X Revealed Their Secret Algorithm on Github #algorithm #twitter #tech

Why It Matters

Open‑sourcing X’s feed algorithm democratizes advanced recommendation technology, enabling new products while spotlighting privacy and competitive implications.

Key Takeaways

• XAI open‑sourced its entire feed recommendation algorithm on GitHub
• Two candidate pools: “thunder” (in‑memory) and “phoenix” (global ML search)
• Merged candidates filtered, then ranked by a Grok transformer model
• Model predicts likes, replies, reposts, clicks, and blocks for ranking
• Enables developers to build custom recommendation engines or viral‑post analyzers

Summary

The video announces that XAI has published the full source code for the algorithm that curates the X (formerly Twitter) home feed, making the previously proprietary recommendation engine publicly available on GitHub.

The system draws content from two distinct sources – “thunder,” an in‑memory cache of recent posts from accounts you follow, and “phoenix,” a machine‑learning engine that scans the entire platform for posts likely to interest you even from unknown accounts. After merging, the pipeline filters out spam, blocked users, and already‑seen items before feeding the candidates to a Grok‑based transformer that estimates the probability you will like, reply, repost, click, or block each tweet.

The presenter highlights that positive interactions boost a post’s rank while negative signals demote it, and that the open code lets developers replicate the candidate‑sourcing, ranking, and diversity‑scoring stages. He suggests building niche newsletters, community feeds, or tools that reverse‑engineer why a particular tweet went viral using the same architecture.

By exposing the core recommendation logic, XAI lowers the barrier for startups and researchers to create sophisticated feed algorithms, potentially intensifying competition while also raising questions about data privacy and algorithmic transparency.

Original Description

xAI just open-sourced the algorithm powering X's For You feed and it's a fascinating look under the hood.

The system pulls content from two sources: Thunder, an in-memory store of posts from people you follow, and Phoenix, an ML engine that finds content from accounts you've never seen. Both get ranked by a Grok-based transformer that predicts how likely you are to like, reply, repost, or even block a post.

The real opportunity?

You can study and adapt this same pipeline to build your own recommendation engine for a newsletter, community platform, or startup.

Full code on GitHub: https://github.com/xai-org/x-algorithm

#xAI #XAlgorithm #OpenSource #MachineLearning #AINews #ForYouFeed #Grok #TechShorts #ArtificialIntelligence #BuildWithAI