What is Apple Intelligence? A detailed explanation of Apple's AI system.
Apple Intelligence is a personal intelligence system developed by Apple and integrated into iPhone, iPad, and Mac platforms. It was announced at WWDC in June 2024. This AI system combines powerful generative models...
In today's digital age,artificialintelligent(AIApple has become a key force driving technological progress and innovation. As one of the world's leading technology giants, Apple has continuously explored and innovated to provide more...intelligentAnd a personalized user experience. At WWDC 2024, Apple unveiled its...up to dateUnveiling the secrets of Apple Intelligence – a brand new personal…intelligentThe system aims to fundamentally change the way users interact with devices.
What is Apple Intelligence?
Apple Intelligence is a personal intelligence platform developed by Apple Inc. and integrated into the iPhone, iPad, and Mac platforms.intelligentThe system was announced at Apple's Worldwide Developers Conference (WWDC) on June 11, 2024 (Beijing time).AIThe system combinespowerfulApple Intelligence uses generative models and personal backgrounds to provide users with useful and relevant services and experiences. At the heart of Apple Intelligence lies its advanced... AI A focus on technology and privacy not only enables us to understand user needs but also to predict user intentions, providing more personalized services while protecting user privacy.
Apple Intelligence's core functionality
AIDriven writing tools
The writing tools are a system-level feature of Apple Intelligence that helps users rewrite, proofread, and summarize text across a variety of applications. Whether composing emails, polishing blog posts, or organizing class notes, these tools give users more confidence when writing. Users can adjust the tone of their text, check for grammatical errors, and even get concise summaries of their content.
Genmoji and Image Playground
Apple Intelligence also brings fun and creativity to users' images and emojis. With the new Image Playground, users can generate fun images in three styles: animated, illustrative, or sketched. Users can create custom emojis, known as "Genmojis," perfectly capturing moments and unique expressions. Furthermore, using the Image Wand, users can transform sketches into sophisticated images.
Improved photo and video experience
The Photos app has also been enhanced, adding a natural language search function that allows users to easily find specific moments in videos. A new cleanup tool can remove distractions from the background without altering the subject. With the "Memory" feature, users can simply...SimpleYou can create a custom story simply by providing a description.
Apple Intelligence AI The domain sets a new standard for privacy. First, it has small models that run entirely on the device to handle requests locally. When more complex tasks require cloud processing, they have a new private cloud computing feature, ensuring that user data is not retained or exposed.
Siri's comprehensive upgrade
Siri has received a complete upgrade with Apple Intelligence. Now, Siri understands users better, follows their train of thought, maintains context between requests, and even allows users to type instead of speaking. Siri can also perform hundreds of new actions within apps, such as fetching books from friends.recommendOr check your parents' flight status.
ChatGPTintegrated
Apple has also integrated Open on its platform.AIChat under its umbrellaGPTSiri can utilize Chat with the user's permission.GPT Use your expertise to answer questions.GPT It will provide assistance within the system-level writing tools, generating content and images. ChatGPT Users can also link their accounts to access paid services.
Technical details of Apple Intelligence
Apple Intelligence is built upon a suite of generative models created by Apple, including on-device and server-based foundational models, diffusion models for image generation, and encoding models. Furthermore, Apple Intelligence can invoke third-party models, such as Chat, as needed.GPTTo handle more complex requests.
Device-side model
The on-device model boasts approximately 3 billion parameters and a 49K vocabulary, employing low-bit quantization and grouped query attention techniques to enhance speed and efficiency. On the iPhone 15 Pro, the model achieves a latency of 0.6 milliseconds from each hint token to the generation of the first token and a generation rate of 30 tokens per second.
Server-side model
The server-side model boasts a 100K vocabulary, enabling it to handle more complex tasks using private cloud computing while ensuring privacy and security. The model employs advanced techniques such as speculative decoding and context pruning to improve performance. It is built on a reinforced subset of the iOS foundation, through...powerfulThe encrypted and secure startup process ensures the privacy of user data.
Apple's models are trained on carefully curated datasets that do not contain any personal user data. Training data includes licensed data, publicly available data collected by AppleBot, and synthetic data. After training, Apple uses novel algorithms such as rejection sampling fine-tuning and reinforcement learning based on human feedback to improve the model's ability to follow instructions.
Apple has implemented a range of cutting-edge technologies to ensure optimal performance and efficiency on mobile devices. These include the use of grouped query attention, shared embedded tables, low-level palletization, and...High efficiencyThrough methods such as key-value caching updates, Apple has successfully created highly compressed models that maintain quality while meeting the memory, power, and performance limitations of mobile devices.
Low-rank adaptation (LoRA)
With Google Gemini Unlike general-purpose models such as Nano and Microsoft's Phi, Apple's model is fine-tuned for everyday activities like summarizing, replying to emails, and proofreading using a technique called "low-rank adaptation." This technique involves adapting small...Neural NetworksModules are inserted into various layers of the pre-trained model. This allows the model to adapt to different tasks while retaining its general knowledge. Importantly, these adapters can be dynamically loaded and swapped, allowing the base model to specialize for the task at hand.
How to use Apple Intelligence
Apple Intelligence for UsersfreeAvailable in beta, this version will launch this fall as a built-in feature of iOS 18, iPadOS 18, and macOS Sequoia, and will only support English (United States). Additional features, software platforms, and other language support will be rolled out next year. Apple Intelligence will only support iPhone 15 Pro, iPhone 15 Pro Max, and iPads and Macs with the M1 or later chip, requiring Siri and the device's language to be set to English (United States).
- Apple Intelligence Official Website
- Press Release – Apple Intelligence available on iPhone, iPad, and Mac
- appleintelligentThe underlying model has been revealed: the 3B model outperforms the Gemma-7B, and the server model is comparable.GPT-3.5-Turbo