Dave Limp, SVP of Amazon Devices & Services, revealed during the company’s 2023 Devices event on Wednesday that Alexa will soon undergo a significant update that will put its conversational capabilities more in line with contemporary chatbots like Google Bard or OpenAI’s ChatGPT. Soon, a huge language model specifically designed for the purpose will power the venerable digital assistant and be present in almost all new Echo devices.
“Today’s model is voice-optimized,” Limp said the gathered assembly. “It does all the things we know our customers love, like real-time information access, smart home control, and making the most of entertainment at home.”
Having studied its “ambient intelligence” systems for more than ten years, Amazon is no stranger to genAI technology. The background operations of Alexa devices have long been powered by generative AI models, particularly Alexa Teacher. About nine years ago, Limp stated, “we started doubling down on the home and we had an epiphany with generative AI within reach.” We came to understand that all of the money spent on R&D in the consumer electronics sector was going directly into the development of mobile phones. Everything was designed with the phone in mind, including the SOCs, screens, chip sets, and sensors.”
He admitted, “That was understandable.” It’s an industry with annual revenue in the billions. However, your house, where you have spent the majority of your life, was all but forgotten at the same moment.”
Limp stated that the new model will “assist us in taking the next steps towards a remarkably different customer experience” and that it will be “larger and more generalized.” In order to achieve this, Amazon designed the LLM with five core functions in mind, then fine-tuned the model for voice applications as opposed to mobile screens.
- Conversational: Over the past nine years, we have researched the elements of a successful discussion. It’s not only words; it’s also body language, knowing who you’re speaking to, gestures, and eye contact.
- Real-world applications: Unlike browser tabs, Alexa is a part of the real world. Additionally, one of these LLMs’ unresolved issues is how to properly communicate with APIs.
- Personalization: Your family and you must be the focus of LLM in your house.
- Personality: “Alexa, powered by this LLM, will have opinions—and it will definitely still have the jokes and Easter eggs you’ve come to love from Alexa. We’ve always said that the most boring dinner party is one where nobody has any opinions.”
- Trust: We need both performance and reliability in order to create an AI that lives up to its promises. “My home is highly Alexa-enabled, and I wouldn’t bring anything in that I felt would jeopardize the privacy of my family.”
Simply put, the speech optimizations eliminate the need for you to repeat yourself to Alexa each time you speak to it. Clients who have registered for the business’s Visual ID program just need to face the screen in order to initiate communication. Furthermore, the new Alexa will soon adjust its tone and mood according to the topic of the conversation and be more understanding of speech that falters or has pauses in it.
This LLM will also be “connected to hundreds of thousands of real-world devices and services via APIs,” according to a press release from the business. “.Additionally, it improves Alexa’s capacity to comprehend ambiguity and subtlety, much like a human would, and to make wise decisions.”. Consequently, users will soon be able to schedule sophisticated requests, such as “Alexa, every weeknight at 9 PM, turn on the porch light, dim the lights upstairs, and turn on the fan in the bedroom, all with spoken commands.”
During an on-stage presentation on Wednesday, Limp attempted to showcase such natural conversational talents, but Alexa was not very helpful. In fact, the AI clearly ignored two of Limp’s spoken cues, forcing him to embarrassingly repeat himself.
The new model is by no means the sole genAI initiative from Amazon. The business recently added a number of AI-based features to its Thursday Night Football broadcasts during the NFL season, and it also published a generative model to assist its e-commerce vendors in creating product listings. The Writers Guild of America has also criticized the corporation for allowing the shop to use AI-generated book listings that significantly violate copyrighted works and sometimes suggest eating questionable mushrooms.
Beginning in 2024, current Echo owners will be able to download the updated LLM as part of a complimentary preview on their current devices as well as on every newly sold Echo device.
