Apple has made waves in the field of artificial intelligence (AI) by integrating Gemini, a model developed by Google, to enhance its Siri assistant.
This collaboration was initially seen as a surrender on Apple’s part, known for its focus on privacy and processing data locally. However, new reports reveal that Apple has gone beyond simply using a Gemini API; it is implementing a technique called distillation to train smaller and optimized versions of the model.
Is Apple redefining AI with a localized version of Google’s Gemini?
Distillation is a method that allows smaller models to learn from larger models by solving tasks and explaining their reasoning step by step. This provides Apple with a model that mimics the functionality of Gemini but can run locally on its devices, maintaining the speed and privacy that characterize the brand.
Apple has full access to Gemini in its own data centers, allowing them to explore the model in depth and adapt it to Siri’s specific needs. However, this adaptation has presented challenges, as Gemini was primarily designed as a chatbot, not for the various tasks involved in the operation of Siri, such as managing reminders or interpreting emails.
As the WWDC on June 8 approaches, Apple is expected to present significant advancements in AI, including a new version of Siri for iOS 27, which could feature conversation memory and proactive functions. Rumors suggest that if Apple manages to implement these improvements, it could redefine its position in the AI landscape, challenging the narrative of having ceded to Google.
Therefore, the question is not just whether Apple is using Gemini, but how it is leveraging it to enhance the user experience while upholding its commitment to privacy and local processing efficiency.