True to Elon Musk's transparency pledge, X has officially open-sourced the code for its "For You" recommendation engine. The full technical details are now available on the xAI GitHub repository, offering a rare look under the hood of how the platform decides what users see.
The Two Pillars: Thunder vs. Phoenix
The X recommendation system operates through two distinct pipelines designed to balance familiarity with discovery:
Thunder: This internal candidate generator focuses exclusively on the accounts you follow, ensuring you don't miss updates from your primary network.
Phoenix: This discovery engine hunts for "Out-of-Network" content, identifying viral or relevant posts from users you don’t follow based on your current interests.
The Scoring Process & Grok’s Role
Once Thunder and Phoenix generate a list of potential posts, they undergo a rigorous filtering process. The system checks for "Blue Check" status, removes duplicate content, and applies filters for blocked accounts or restricted keywords.
Finally, Grok, X's proprietary AI, processes these candidates and assigns a score to each post based on various engagement factors, ultimately ranking them to create your personalized feed.
Key Takeaways for Creators (Algorithmic Insights)
The source code revealed several critical behaviors that every user should know:
Recency is King: Phoenix only analyzes your last 128 interactions. This means your "For You" feed is heavily influenced by what you liked or replied to in the last hour, rather than long-term history.
Quality over Frequency: Due to the "Anti-Spam/Duplication" filter, posting too frequently from a single account can actually diminish visibility.
Link Penalties: The code confirms that posts containing external links may receive a lower score, as the platform aims to keep users within the X ecosystem.
Elon Musk acknowledged that the current algorithm is "far from perfect" and has committed to monthly updates to make the system smarter and more transparent.
- The code indicates that "Replies" and responses from the post owner have significantly more weight in pushing a post to greater visibility than simply liking it.
- The algorithm gives extra points to posts with images or videos, especially videos with a high watch time, which greatly increases the post's "For You" score for other users.
- Things that immediately negatively impact a post's score include being muted, blocked, or reported, as well as posting content that the AI deems "clickbait."
- Although the algorithm prioritizes visibility based on interest, paying X Premium subscribers receive a significantly higher ranking multiplier than regular users.
cURL Scraps Bug Bounty Program: Creator Cites AI-Generated "Noise" as Primary Reason
No comments:
Post a Comment