Preface:

In today's era of large models and AIGC, as a traditional dominant player in the smartphone market, how Apple maintains its market leadership has become a key issue urgently needing to be addressed.

However, on some level, Apple's high emphasis on its own ecosystem moat has led it to choose to open cooperation with OpenAI, which undoubtedly has put it in a relatively passive position.

Author | Fang Wensan

Image Source | Network

Apple Opens the AI Blind Box, Bringing in OpenAI

At the recently concluded Apple Worldwide Developers Conference (WWDC24) in 2024, Apple officially announced a wave of highlights including iPhone call recording, iOS 18, macOS 15 Sequoia, and the release dates of Vision Pro in China and other markets, but the main attraction was still Apple's AI capabilities.

Apple's AI "big package" mainly includes two parts:

① Apple reached a cooperation with OpenAI, integrating GPT-4o capabilities, allowing users to freely access ChatGPT without needing to link accounts;

② Introducing a set of AI capabilities based on personal contexts - Apple Intelligence.

According to Apple CEO Tim Cook's expectations, future Apple products will be upgraded to personal intelligence, and Siri empowered by GPT-4o will become an AI agent assisting users in taking action.

However, it is worth noting that Apple's current AI capabilities mainly rely on close cooperation with OpenAI to achieve.

Given the relative closed nature of Apple's ecosystem, this strategy requires Apple to maintain the uniqueness of the iOS ecosystem while actively collaborating with other platforms to launch innovative applications.

Industry developers have clearly pointed out that no other smartphone company has, to such an extensive extent and with such fine granularity, successfully embedded AI large models into their products as Apple.

For OpenAI, teaming up with Apple, a heavyweight partner after Microsoft, undoubtedly adds the most solid cornerstone to its layout in the large model field.

Currently, Apple has not released a self-developed large model representing its AI hard power. Apple's definition of AI is more reflected in its views on AI+terminals and the application of AI;

Regarding the innovation of AI underlying technologies and algorithms, Apple has not disclosed any information, seemingly relying more on players like OpenAI.

However, it is worth noting that Apple's current cooperation with OpenAI may not mean that it is willing to bind itself to OpenAI for a long time, but may be a strategy to buy more time for its self-developed process.

Therefore, to some extent, the cooperation between OpenAI and Apple may be limited to the short to medium term.

Ahead of the WWDC event, Apple's market capitalization achieved significant growth, successfully breaking through the $3 trillion mark. However, after the event, the capital market's reaction to this matter was relatively subdued.

As of the close of trading on June 10, Eastern Time, Apple's stock price closed at $193.12, down about 2% from the previous trading day, with a market capitalization evaporation of up to $58 billion.

Apple Intelligence Launched, Apple Wants Not Much

At the WWDC2024 conference, Apple Intelligence, as the core AI function of Apple, is mainly reflected in the following aspects:

① It possesses capabilities such as image generation, video generation, and erasure. Users can easily create visual content such as animations, illustrations, and sketches through the Image Playground platform and widely apply them in applications such as Messages, Keynote, Freeform, and Pages.

② With the Genmoji feature, users can independently design and generate personalized emojis;

while the Clean Up tool can intelligently recognize and remove unwanted elements in the selected content.

At the same time, the introduction of search scenario functions enables users to retrieve specific photos more conveniently.

③ Apple Intelligence also has functions such as text generation, video generation, and search Q&A. On iPhone, users can rewrite, proofread, and summarize text through applications such as Mail, Notes, Pages.

Priority Messages and Smart Reply features help identify issues in emails and provide intelligent responses.

The Memories feature allows users to create short videos based on themes or storylines.

It is worth noting that Apple is not pursuing a large model with comprehensive coverage of functions but is committed to building a smart assistant system spanning its own data, software, and hardware.

In terms of user interface design, Apple hopes to present more small windows similar to widgets on the iPhone screen, precisely displaying the information users need most through AI technology, avoiding the screen being occupied by too many application buttons, and reducing user operation costs.

In short, Apple hopes that its smart assistant system can become a "portal" on users' phones, directly connecting users to the desired destination through AI technology, rather than just serving as a "room key" to enter different functional modules.

In summarizing AI characteristics, Apple emphasized five keywords: powerful performance, intuitive ease of use, functional integration, personalization, and privacy and security.

These characteristics together constitute the core advantages of Apple Intelligence, enabling it to demonstrate excellent capabilities in text processing, image processing, and human-computer interaction.

More Emphasis on Customization Compared to Common Large Models

In terms of application scope, many well-known manufacturers in the industry adopt a "scale-first" strategy when developing AI models, aiming to build their AI systems into comprehensive platforms that can cover global information.

However, Apple has chosen a more pragmatic technical path. Apple Intelligence, as a generative AI method, is characterized by a high degree of customization and is closely built around Apple's various operating systems.

But from another perspective, this also reflects Apple's vision of seamlessly integrating generative AI into its operating systems.

Even if users have no knowledge of the underlying technologies driving these systems, it will not affect their enjoyment of the smooth experience brought by Apple products.

In this process, the most crucial point is to maintain the lightweight nature of the model.

Specifically, Apple Intelligence is trained only on customized datasets designed for the types of functions required by users of its operating systems, ensuring moderate model size.

In short, Apple Intelligence is a hybrid cloud-end intelligent system.

After the latest iPhone, iPad, and Mac product upgrades, two end-side models with relatively small parameter amounts will be built-in.

When users have AI needs or Apple Intelligence detects the need to perform related tasks, the system will prioritize using the local end-side models for processing, such as simple tasks like text-to-image generation.

Relying on OpenAI to Revive Siri is Not the Ultimate Solution

Integrating GPT technology into Siri significantly enhances its intelligence level while also facing the challenge of how to reasonably use personal data while ensuring user privacy and security.

To address this problem, Apple has proposed and implemented two effective solutions.

① Relying on the powerful NPU computing capabilities of the M-series chips and A17 Pro chips, most AI functions can be realized on the end-side, thereby ensuring the privacy and security of user data.

② When users need to utilize the powerful capabilities of cloud-based large models, these models will run on servers dedicated to Apple Silicon, further protecting user privacy from infringement.

From Apple's perspective, invoking large models through cloud-end collaboration can not only protect user privacy but also maximize the AI performance of Apple products.

It should be noted that although the iOS version of the ChatGPT app update has supported Siri and Shortcuts on iPhone 15 Pro and above, due to restrictions on ChatGPT services in China, domestic users cannot enjoy a complete service experience for the time being.

Apple's cooperation with OpenAI can be described as deep to the "granularity level".

However, for Apple, the impact of this upgrade is still unpredictable, bringing both opportunities and challenges.

The most decisive move that embodies Apple's determination is undoubtedly the introduction of GPT-4o into Siri, significantly enhancing its intelligence level.

Relying on Apple Intelligence capabilities, Siri now possesses richer semantic understanding and large model context analysis capabilities, enabling it to accurately capture the specific references of pronouns such as "that time," "then," and "there" in users' expressions.

In addition to the original voice interaction function, Siri will also expand new capabilities such as text interaction and cross-app execution.

Nowadays, users can communicate with Siri through text or voice, and Siri can accurately understand the subtle differences and hesitation in users' expressions.

When faced with complex questions, such as when a user inquires about the preparation method of a complex dish, Siri will intelligently judge whether to invoke ChatGPT for a more detailed answer.

Moreover, with the powerful capabilities of GPT-4o, users can also interact with Siri in multimodal ways such as documents and PDFs, and Siri can provide precise responses based on applications and database information within the iPhone.

In terms of cloud-based large models, Apple has access to OpenAI's GPT-4o. During text processing and Siri usage, if users wish to invoke more powerful cloud-based models, they can switch to GPT-4o to obtain richer information.

However, there is currently no clear information about which cloud-based model will be used for the Chinese version.

Phones with AI Ultimately Serve Sales

With changes in the large model market and the convergence of smartphone functions, the impact of AI on smartphone increments is questionable.

On the one hand, domestic consumers have limited understanding of large models and have low actual usage demands.

In the terminal market, sales still emphasize phone configuration and smoothness.

The new energy vehicle market is similar, with customers paying little attention to AI. More than half of global respondents are unfamiliar with generative AI tools.

On the other hand, scenarios such as text-to-text, image-to-image, and search Q&A are widely used and perceived significantly.

Smartphone manufacturers profit from large models in areas such as imaging and photo editing, and the market is not lacking in phenomenal image-based large models.

However, users still worry about privacy protection in AI phones.

In addition, large model vendors maintain competitiveness through price wars, gradually concentrating market share and forming a Matthew effect.

General large model vendors reduce prices and improve technology, making value maximizers and pragmatists unwilling to replace their devices.

Currently, it is difficult to determine Apple's position in the AI field, but it has launched some excellent applications within the iOS system, narrowing the gap with the Android camp.

In addition to user privacy protection, the launch of Apple Intelligence also faces a challenge: not all devices support its AI functions.

To ensure smooth operation of AI functions, devices need to have sufficient processors and memory. Taking the Android market as an example, running AI large models requires at least 12GB of memory.

When deploying the Gemini Nano model, Google excluded the Pixel 8 because its 8GB of memory may be insufficient.

While not completely denying its possibility, it noted that enabling this feature may affect the operation of other applications.

Although Apple devices are durable, 8GB of memory still poses a limitation for LLM functions.

Apple's update strategy is limited, and users have a narrow range of choices. Old flagship models may not meet AI demands, while new models are expensive. Apple needs to focus on how to make users accept and promote a device replacement wave.

Facing privacy regulation and market competition, investors have a cold attitude towards AI technology.

Conclusion:

Goldman Sachs analysis pointed out that Apple's introduction of intelligence and generative AI at the conference marks an important milestone in its AI strategy, laying a solid foundation for the future integration of large language models (LLMs).

However, objectively speaking, in the innovation race in the AI phone field, Apple has not yet formed a clear leading position compared to its competitors, and we have not observed any unique and sustainable competitive advantages.

Some reference materials: Top Headlines: "Why Did Apple Choose Small Models on the Path of Generative AI?", APPSO: "Apple Intelligence Plays a Word Game", Tencent News DeepWeb: "Apple Cannot Rely Solely on OpenAI to Revive Siri", All-day Technology: "Apple AI's First Shot Missed", Intelligence Emergence: "Apple's AI is Not Competing with AI Companies", DoNews: "Apple AI, Is That It?", Finance and Economics Magazine: "Apple AI, No Surprises".