately, it could help move the world forward in ways that the people want.
Following the path we used to develop Community Notes, we’re building in public with a small pilot so that this concept can be shaped by the people. We look forward to learning and iterating with you all as we do with Community Notes every day.
Learn more at https://t.co/PvWa3neupo
TikTok’s rolling out Community Notes too! They’re calling it Footnotes although it seems pretty similar
They are using an algorithm very similar to ours (finding agreement from users who typically disagree)
If the reports that the European Commission is considering enforcement actions against X are accurate, it represents an unprecedented act of political censorship and an attack on free speech. X has gone above and beyond to comply with the EU’s Digital Services Act, and we will use every option at our disposal to defend our business, keep our users safe, and protect freedom of speech in Europe.
Under the hood update: Turns out people like to write notes about scams & behaviors they think are potentially not in accord with policies. We rolled out a new topic model to score these independently. Doing so better models perspectives on these notes in particular, and improves modeling of perspectives on notes outside of the topic. This helps the scorer better determine which notes will be found helpful by people from different viewpoints.
As always, code is open source: https://t.co/PlpTud7cFv