Apple will be able to distill the Google Gemini models in order to train smaller models for its devices

Apple has full access to Google’s artificial intelligence (AI) models Gemini within the technology company’s own data centers, allowing it to use these models to train more efficient and smaller versions optimized for your devices, using the distillation.

Within the framework of the multi-year agreement that both companies announced in January of this year, it is expected that the iPhone maker will base its Foundational Models on Gemini models and Google cloud technologyin order to have the necessary technology to boost its Apple Intelligence functions and its renewed version of Siri.

At the time of the announcement, those from Cupertino did not offer many details about what this collaboration would be like, although close sources indicated they could independently modify Gemini models to adjust them to your needs and offer a user experience in which the Google brand does not appear.

Now, it has been learned that the agreement goes further and allows Apple has full access to Gemini models through propsios Google data centers. In this way, the firm can directly experiment with the technology giant’s AI to train its own Foundational Models.

As a result, Apple will be able to use the more complex Gemini models to train smaller, more efficient AI modelsoptimized specifically for your devices, using a distillation processas The Information has learned.

This is, it is a ‘machine learning’ practice that allows learning to be transmitted from one AI model to another. Thus, it makes it possible for a smaller model (student model) to imitate the behavior of larger neural networks (teacher model), learning their predictions and using their data, but without having to carry out great computing power.

That is, Apple will be able to use Gemini as a teacher model for his own smaller models and focus its use on specific devices such as the iPhone, iPad or Mac, requiring less computing power but guaranteeing optimal results.

This distillation technique would allow Apple to perform AI task processing seamlessly. local on the device itselfinstead of having to resort to the cloud. Thus, the company’s intention of keep Apple Intelligence and Siri working on devicesusing Private Cloud Compute.

All of this, with the aim of preserving the company’s characteristic privacy standards, using user data securely.

Likewise, the aforementioned media has pointed out that Apple’s intention is not to train small models to generate a larger one that competes among the main AI models in the sector.

STANDALONE APP FOR THE NEW SIRI

Part of Apple’s investment in Gemini AI is intended to boost its new version of your assistant Siriwhich will offer smarter capabilities and more useful services to users.

For example, will improve your answers providing concrete information about the applications instead of links to websites and will be able to offer more complete conversational responses, as well as run functions on the device such as sending a text message to a specific contact or setting reminders.

However, Apple is also testing a standalone app for Sirialong with a new ‘Ask Siri’ feature that will work with all of the company’s software options, as shared by Bloomberg analyst and journalist Mark Gurman.

The dedicated application for Siri will be available on iPhone, iPad and Mac later this year and will show an interface with previous conversations from the assistant, with a look similar to Apple’s Messages application, mixing message bubbles and a voice interaction mode.

The ‘Ask Siri’ function, for its part, will be based on a switch that will appear in the menus of integrated applications and that will allow users ask about on-screen contentstarting a new conversation with the assistant. For example, about certain texts on a web page or email queries.