tech
Apple working to cram massive Gemini model into iPhone to power new Siri
As Apple tries to shrink Gemini for the iPhone, a cloud component is probably inevitable.
TL;DR
- Apple's AI-enhanced Siri, powered by Google's Gemini, will operate using both on-device and cloud processing.
- The move to cloud processing appears to contradict Apple's previous preference for local AI and its focus on privacy.
- Smartphones, including iPhones, have limitations in processing large AI models due to smaller AI models and insufficient RAM.
- Large cloud-based AI models, like Google's Gemini, have trillions of parameters, vastly exceeding what can run locally on a phone.
- Google's Gemini Nano is designed for specific mobile features, not for a conversational assistant like Siri, which requires different models.
- Apple is reportedly using AI model distillation to shrink Google's Gemini models for potential local processing.
- Apple has faced challenges in running large, undistilled Gemini models on its own Private Cloud Compute infrastructure.
- Complex Siri tasks may be routed to Google's cloud, potentially using Nvidia's Confidential Computing platform for encrypted processing.
- Nvidia's Confidential Computing encrypts data during processing, which may help Apple maintain a perception of user privacy.