Amazon says that Alexa+, the new and improved Alexa unveiled on Wednesday, is powered by a “model agnostic” system that always uses the “best” AI model for a given task.
On stage at a New York City press event, Amazon VP Daniel Rausch explained that Alexa+ draws on Bedrock — the company’s cloud platform designed to let organizations experiment with generative AI models — to power its various capabilities.
Among the models Alexa+ uses are Nova, Amazon’s in-house generative AI model family; as well as models from close partner and collaborator, Anthropic.
This “model agnostic” approach enables what might be considered “agentic” capabilities. Alexa+ leverages a new system called “experts” for particular tasks, Amazon says. Experts orchestrate and execute Amazon services as well as those from third parties, like AI startup Suno’s music-generating tools.
Rausch said on day one, Alexa+ will work with “tens of thousands” of devices and services.
Some of these include news services. Amazon says that it’s partnered with publications including Time, Reuters, and the Associated Press to supply info for answers related to current events, financial market movements, and more.
That only scratches the surface of what Alexa+ can do. According to Rausch, Alexa+ can navigate websites to complete tasks on a user’s behalf. For example, it can find a professional to repair a broken oven by searching the web, and even contact the repair shop directly.
“Alexa takes a couple of minutes to navigate a website [and it] comes back and tells me it’s done,” Rausch said on stage. “I’m going to get a notification on my phone that lets me know [when Alexa is finished.]”
Experts also allow Alexa+ to coordinate multiple services at once, Rausch said. For example, making a dinner reservation via OpenTable, then booking an Uber to the restaurant and texting the plans to a contact. Developers will be able to tap into this functionality via Alexa AI Multi-Agent SDK, which is launching soon in preview.
Post a Comment