Ex-Netflix engineer's 3 questions debunk 'terrible recommendations': your profile is polluted by passive data
Aiden_Tech_Ai · x · 2026-09-16
A user complained his Netflix recommendations kept repeating — until a friend who spent 4 years on Netflix's recommendation team asked three questions: does anyone else use your profile (his girlfriend watches K-dramas, his kid watches cartoons), do you press thumbs-down (no), do you rate anything (no).
The verdict: the algorithm isn't broken, it's uninformed. Three years of passive clicking with zero active feedback made the system conflate the whole household's tastes with his. A neat analogy for AI use: output quality depends on the context you feed it, not just the model.
More from Fun
- Star Wars AI Video Parody Hailed as the Funniest AI Creation Yet — DeryaTR_ · 2026-09-16
- "AI Plays Doom" Demo Debunked: Text State Input, Solvable in ~30 Lines of Code — banteg · 2026-09-16
- Gary Marcus calls Nate Silver's AI takes a 'strawman crusade', weighs in on a rebuttal — GaryMarcus · 2026-09-16
- From Camera Roll to Boxing Game: GPT-6 Builds 3D Character via MCP — heyshrutimishra · 2026-09-16
- NoSpoon agent cranks out hundreds of AI microdrama episodes daily before going internal — sachinmaya1980 · 2026-09-16
- No AGI means 100% asteroid doom: the quip puncturing p(doom) debates — seanmcdonaldxyz · 2026-09-16