What makes AI phones different? More details of Apple's AI features leaked out

In this round of AI competition, Apple is catching up by "speeding up". In last week's earnings call, Cook, who had previously been cautious, enthusiastically talked about artificial intelligence and said that "exciting" AI features and products will be launched soon.

What makes AI phones different? What are the advantages?

Although Apple has not yet revealed details to the outside world, plans to incorporate AI features into iOS 18, iPhone 16, etc. have gradually emerged.

According to the latest media reports and Bernstein's report:

iOS 18: One of the major focuses will be giving Siri the ability to summarize text, allowing it to analyze and summarize text messages in the Messages app, providing users with a more intelligent voice interaction experience. Safari browser will also introduce "smart search" functionality, allowing users to summarize the content of web pages they search for.

iPhone 16: Bernstein predicts that iPhone 16 will at least provide AI features similar to Google Pixel 8 and Samsung Galaxy 24, and may have better performance because Apple may use localized large models to achieve faster performance.

Google and Samsung's latest flagship models, Pixel 8 and Galaxy 24, both use Google's large model "Gemini", providing a range of Gemini-based AI features, including instant search, conversation summarization, smart translation, and more. In addition, they also have some AI features based on traditional machine learning, such as magic eraser, video enhancement, etc.

Importantly, Apple plans to run large models for AI functions on the device side, rather than adopting a hybrid cloud + local mode like Google and Samsung. With Apple's powerful neural network acceleration capabilities in its chips, as well as the company's expertise in model compression and localized deployment, iPhone 16's AI features are expected to demonstrate lower latency and faster response times than competitors.

Software with integrated AI features: Apple's AI features may be integrated into software such as Apple Music, Xcode, Pages, and Keynote in the future. For example, AI content creation in Pages and Keynote, as well as improved Spotlight search. Apple may also offer backend features for developers, such as integrating AI into XCode to create content. There are also rumors that Apple is internally testing an AI tool called "Ask for AppleCare" to assist technical support staff.

Details about the AI plan may need to wait until the Worldwide Developers Conference (WWDC) on June 10th to be revealed, and most AI features are likely to be released in September along with the new iPhoneAnd driving these new features may be Apple's self-developed large-scale model Ajax. Several Apple papers show that the company is implanting large models into chips to achieve on-device local operation, with Ajax reportedly generating simple responses through localized processing.

Ajax has 200 billion parameters, more powerful than GPT 3.5, but still behind leading models such as GPT 4 and Google's Gemini Ultra1. With rumors of GPT-5 set to be released this summer, the gap may further widen, prompting Apple to discuss the possibility of using external large models with potential partners such as Google, OpenAI, and Baidu.

In addition, according to reports from the media last week, Apple researchers have released four open-source "small models" on Hugging Face. Apple stated that OpenELM demonstrates extremely high computational efficiency in text-related tasks such as email writing.

The parameter magnitudes of the four models are 270 million, 450 million, 1.1 billion, and 3 billion. In comparison, Microsoft's latest phi-3 model starts at 3.8 billion parameters, while Google's Gemma only offers a 2 billion parameter version.

From a commercialization perspective, Bernstein's report points out three paths:

1. Apple can drive hardware sales through AI features on the iPhone 16, with even minor efficiency improvements potentially leading to significant revenue growth;

2. AI search could bring in revenue for Apple through increased advertising income, but this depends on whether AI search can generate additional revenue rather than cannibalizing traditional search revenue;

3. If AI drives the development of new applications, Apple may earn more revenue through the App Store;

Overall, Bernstein believes that Apple is actively promoting the integration of AI technology and products, which is expected to become a new profit growth point for the company. AI is gradually becoming an important part of its long-term growth story.