← back
Netflix's Big Bet: One model to rule recommendations: Yesu Feng, Netflix
Takeaway
Netflix is replacing many specialized recommendation models with one transformer foundation model over rich user-event tokens, applying LLM scaling laws to recsys.
Summary
- Yesu Feng describes Netflix's bet on a single autoregressive transformer foundation model for all recommendation use cases (rows, items, pages, games, live streaming).
- Historically each surface had its own specialized model with duplicated feature/label engineering — unscalable as content types expand.
- Two hypotheses: (1) scaling laws apply to recommendation via semi-supervised learning; (2) one foundation model lifts all downstream canvas-facing models simultaneously.
- Tokenization is critical: each token is a user-interaction event with many fields (not a single ID), so granularity decisions propagate through model quality.
recommendationsfoundation-modelsnetflix
Original description
Discuss the foundation model strategy for personalization at Netflix based on this post https://netflixtechblog.com/foundation-model-for-personalized-recommendation-1a0bd8e02d39 and recent developments. About Yesu Feng Yesu Feng is a staff research scientist/engineer at Netflix, his work focused on generative foundation models for personalized recommendation. Before Netflix, he was at Linkedin and later Uber, worked on homepage feed and marketplace optimization, respectively. Recorded at the AI Engineer World's Fair in San Francisco. Stay up to date on our upcoming events and content by joining our newsletter here: https://www.ai.engineer/newsletter