X Algorithm “Is Dumb,” Musk Says as Platform Fully Open-Sources Its Ranking System

X has taken another significant step toward algorithmic openness by releasing the latest version of its ranking system, the engine behind the platform’s “For You” feed. 

Powered by the same transformer architecture used in xAI’s Grok model, the new X algorithm is now fully open-sourced on GitHub, marking one of the most transparent moves from a major social platform in recent years.

Elon Musk set the tone with a candid post acknowledging both the shortcomings and the openness of the system. 

“We know the algorithm is dumb and needs massive improvements,” he said. “But at least you can see us struggle to make it better in real-time and with transparency. No other social media companies do this.”

X Engineering Highlights Grok-Based Architecture

The engineering team behind X confirmed the broader technical shift: the platform’s recommendation infrastructure is now powered by transformer models similar to those at the core of Grok.

“We have open-sourced our new X algorithm, powered by the same transformer architecture as xAI’s Grok model,” the @XEng account announced. The system is now available for anyone to inspect, critique, or build upon through the public repository.

This release marks a follow-up to previous open-source efforts from X, but this version features significant upgrades built to handle the scale and complexity of personalized timelines.

Grok Breaks Down How the X Algorithm Works

Shortly after the release, a Grok explanation of how the new system selects and ranks posts was published on X. 

According to Grok, the algorithm’s workflow begins the moment users open their “For You” feed. The system first gathers personal interaction data, who the user follows, which posts they have engaged with, and how they have historically interacted with content. This forms the baseline for personalization.

Next, the X algorithm collects posts from two sources: accounts the user follows (in-network content) and wider content from across the platform (out-of-network). 

In-network posts come from a high-speed memory system optimized to fetch fresh updates instantly. Out-of-network content is selected via AI-driven matching that compares a user’s interests with posts across all of X.

How Posts Are Enriched, Filtered, and Scored

Once the algorithm compiles potential posts, each candidate is enriched with additional metadata including text, images, videos, author details, verification status, and whether the post is behind a paywall.

The system then removes duplicates, old posts, content from blocked or muted accounts, and anything the user has already viewed. It also filters out keywords the user has asked not to see.

The next stage involves the Grok-based scoring model. This AI evaluates the user’s historical behavior and predicts the likelihood of various interactions, likes, replies, reposts, clicks, or negative responses. Each prediction contributes to an overall ranking score.

Those scores are then adjusted to ensure diversity, preventing overexposure to any single account and maintaining a balance between familiar and new content.

Final Feed Assembly and Moderation

After sorting the posts by score, the algorithm selects the highest-ranking candidates for the final “For You” feed. Before the feed is delivered, the system conducts another moderation sweep, removing deleted posts, spam, violent material, and any inappropriate content.

The result is a curated feed shaped by personalization, AI matching, ranking predictions, and multi-layer filtering. With the code now public, users and developers alike can see exactly how decisions are made under the hood.

A Transparency Strategy That Sets X Apart

While Musk has made it clear that the algorithm still has flaws, he emphasized that the openness of the process is the core principle. The public release allows researchers, engineers, and users to inspect and critique the logic behind the feed—something no other major social media platform currently offers.

For X, the project is both a technical milestone and a branding statement: even if the algorithm isn’t perfect, it is visible, inspectable, and improving in real time.