Jeff Bezos opens up on Amazon Go and Alexa
Jeff Bezos has posted an update on Amazon Go, the e-commerce giant’s new cashierless store which opened to the public earlier this year in Seattle.
In his annual letter to shareholders, Bezos said: “Since opening, we’ve been thrilled to hear many customers refer to their shopping experience as “magical.” What makes the magic possible is a custom-built combination of computer vision, sensor fusion, and deep learning, which come together to create Just Walk Out shopping. With JWO, customers are able to grab their favourite breakfast, lunch, dinner, snack, and grocery essentials more conveniently than ever before. Some of our top-selling items are not surprising – caffeinated beverages and water are popular – but our customers also love the Chicken Banh Mi sandwich, chocolate chip cookies, cut fruit, gummy bears, and our Amazon Meal Kits.”
Elsewhere, Bezos said that customer embrace of Alexa continues, with enabled devices among the best-selling items across all of Amazon. “We’re seeing extremely strong adoption by other companies and developers that want to create their own experiences with Alexa,” he commented. “There are now more than 30,000 skills for Alexa from outside developers, and customers can control more than 4,000 smart home devices from 1,200 unique brands with Alexa.”
He added that the foundations of Alexa continue to get smarter. Amazon has developed and implemented an on-device fingerprinting technique, which keeps a device from waking up when it hears an Alexa commercial on TV (this technology ensured that the Alexa Super Bowl commercial didn’t wake up millions of devices).
"Far-field speech recognition (already very good) has improved by 15% over the last year; and in the US, UK, and Germany, we’ve improved Alexa’s spoken language understanding by more than 25% over the last 12 months through enhancements in Alexa’s machine learning components and the use of semi-supervised learning techniques," Bezos added. "These semi-supervised learning techniques reduced the amount of labelled data needed to achieve the same accuracy improvement by 40 times. Finally, we’ve dramatically reduced the amount of time required to teach Alexa new languages by using machine translation and transfer learning techniques, which allows us to serve customers in more countries (like India and Japan).”