Why Apple is taking a small-model approach to generative AI

Apple's bespoke approach to generative AI prioritizes user experience and privacy, focusing on smaller models tailored to their operating systems.

: Apple's new Apple Intelligence emphasizes a smaller, pragmatic approach to generative AI, tailored for their OS. The models focus on seamless integration and user privacy. Apple also opens up to third-party AI for extended functionalities.

Apple Intelligence, introduced at WWDC 2024, focuses on a more pragmatic and bespoke approach to generative AI. Unlike other companies that take a 'bigger is better' approach, Apple's models are specifically designed for its operating systems, aiming to provide a seamless and frictionless user experience.

These smaller models are trained on customized datasets, which may improve transparency and reduce complexity. Apple also includes feedback mechanisms for users to report issues and relies on specialized 'adapters' to handle a variety of tasks. This focused approach makes Apple Intelligence practical and efficient, though it may sometimes require third-party AI for broader tasks.

Apple emphasizes stringent privacy standards, treating both on-device and cloud-based processes with equal security. The company has also introduced mechanisms for websites to opt-out of AI data training. This commitment to transparency and user privacy is a balancing act that Apple aims to maintain as its AI systems evolve.