X for You Feed Algorithm (Updated May 15th) (github.com)

X has announced an update to its "For You" feed algorithm, which now utilizes a Grok-based transformer model to enhance its recommendation system. This system blends in-network content—posts from followed accounts—with out-of-network posts surfaced through machine learning. The update includes a new end-to-end inference pipeline that consolidates the retrieval and ranking processes, allowing for more efficient performance and streamlined integration in real-time user feeds. The significance of this update lies in its reliance on the Grok-based transformer, which completely eliminates manual feature engineering, enabling the model to autonomously learn from user engagement patterns. Key improvements include a pre-trained mini Phoenix model packaged for instant use, a Grox content-understanding pipeline for tasks like spam detection and post classification, and an ads blending system that respects user content preferences. This architecture not only boosts the relevance of posts shown to users through better engagement predictions but also enhances ad positions in a user-friendly manner, ensuring diverse content while maintaining brand safety. This represents a major step forward in AI-driven content curation, highlighting the growing reliance on sophisticated machine learning techniques to personalize user experiences.