Amazon is racing to match Google in the AI-powered voice assistant market. Unable to develop sufficiently advanced AI models in-house, the company has partnered with Anthropic, an OpenAI competitor.
Five sources familiar with the plans told Reuters that Amazon will use Anthropic's Claude AI models for most of the upcoming Alexa redesign, instead of relying solely on its own AI technology.
The new "Remarkable" Alexa version is scheduled for an October launch. Users can opt for the more capable Claude-powered version for $5-10 monthly, while the free "Classic" Alexa will still be available. Some Amazon employees, however, question whether customers will pay for a service that was previously free.
By incorporating Claude, Amazon aims to offer more natural conversations, personalized shopping suggestions, news summaries, and enhanced smart home controls.
Amazon's own AI model falls short
Amazon, like Microsoft with OpenAI, is partnering with an external AI provider. This decision seems driven by the underwhelming results of Amazon's in-house AI development.
Sources report that Amazon's initial tests using its own AI for the Alexa upgrade yielded disappointing outcomes. The AI had trouble generating responses and required up to seven seconds to process and answer prompts. Amazon's ambition to outperform Claude 3 with its "Olympus" model appears unmet.
The specific Anthropic model Amazon will use for Alexa is not yet known. Compact, efficient models such as Claude 3 Haiku or Claude 3.5 Sonnet are likely candidates, offering quick responses and lower inference costs.
To date, discussions have focused on software updates, not new hardware. Whether the Claude-powered "Remarkable" Alexa will need specialized devices or function on existing hardware remains unclear.
Cloud processing likely, local AI unlikely
Amazon will likely process all Alexa requests via the cloud, independent of device capabilities. This approach differs from Google's strategy of running local Gemini voice models on its latest Pixel 9 smartphones.
Future smart speakers with built-in neural processing units (NPUs) could offer better privacy by handling requests locally. However, releasing such a product would be an unusual move for Amazon.
Integrating generative AI into a voice assistant is a significant step. One of Alexa's primary functions is controlling smart home devices - a task that demands reliability.
Current speech recognition systems using predefined patterns work adequately for this purpose. It's unclear whether generative AI will improve upon this or potentially introduce issues like refusing to activate lights due to hallucinations. The real-world performance of AI-powered Alexa in smart home control remains to be seen.
Will Amazon's investment pay off?
Amazon and Anthropic have an existing partnership, with Amazon providing computing power and investing up to $4 billion in the AI company. This deal gives Amazon access to Anthropic's early technologies.
It's unclear whether Amazon will pay additional fees to use Claude in Alexa. By comparison, Apple reportedly isn't paying OpenAI for ChatGPT integration, with exposure in Apple's ecosystem serving as compensation.
The Amazon-Anthropic deal has drawn regulatory scrutiny. International competition authorities are investigating this agreement, along with other AI partnerships such as Microsoft and OpenAI, for potential effects on market competition.
