Alexa just went from being a somewhat ordinary, if trusted, digital assistant to a powerful enigma. Alexa+ replaces the plodding of the original with conversational intelligence, proactivity, and true AI. It’s a lot to take in and raises a lot of questions. I managed to get some answers from Daniel Rausch, Amazon’s VP of Alexa and Echo and of the people who helped build Alexa Plus, to get some answers.
Even though I caught up with Rausch at the end of a long day, he seemed energized and clearly quite proud of Amazon’s creation.
I started by asking him about something bugging me since the launch: How much of Alexa+ is Claude? Anthropic, Claude’s developer, was touted as a partner in the development of Alexa Plus and listed during the presentation alongside Amazon Nova, Amazon’s own large language model.
Rausch, though, quickly disabused me of that notion.
“You saw Amazon Nova models up there and they are definitely where we start at Amazon,” Rausch began, “we always start with our own technology.”
Raushe explained that there is “unbelievable price performance in those models, unbelievable latency, unbelievable accuracy.”
Bedrock principles
It goes further than that, though. Rausch explained that Amazon’s Bedrock is a sort of cloud-based foundation for all its generative AI work. Anthropic is a “really important partner,” said Rausch, but the system Amazon’s built is “model agnostic.”
“Bedrock’s goal is to serve highly capable, state-of-the-art models,” he said. That means the system will pick and choose the best models for the job.
It makes sense this also may have been Rausch’s way of not identifying which models, including Claude, are used when. “With Alexa,” he added, “we have access to the full suite.”
It goes further than that, though. Large language models are not the end or destination. Rausch told me the Alexa+ experience extends and builds upon them. He discussed the “information experts” idea introduced during the launch event. The models use these experts to gather relevant and factual information. Rausch used the example of his constant queries about baseball, specifically the Yankees, but added that the system is smart enough to know that, in his house, only he likes to talk endlessly about baseball while his daughter has no interest.
“I would say the models are helping orchestrate the overall experience and are the foundation and are helping us build the rest,” Rausch told me.
That art
During the kids’ section of the unveiling, or the part that described the new “Explore and Stories with Alexa,” I noticed what looked like generative AI art.
Rausch confirmed that Alexa+ generates that art on the fly based on kids’ ideas. During development, Rausch put the tools in front of some relatives’ kids to gauge their reactions: “It is super-fun. A kid is describing the story, Alexa is helping them explore, ‘Hey, what would you like to write a story about?'” For instance, the kid describes a bearded dragon playing a saxophone. “Alexa’s off drawing some creative artwork, asking about the path of the story, ‘Where does the bearded dragon live?’ or ‘What city is the bearded dragon visiting?’ Of course, kids are unlimited in their imagination. They’re riffing. They’re saying, sometimes, things adults can’t come out with.”
It does sound like fun. I noticed that the images I saw looked like generative AI, and Rausch told me that was exactly what they were. He would not, however, reveal which generative image model Alexa+ is using. All Rausch would only say that “it’s from the models on Bedrock.”
Security
As with any generative AI, the key to Alexa+’s utility is data or rather your data. Generative tasks will be handled in the cloud, but Raush told me that all of it will be encrypted “in transit.”
“It is incredibly secure and meets our standard practices of trust generally, which include deep security and privacy,” he added.
“We always start with our own technology.”
Daniel Rausch, Amazon VP Alexa & Echo
Naturally, this led me to ask about the kids’ technology and what safeguards Amazon has built around the generative image creation.
Rausch described it as “incredibly safe,” and said there are many safeguards in place to ensure “children are always kept safe.”
I know many companies say their generative image platforms are safe, but Amazon has a track record in developing kid-friendly platforms and systems. The “Explore and Stores with Alexa” is an extension of all that work.
Not always a screen
It was hard to ignore the pervasive use of the Echo Show 21 smart display throughout the Alexa+ demo. I’ve naturally been wondering about all the Echo Smart speakers out there, at least the ones new enough to support Alexa Plus. What will the experience be like with them?
During the demos, Amazon actually cranked up the length of responses for the Echo 21 screens to show off what Alexa knows. But no one wants to stand around staring at a speaker waiting for these responses. Rausch told me that they designed Alexa Plus to deliver more concise responses on the speakers. All of this is customizable.
“The idea is we picked the right kinds of responses and the right interactions for the device and the modality that you’re in,” he added.
Is Amazon ready
When Amazon delivers Alexa+ in March, that could mean millions of Echo owners suddenly having generative AI conversations with the newest chatbot on the block. It’s potentially a big computational lift. Is Amazon ready?
“Yes,” Rausch smiled, “It is very nice to have AWS at Amazon.”
Amazon’s massive cloud computing platform supports countless websites and services and will now supply the bandwidth for Alexa+. Still, it won’t be everything, everywhere, all at once. Amazon is planning on rolling out Alexa+ in waves.
And now I feel like I’ll understand Alexa+ a little better when those waves first hit the digital shore.
Leave a comment