X open-sources feed algorithm, with AI-driven scoring at its core

X open-sourced its "For You" feed algorithm on Tuesday, which is powered by the same transformer architecture as xAI's Grok model.

Instead of following manually crafted rules, the feed system filters and ranks posts using an AI model, with scoring rules based entirely on the user's past interactions.

Content with higher scores are selected to create the feed. The scoring tool analyzes users' past behavior to predict preferences, aiming to boost positive actions, such as likes, and minimize negative ones, for instance, blocks.

The algorithm collects potential posts from two main sources: in-network and out-of-network. The former refers to posts from people the user already follows, while the latter gathers new discoveries from across all of X that might interest the user.

To maintain content quality, the system removes recently viewed content and posts from blocked accounts before scoring. After scoring, it filters out deleted or inappropriate content. Additionally, the algorithm ensures content variety by preventing repeated posts from the same creator.

Elon Musk posted on X, "We know the algorithm is dumb and needs massive improvements, but at least you can see us struggle to make it better in real-time and with transparency."

The full code is live on GitHub and will be updated every four weeks.