Netflix tests language model as alternative to hand-built recommendation logic
Netflix swapped thousands of hand-built recommendation rules for a language model. Early results beat the legacy engine.
Why it matters
A major streaming platform is moving from engineered feature logic to LLM-based personalization—a shift that could reshape how recommendation systems work at scale and influence other consumer platforms to follow.
The key facts
8 to knowNetflix developed internal LM called GenRec
GenRec converts viewing behavior to plain text instead of relying on hand-crafted features
GenRec outperformed Netflix's legacy recommendation engine in testing
Netflix describes result as 'early but promising'
Shift from feature engineering to LLM-based personalization
Netflix internal model GenRec outperformed legacy recommendation engine
GenRec converts viewing behavior to plain text instead of hand-crafted features
Test replaces years-old recommendation logic with language model approach
Go to the source
The Decoderthe-decoder.com
Publisher excerpt: Netflix pitted its years-old recommendation engine against an in-house language model called GenRec and says it got better results. Instead of relying on thousands of hand-crafted features, GenRec converts viewing behavior into plain text. Netflix itself calls it "an early but promising step."