Apple made a significant leap in the field of artificial intelligence (AI) by introducing its personal intelligence system, “Apple Intelligence,” at the 2024 Worldwide Developers Conference (WWDC 2024). Integrated into iOS 18, iPadOS 18, and macOS Sequoia, this system aims to provide users with more efficiency and effectiveness in their daily tasks. Apple Intelligence includes two powerful language models: on-device and server-based. So, what’s behind this AI? How does Apple Intelligence work?
Apple Intelligence and core models
Apple Intelligence consists of two main components: the on-device language model and the server-based language model. The on-device model has approximately 3 billion parameters, while the server-based model operates on Apple’s cloud infrastructure. These models perform tasks such as writing text, prioritizing notifications, creating fun images in chats, and simplifying in-app actions.
Responsible AI development
Apple adheres to four main principles of responsibility in its AI development processes:
- Empowering users: AI tools are designed to make users’ lives easier.
- Representing users: AI models work by considering users’ needs and preferences.
- Careful design: AI tools are designed carefully and ethically.
- Protecting privacy: User data is protected and not used in the AI development process.
Apple unveiled its new "Apple Intelligence" initiative, representing a major push by the tech giant into artificial intelligence.
Adapters integrate into various layers of the pre-trained model to fine-tune it for specific tasks. Adapter layers’ parameters are represented in 16-bit and require about 10 megabytes. These adapter models can be dynamically loaded and changed, allowing the core model to adapt instantly to the task.
Performance and evaluation
Apple’s models have been preferred by human evaluators and have been comprehensively evaluated in terms of overall capabilities for both on-device and server-based models. In these evaluations, Apple’s models have shown superior performance compared to open-source and commercial competitor models.
Future perspectives
Apple aims to continuously improve and optimize the performance of its AI models. User feedback is highly valued in this process, and the capabilities of the models are continually enhanced. Apple’s core models and adapters make AI more effective and useful in daily life.
Apple’s core models and adapters form the foundation of Apple Intelligence, providing powerful capabilities to iPhone, iPad, and Mac devices. These models are developed to help users perform their daily tasks more efficiently and effectively. Designed meticulously at every stage with Apple’s responsible development principles, these models open the doors to a new era in artificial intelligence.
