Home About us

There is no suspense in the battle for Apple's AI's domestic large model

Intelligent relativity 2024/07/03 21:38

Intelligent relativity, the battle for Apple's AI's domestic large model, there is no suspense

Text | Intelligent relativity

Author | Chen Bocheng

Apple has finally unveiled its latest AI progress.

A month ago, as previously expected, artificial intelligence was the focus of this year's WWDC conference. Of the 105-minute keynote speeches, more than 40 minutes were devoted to Apple's AI achievements.

Apple seems to have deliberately played a "homophonic meme" to redefine AI as "Apple Intelligence".

But in any case, the appearance of Apple's AI this time really answered a lot of doubts for the market.

Previously, there was news that Apple's mobile phone may have two versions of the domestic market and overseas markets due to the different AI models of the access system. What's more, there are people on the Internet who call the domestic version "castrated version".

Is that really the case? The answer may not be.

The basic disk of AI is firmly in the hands of Apple

Apple Intelligence (Apple Intelligence) is referred to as AI, which is not a "homonym" for Apple, but its understanding and design of AI functions has its own set of logic.

When introducing Apple Intelligence, Cook emphasized that what they wanted to build was "Personal Intelligence" that went beyond "Artificial Intelligence".

What do you mean? We can infer from Apple's conventional settings in the past that Apple's AI is likely to be highly synergistic with its own system and the applications and ecology within the system to fully meet the intelligent needs of Apple users. For example, deep integration with Apple services such as Siri, FaceTime, iMessage, Apple Pay, iCloud, etc., to provide a seamless cross-device AI experience.

To put it simply, what Apple AI wants to achieve is the advantages of system-level applications.

From the information released so far, we can see some application prototypes. Like AI images, Apple's AI expects not only to generate images with prompts, but also to create image content that is closely related to the user's personal life and interests based on the user's actual experience and personalized needs.

Similar to the "Photo Memories" function in Apple's mobile phones, Apple is trying to provide users with a brief AI video editing service, but the premise is to open up the user's photo library, so that the Apple system can repeatedly identify the photo content, in order to edit the video content of different themes.Apple Photo Memory feature

In order to achieve such or even better services and effects, Apple's AI must have system-level capabilities. As a result, the large model behind Apple's AI will most likely use its own solution.

Judging from the technical documentation of the large model released by Apple, it is indeed developing its own large model. According to reports, Apple's device will use a small model with 3 billion parameters, which is responsible for local computing and data processing, and at the same time ensures that users' personal habits and personal data are learned and trained, so as to provide more intimate and personalized services.

When it comes to the cloud, Apple has not yet announced the parameters of its self-developed cloud large model, but its performance is said to be comparable to that of GPT-4 Turbo. Moreover, the access to third-party large-scale model products such as ChatGPT is also a fact, but it is not the only one.

According to Craig Federighi, Apple's senior vice president of software engineering, Apple plans to let users choose their favorite large models in the future, including Google's Gemini.

In addition, Apple has released OpenELM, a set of small-form, open-source AI models designed to run on devices. These models range in parameters from 270 million to 3 billion and are designed to improve the processing power of AI tasks on mobile devices.

Obviously, Apple already has the ability to build an internal closed-loop AI system, and the basic disk of AI has always been firmly in Apple's hands. Third-party large-scale model products like ChatGPT are just a dose of seasoning.

The "castrated version" of apples is nothing

Judging from the current situation, neither OpenAI's ChatGPT nor Baidu's Wenxin Yiyan will affect the basic functions of Apple's AI. The positioning of these third-party large-scale model products is more like an App in the App Store for Apple.

It may be due to different regions and different permissions to use the app, but this does not define Apple's version features. Just as we can't say that the current Apple phone is a "castrated version" because it can't be on the "X" (i.e., Twitter), the same is true for third-party large-scale products.

Whether it is "X" or ChatGPT, if these products cannot be used in China, it is a problem of the product, not the "pot" of the channel. What's more, Apple's self-developed large model is quite comprehensive from the device side to the cloud, providing a relatively complete AI experience.

As far as the current thinking is concerned, the next thing that will affect the user experience more is the integration and application of Apple's small model on the end side. According to Apple's concept of prioritizing user experience, it emphasizes the customization of AI models, similar to the sparse activation mechanism of MoE models, and solves daily tasks in the process of user use faster through the mobilization and combination of a variety of small models.

As we all know, large models have larger parameter scales, higher computing resources, storage space, and higher network speeds, and if the hardware and network conditions cannot keep up, it is easy to have problems such as high latency and packet loss, which cannot bring a good AI experience. The small model has the characteristics of small size, low computing resource requirements, and high training efficiency, which is suitable for running on smartphones and other terminals.

Nowadays, the MoE architecture that has emerged in the field of large models also corresponds to Apple's thinking on the deployment model of smartphones from the device side to the cloud. The high efficiency and high adaptability of the small model are inevitable, while the large model, whether it is developed by Apple or the intervention of third-party manufacturers at home and abroad, only provides an advanced personalized experience and does not affect the main body.

Of course, through the experience of "intelligent relativity", it was found that when we really use large-scale model products, domestic large-scale models are actually not so stretched. On the contrary, due to local computing power, Chinese semantic understanding and high integration with the Chinese Internet, domestic large models have more advantages for local users.

On the one hand, the understanding of language and culture, the application of AI large models is closer to the form of dialogue. In the process of experience, "Intelligent Relativity" found that the user's personal expression is very casual and can be said to be not so standardized, so the domestic large model has a more obvious advantage in understanding the user's semantics, so as to respond more accurately to the user's needs.

On the other hand, there is the issue of obtaining data resources. AI models are constantly evolving, especially in the process of user use, they will continue to obtain new data for training and learning, and finally present a more ideal state. In this regard, the domestic large model is backed by a huge Chinese Internet and user group, which can maximize the acquisition of local data for learning and improvement, while OpenAI and Google are limited and difficult to access.

To put it simply, we should not only look at its performance in the laboratory, but also look at its application space and development prospects in the market and actual industry scenarios.

Therefore, it is not a bad thing for Apple AI to access local large models such as Baidu Wenxin Yiyan in China. For local users, based on the joint creation of Apple and Baidu, they will be able to obtain services and experiences that are more in line with local needs.

Perhaps, we don't need to stare at OpenAI, a product that is already difficult to experience, and entangle the choice of Apple's AI.

Write at the end

In general, there is no doubt that Apple's AI basic disk is supported by a self-developed large model.

In addition, the performance of the domestic large model is not bad in local applications, which can be waited and seen in the next time.

Apple's mission is not to introduce OpenAI's ChatGPT to China, but to create a more efficient, convenient, and powerful AI system based on its own system and ecology to provide advanced services for its own users.

Therefore, the controversy over the "castrated version" and the controversy over the large model is not so important. As Apple revealed, the access to third-party large model products is not the only one, and Apple will find a way to let users choose their favorite large models.

In the future, we look forward to Apple's promise to give users the right to choose.

*The pictures in this article are all from the Internet

This content is original to [Intelligent Relativity],

It only represents a personal opinion and is not allowed to be used by anyone in any way, including reprinting, excerpting, copying or mirroring without authorization.

Some of the pictures are from the Internet, and the copyright ownership has not been verified, not for commercial use, if there is any infringement, please contact us.

•New media for the AI industry;

•Monthly TOP5 of The Paper Technology List;

•The article has long "occupied" the TOP10 of the top 10 popular articles in titanium media;

•Author of "Artificial Intelligence: 100,000 Whys"

•[Key focus areas] Smart home appliances (including white and black appliances, smartphones, drones and other AIoT devices), intelligent driving, AI + medical, robots, Internet of Things, AI + finance, AI + education, AR/VR, cloud computing, developers, and the chips and algorithms behind them.

This article is from Xinzhi self-media and does not represent the views and positions of Business Xinzhi.If there is any suspicion of infringement, please contact the administrator of the Business News Platform.Contact: system@shangyexinzhi.com