Amazon releases new version of Alexa, utilizing generative AI technology

"Alexa, let's chat" - After nearly a decade since the introduction of its voice-activated Echo devices, e-commerce and cloud computing giant Amazon announced today that its iconic voice assistant Alexa will be replaced by a new, customized large language model (LLM) that leverages the generative AI trend in Silicon Valley to give Alexa more capabilities and human-like conversational qualities. This announcement was made by David Limp, Senior Vice President of Amazon Devices and Services, at the company's luxury "HQ2" headquarters in the suburbs of Washington, D.C., to attendees of VentureBeat. The new Alexa LLM will soon be available for free preview on Alexa devices in the United States, with Amazon claiming that it is "smarter, more conversational," and its voice is more "realistic" and "casual." According to Rohit Prasad, SVP and Chief Scientist of Amazon's Artificial General Intelligence, this news signifies that "our favorite assistant has undergone a significant transformation." Amazon aims to surpass OpenAI's ChatGPT with more "real-world" capabilities OpenAI's powerful ChatGPT application has shocked the world and become a household name overnight, while Amazon's entry into the conversational LLM field comes almost a year later. However, the company claims that the new Amazon LLM is worth the wait. Amazon states that unlike ChatGPT, which closed its knowledge base at the end of 2021 or early 2022, Alexa LLM provides "real-time information" and is "more interactive," containing more information than previous versions of Alexa. During the event, Amazon directly mentioned ChatGPT, stating that its Alexa application "goes beyond ChatGPT on browsers or mobile devices" and provides users with "real-world applications" such as recipes, travel ideas, and even writing poetry for them. Prasad said, "What sets our LLM apart is that it not only tells you things, but also teaches you how to do things." To illustrate this idea, Limp conducted a live demonstration in front of the media and Amazon employees, asking Alexa about the performance of his "favorite football team." Alexa remembered that he was referring to Vanderbilt University, showcasing its personalized attributes. If the user's favorite team wins, Alexa responds with a "happy" voice. Limp also asked Alexa to write a message and send it to his phone, reminding his friends to watch an upcoming Vanderbilt football game. Alexa completed this action within seconds. Amazon also showcased a promotional video that implies the new Alexa LLM is a "family member" for users. Four core components and third-party applications Prasad stated that the new Alexa LLM is built around four key components: a large language model, real-world devices and services, personalized context, and responsible AI. In fact, Heather Zorn, Vice President of Alexa Connectivity and Infrastructure at Amazon, mentioned that developers can already integrate their own "customized, specially designed" third-party LLMs into Alexa. One popular developer in this space is Character.ai, which allows users to create and interact with various fictional characters and prototypes, offering 25 different personality types. Another developer, Splash, provides users with the ability to create and preview songs through its Alexa integration. Impressive technology Prasad mentioned that Alexa's text-to-speech engine now "perceives emotions and tones better and conveys similar emotional changes in the output based on the speaker's tone." It also includes a new automatic speech recognition system designed for conversations, which uses a massive Transformer model that is the best in its class and will continue to improve. For Amazon Echo Show devices with built-in screens and cameras, Amazon users participating in the Visual ID program can simply look at their devices and speak to them without having to repeatedly say "Alexa." They can interact with the assistant as if they were interacting with another person. Prasad said, "This is thanks to the collaboration between visual processing and acoustic models on the device, so it knows whether you are talking to Alexa or someone else in the room."