20 years later, modern deep learning barely hits the Netflix Prize 10% target at 9.98%
sedielem · x · 2026-09-21
Two decades after the Netflix Prize, a team with first-hand competition experience retried the challenge using modern hardware, deep learning, AI-assisted coding, and published solutions — yet only reached the 9.98% improvement mark. The thread shows the famous 10% RMSE target remains genuinely hard even with today's tooling, a useful data point for RecSys and ML practitioners.
More from Research
- Study of 1,200 Big Tech acquisitions finds patents get more citations, not killed — soumitrashukla9 · 2026-09-21
- Rethinking classifiers: conversation+tool templates as special cases of inference-time specs — austinvhuang · 2026-09-21
- Fruit-fly connectome with 166,000 neurons completed; open-source voice cloning in 646 languages — thione · 2026-09-21
- Odyssey-3: one frozen world model drives humanoids, cars, drones, and games — thione · 2026-09-21
- Recap: OpenAI launches Astra for Law with 230M-URL legal index; DeepMind unveils Dream-RSI — thione · 2026-09-21
- Recap repeat: DeepMind Dream-RSI cuts discovery-agent calls up to 162x — thione · 2026-09-21