谁会是苹果的Plan B?

Produced by | Tiger Sniff Technology Group

Author | Maruto Mountain

Edit | Miao Zhengqing

Header | Visual China

There is growing evidence that Apple is looking for a spare tire for Siri other than OpenAI.

At the recent WWDC, Craig Fedrighi, Apple's senior vice president of software, publicly stated that Apple is looking forward to cooperation plans with third-party models in the future, such as Google's Gemini.

This week, the Wall Street Journal published another article suggesting that Apple is in talks with Meta to integrate Llama into iOS.

But the question is, can Apple accept the terms of a "marriage" with the two companies?

"Apple's cooperation with OpenAI can make real money, and iOS users will also take a 30% commission (i.e., Apple tax) when subscribing to GPT Plus," Renee, an independent developer of iOS, told me that most open-source models still end up serving the core business, such as Llama for Meta Adavantage (Meta's AI ad optimization tool).

This is the contradiction, if Apple chooses to cooperate with Google and Meta, then it means that Apple will not be able to earn commissions in the open source model, and at the same time have to endure the latter two to help advertisers configure target audiences, and it is very likely to bypass the IDFA (Advertising Identifier) - Apple's core means of restricting advertisers in the process of invoking third-party large models.

While Apple has good reasons to reject Meta, the reality is that the company can't hang itself from a tree.

Is ChatGPT not good enough?

"People may prefer to use certain large language models, such as those that are good for writing or those that are used for programming, but we still want users to be free to choose." In a recent podcast abroad, Craig Fedrighi gave the above explanation of why Apple uses a third-party model.

However, Renee, who has more than ten years of experience in iOS development, expressed another view to the author, "Judging from the features announced by Apple Intelligence so far, Apple still focuses more on the application level than the creative level of AI. ”

And from the application level, OpenAI doesn't seem to be the best choice for Apple.

"When ChatGPT is combined with third-party plug-ins, it never makes separate matching and interfaces for it, even Microsoft's Copilot directly uses APIs, and these plug-ins describe their own needs for the model to call." Renee said that with OpenAI's simple and crude business model, there is a high probability that it will not be deeply optimized for Apple.

"That's one of the reasons why Apple uses the base model for built-in apps like Gallery, Notes, Mail, and more." Renee added.

In addition, compared with ChatGPT, the open-source Gemini and Llama are also more friendly to iOS developers, such as the latter, which has a fine-tuned environment for inference and training configured in the model. This is also more in line with Apple's current expectations for Apple Intelligence: with the help of the Apple Intents API, developers can define various actions in the app, and this process may only require a few simple lines of code.

In addition to the technical and application aspects, from the perspective of Apple's strategy itself, it is also necessary to find several more third-party large models.

On the one hand, Apple will continue its consistent supply chain management principles in the application of AI: to avoid any link to form a scale advantage, the most famous case is Apple's perennial support for LG Display, the meaning behind which is to control Samsung.

Again, if you look at OpenAI as Apple's technical support company for Apple Intelligence, it's not hard to understand Apple's decision.

On the other hand, the cooperation with Google and Meta can also avoid some legal and policy risks.

For example, in March this year, the European Union issued a sky-high fine worth 1.84 billion euros to Apple on the grounds of "suspected application distribution monopoly in the streaming media market". Choosing to call a variety of large models on iOS can avoid the accusation of industry monopoly to a certain extent.

To sum up, whether it is the introduction of Genmin, or the hand-in-hand Meta, or other third-party models, it is necessary for Apple, but for now, Llama's option may have been ruled out.

On June 26, Bloomberg reported that Apple's cooperation with Meta on artificial intelligence has not made a breakthrough, and it is reported that the latter has rejected Meta's cooperation proposal because "Apple believes that Meta's privacy protection measures are not strict enough." ”

However, this high-sounding reason does not seem to hold water, because Google, which is at odds with Apple, is not much better than Meta when it comes to privacy protection.

The difference between the two is that Meta can really touch Apple's life.

Bring in a third party and rest easy?

One fact that may be overlooked is that Meta is currently the only company among the traditional Internet giants that makes money with AI. According to the previously announced financial report, Meta's revenue in the first quarter of this year was $36.455 billion, a year-on-year increase of 27%; Net profit was US$12.369 billion, a year-on-year increase of 117%.

This impressive performance is partly due to the recovery of the Internet advertising industry, and partly to the success of Meta Advantage and AI Sandbox, which together form Meta's cutting-edge AI for Advertising feature, which successfully applies machine learning capabilities to the creation and delivery of ads, especially to extend ad targeting beyond the target group.

This is inseparable from the capture of massive user data and the analysis of user behavior habits. For Apple, it is unacceptable, either from the perspective of traffic distribution, or from the perspective of user privacy protection that it advertises.

Compared with the-for-tat Meta, Apple and Google have a symbiotic relationship. For example, Pichai, the CEO of Google's parent company Alphabet, disclosed in an antitrust lawsuit last year that Google paid Apple 36% of the revenue from running search ads through the Safari browser, and Google would pay Apple billions of dollars a year for this alone.

It is foreseeable that if Apple introduces Gemini in iOS in the future, there is a high probability that it will land around search-related applications.

However, although Apple has a positive attitude towards third-party large models, this will not be less troublesome.

Most typically, there is a problem with calls between models.

Renee, a veteran iOS independent developer, said, "The current trend of smart terminals is that the traditional GUI (graphical operation interface) is constantly being diluted, accompanied by the enrichment of interactive forms and the reduction of application operation levels, which require high-frequency access of different models. ”

"The future device-side model is likely to be the mutual call between several models, and according to Apple's tonality, the company may create a closed and deeply integrated ecological structure, which is a big challenge for Apple." Renee added.

In addition, it has to be mentioned that the problem of the scope of content available for third-party large models.

Due to well-known reasons, neither ChatGPT, nor Gemini, nor Llama can be used on the national version of Apple's terminals, and in the future, after Apple Intelligence enters China, it is likely to use the customized functions created by Baidu and the Meitu App like the Samsung S24 model.

Regardless of the difference in technical capabilities, only from the search source, the Gemini Nano model equipped with the overseas version of the S24 model has a search source of "all web pages that can be found on Google", compared to the domestic customized model, which is much more limited in terms of search source.

Renee believes that because of the above problems, Apple will eventually focus on the basic model developed by itself, and the third-party model is just the icing on the cake.

"At present, the cloud model has not yet been shown on the mobile phone, and the so-called killer applications in the industry in the past, even if they are as strong as GPT-4o, only exist in the form of edge applications. Renee added.

People who are changing and want to change the world are all in the Tiger Sniff APP