#MachineLearning

xAI released major updates to the open-source X recommendation algorithm. The commit introduces the next-generation 'Phoenix (xrex)' recsys framework, video embedding hydrators, post v8 search embedding clients, and revamped safety label processing.

The latest open-source release integrates the next-generation 'Phoenix' candidate pipeline and scorer with a specialized GPU scoring runtime (BDSM). Key updates also introduce post-engager feature hydration, a zero-favorite codivert filter, and an overhaul of confidence-based abuse enforcement controls.

The open-source X recommendation algorithm has been updated to detail the Phoenix-based candidate pipeline within the Rust home-mixer, dedicated immersive video scoring and carousel filtering, and an end-to-end visibility filtering subsystem governing content safety and action limitations.

xAI has open-sourced the modernized X recommendation algorithm under the 'Phoenix/xrex' stack. The update introduces a full Transformer-based ranking architecture leveraging CuTe DSL, FlashAttention-4 variable-length attention kernels, and sequence packing to process user engagement sequences efficiently.

The xAI and X engineering teams have officially released an updated open-source codebase featuring the Rust-based 'home-mixer' architecture. This update details the modern 'For You' and 'Following' candidate pipelines, integrating SimClusters ANN retrieval, brand safety scoring, out-of-network filters, cold-start creator boosts, and value model (VM) ranking debiasing.

A major open-source update to the X algorithm repository reveals the inner workings of the next-generation 'Phoenix (xrex)' neural recommendation framework, Rust-based Home Mixer enhancements for author cold-start scoring and mutual-follow metrics, and low-latency visibility filtering subsystems.