Netflix tests a language-model recommender as RayNeo drops the camera for text
Netflix's GenRec trial delivered small online gains, with no full replacement planned. RayNeo is preparing camera-free text glasses, while Beijing's Jinguyuan gives model credit with meals but says usage trails press attention.
Artificial Intelligence··Evening
Netflix tests an offline gain on the homepage
Netflix's GenRec model turns viewing history into plain text and searches for patterns without thousands of hand-built features instead of representing the history as numeric vectors. It ranked 1.6 per cent above the production system offline. In a 4-week A/B test on about 10 per cent of traffic, short-term homepage engagement rose 0.115 per cent and the long-term core measure rose 0.006 per cent. The results come from Netflix's engineering blog, and a full replacement is not planned.[1]
RayNeo puts text in view instead of adding a camera
The RayNeo iO glasses have neither a camera nor a speaker. A green waveguide display with 97 per cent transparency and about 1,300 nits places text in the wearer's field of view. 4 microphones and a bone-conduction sensor provide input for conversation summaries, action items, calendar suggestions, a teleprompter, and speech-to-text translation in 40 languages. A 9.99-dollar monthly subscription adds access to more models; the report gives neither a device price nor a release date.[2]
Jinguyuan puts model credit beside the meal
Jinguyuan, a shop in Beijing, struck a deal with an AI cloud provider in July and began giving customers token coupons worth 10 yuan, about 1.50 dollars, for model subscriptions. Its owner, Li, also had Alibaba's Qoder build a queue-status site for the 2 branches. Li says the promotion has brought more press attention than actual use. The coupons document a small-business customer experiment; the level of broader adoption was not measured.[3]
