X's recommender is multimodal: CLIP and MultimodalPostEmbedderV5 embeddings feed Phoenix retrieval
cocktailpeanut · x · 2026-11-11
cocktailpeanut shares a TIL: X's recommendation system is multimodal. Per Grok, posts get dense embeddings from text, images, and video frames (plus ASR transcripts) via models like MultimodalPostEmbedderV5 and CLIP; these feed Phoenix retrieval through semantic IDs, while Grox uses vision-language models for content understanding and safety — all confirmable end-to-end in the open-sourced x-algorithm.
More from Infra
- HF tokenizers v1.0.0-rc.4 restores Python APIs and encodes up to 17x faster — art_zucker · 2026-11-11
- MacPaw's Elix claims fastest LLM inference on Apple silicon, winning all 33 benchmarks — JosephJacks_ · 2026-11-11
- Which tasks have you fully moved from cloud models to local ones? — FulcraDynamics · 2026-11-11
- NEAR AI discloses breach: SGLang RCE flaw let attacker reach infra, ~100 users' data exposed — bingxu_ · 2026-11-11
- Bittensor launches Gamma Tokens for native agent payments across subnets — markjeffrey · 2026-11-11
- Meta's Muse Spark Open, 816B params with 1M context, goes live on ai& inference — DavidBennett__ · 2026-11-11