Apple Uses Google Gemini for On-Device AI Training

Apple has secured extensive access to Google's Gemini AI models, according to a report by The Information. This agreement grants Apple full rights to utilize Gemini within its own data centers. The core of this arrangement involves distillation, a process where Apple uses Gemini's sophisticated responses and reasoning chains as high-quality training data for smaller, custom-built models. In essence, Apple is openly employing a strategy that some Chinese AI companies have reportedly pursued covertly: using a powerful AI model to generate superior training material for more efficient models. With unrestricted access to Gemini, Apple can develop smaller, specialized versions that replicate Gemini's output and its underlying logic. These leaner models require significantly less computational power, making them suitable for direct deployment on Apple devices. While Gemini is engineered for chatbots and enterprise applications, its capabilities may not always align with Apple's specific vision for its virtual assistant, Siri. Apple is concurrently developing its proprietary models through its Apple Foundation Models team. The company is expected to unveil new AI features, potentially incorporating this on-device AI technology, at its upcoming Worldwide Developers Conference in June.