Netflix replaced its mature recommendation engine with an llm that treats watching history as a…
netflix replaced its mature recommendation engine with an llm that treats watching history as a language problem, and genrec beat the old system with roughly 40x less labeled data while scoring statistically significant online lifts. spotify’s glide and google’s tiger are building the same transformer-on-interaction-sequences stack, and recommendation-driven plays now cover about 82% of netflix viewing hours.
the collaborative filtering era ended without most users noticing.
Context
Netflix's tech blog (Jul 30, 2026) describes GenRec, an LLM-backed ranker post-trained on Netflix data that it says can match or exceed a mature production recommender; the arXiv paper describes moving from a ranker with thousands of engineered features to one driven by verbalized user histories.
Spotify's KDD 2026 paper covers deploying LLMs for personalized generative recommendations, and its GLIDE paper describes a production generative recommender for podcast discovery using Semantic IDs.
The LLM ranker, verbalized histories and Spotify's GLIDE match the sources. The 40x less labeled data and 'statistically significant online lifts' figures were not seen in the sources read, so unsupported here, not refuted. in the excerpts read; Netflix says GenRec can match or exceed the old ranker. 'Replaced' is stronger than the paper's description of a transition. Google's TIGER was not covered by the sources read. Netflix's 80 percent of hours influenced by recommendations is from a 2015 ACM paper, not a 2026 figure; the post says about 82 percent, which was not seen in the sources read, so unsupported here, not refuted. 'Without most users noticing' is the author's.
Related work
- ACM: the Netflix recommender system ↗2015 paper.
- arXiv: GLIDE at Spotify ↗Mar 2026.
- Netflix Help Center: how recommendations work ↗Help article.
Watch next
- Find the 40x labeled data figure and the 82 percent viewing share.
Sources
- Netflix Tech Blog: GenRec, towards LLM-native recommendationnetflixtechblog.com
- arXiv: GenRec, an LLM-backed recommendation ranker at Netflixarxiv.org
- Spotify Research: from habits to discoveryresearch.atspotify.com
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 1 October 2026 at 20:47 IST. Sources are the papers and datasets the note draws on.
View the original post ↗Embed this note
More notes
The air is now being asked to keep its own ledger
the air is now being asked to keep its own ledger: ecmwf’s aifs compo becomes the first ai model to forecast atmospheric composition globally every three hours, cleanair simulates 365 days of pm2.5 over china in ten seconds, and a unified framework maps six pollutants at one kilometer across the whole country. the air now files its own composition report.
read the note →The current is now being asked to draw its own map
the current is now being asked to draw its own map: china’s langya 2.0 predicts six ocean phenomena including internal waves and mesoscale eddies, a deep net called wenhai resolves eddies globally with air sea flux formulas built in, and scripps infers surface currents from the way temperature patterns deform in satellite images. the ocean now files its own circulation report.
read the note →The soil is now being asked to report its own carbon
the soil is now being asked to report its own carbon: a nix color sensor paired with generative data augmentation predicts soil organic carbon without a lab, random forest drives 74 percent of soil health mapping studies, and sentinel 2 tracks five year carbon change across france and italy from 922 samples. the dirt now files its own carbon account.
read the note →