活久见！苹果和“死对头”合作，国行用户只剩羡慕

Core Viewpoint - Apple and Google have announced a partnership where Google’s Gemini will serve as the "core brain" for Apple's AI, significantly enhancing the AI capabilities of iPhone, iPad, and Mac devices [2]. Group 1: Partnership Details - Apple will integrate Google Gemini into its AI systems, leading to a more personalized Siri expected to launch later this year [2]. - The new Siri will utilize advanced reasoning and personalized services, allowing it to perform complex tasks and understand user intent better than the current version [2][8]. - Apple has previously collaborated with OpenAI, but the new partnership with Google positions Gemini as a primary AI model, while OpenAI's ChatGPT will function more as a supplementary tool [5][3]. Group 2: Technical Aspects - Google Gemini 2.5 Pro, with 1.2 trillion parameters, was chosen over other models due to its cost-effectiveness and engineering advantages [7][6]. - The Gemini model employs a "mixture of experts" architecture, activating only a portion of its parameters for each query, which helps manage computational costs and energy consumption [9]. - Apple will pay Google $1 billion annually for the underlying technology licensing, which is more favorable compared to the estimated $1.5 billion annual cost of using Anthropic's Claude model [6]. Group 3: Privacy and Control - User privacy concerns are addressed as the Gemini model will operate on Apple’s private cloud servers, ensuring that user data remains under Apple's control [11][12]. - The partnership does not imply that Apple will abandon its in-house development; rather, it will use Gemini to enhance its own AI models until its trillion-parameter model is ready by 2027 [13]. Group 4: Market Implications - The collaboration with Google is seen as a strategic move to fill a gap in Apple's AI capabilities while maintaining control over its AI ecosystem [13]. - Domestic users may experience different AI functionalities, as Apple is likely to partner with local firms or use tailored models to comply with regulatory requirements [14].