Apple's artificial intelligence (AI) voice assistant "Siri" is set for a sweeping overhaul based on Google's AI model "Gemini." Apple is pushing to release the next-generation Siri in Sep., U.S. information technology (IT) outlet The Information reported on the 3rd, citing multiple sources.
The new Siri will use On-device AI as its foundational architecture, processing AI computations on Apple devices such as the iPhone. To that end, Apple is reportedly conducting "distillation" work to build a small AI model that can run inside smartphones, based on the Gemini model. Distillation is a technique that transfers the performance of a large AI model to a smaller one, reducing it so it can run on the device.
On-device AI does not send user data to external servers, which is advantageous for privacy. However, due to hardware constraints, it has limits with complex commands or lengthy reasoning tasks. To offset these limits, Apple is expected to adopt a hybrid architecture that handles some high-difficulty tasks on Google Cloud. In this process, it is said to use Nvidia AI chips and confidential computing to maintain data encryption and security standards.
Apple initially sought to handle AI computations with its "Private Cloud Compute" servers based on its in-house M‑series chips, but appears to have concluded that running giant models with trillions of parameters would be too demanding performance-wise. Observers say it is unusual for Apple, which has tightly controlled its technology and ecosystem, to leverage infrastructure from Google and Nvidia.
Apple is highly likely to unveil key features of the new Siri at the Worldwide Developers Conference (WWDC) to be held at its Cupertino, Calif., headquarters on the 8th. Given that the industry has viewed Apple as lagging Samsung Electronics and Google in the AI phone race, attention is on whether this Siri overhaul will become a turning point for its "Apple Intelligence" strategy.