How will the world's major mobile phone manufacturers solve the problem of AI?

Doing what users need and making effective and balanced progress in capabilities, experience, and cost is exactly where mobile phone manufacturers should play a role in the big model era.

Written by: Zhang Peng

Editors: Jesse, Jingyu

**Source: **Geek Park

On March 24, OnePlus released their new cost-effective flagship, OnePlus ACE 3V. It may not seem like a very important launch event, but it is one of the first mid-range smartphones to embrace AI, which has quite important symbolic significance.

At the press conference, they specially invited Zhou Hongyi, the founder, chairman and CEO of 360 Group, who has been gaining popularity recently, to share some future trends about AI and mobile phones. Zhou Hongyi shared 15 development trends of the AI industry, the most important of which is “In the next five years, mobile phones will be the best hardware carriers for AI. Mobile phone manufacturers that do not turn to AI will become the next ‘Nokia’.”

Indeed, as the first batch of AI mobile phones came onto the market, more and more people have come to recognize that “mobile phones are the best carrier for AI.” It has become an industry consensus that every mobile phone manufacturer must embrace AI.

“Embracing AI” is just a slogan. Today, any smartphone can be installed with various large-scale apps, including many products that can be directly accessed with a browser. However, in terms of specific practice, there are still many problems that need to be solved.

Currently, the two leaders in the global mobile phone field, Apple and Samsung, are firmly embracing AI. The same is true for OPPO, which was the first company in China to propose the concept of AI mobile phones and has long expressed its strategic determination to go all-in on AI. The strategic visions of the three companies coincide.

Apple and Samsung have taken two completely different routes in practice. Apple chose to develop its own full stack, while Samsung connected to Google’s Gemini model in a similar way to “shelling”.

For domestic mobile phone companies like OPPO, which are aiming to surpass Samsung and be on par with Apple, how to embrace AI in practice has become a new proposition. On the one hand, overseas models cannot be used, and the competition landscape of domestic large models upstream is not yet clear. On the other hand, the cost of full-stack self-development and training means a huge burden. As a pioneer, every attempt by OPPO is to open up new frontiers for the industry.

For OPPO, combining the new round of AI technology with mobile phones, implementing it into toC applications, and efficiently converting it into commercial value is a more complicated issue.

In fact, it is a good thing to face these problems first, because whoever can answer new questions well will win a new position in the industry.

01 AI and mobile phones need each other

Before discussing how mobile phone manufacturers can implement AI, there is a preliminary question that is worth discussing: “Mobile phones and AI, who needs whom?”

There are two opinions on this issue in the market today. One opinion is that in an era when the growth of the mobile phone market is slowing down or even stagnating, mobile phones need AI to break through the current market ceiling and gain new competitive advantages. The other opinion is that although the AI big model has shown amazing results, the conversion efficiency on the demand side is low. If a good entry scenario cannot be found, it may be difficult to implement on a large scale in the end. Therefore, mobile phones are more needed to assist AI in entering high-frequency scenarios.

Combining the two perspectives shows that AI and mobile phones actually have a “mutual need” relationship.

In recent years, the speed of iteration of mobile phone software and hardware has slowed down significantly. The hardware parameters and software design, which were previously the most competitive, are moving towards maturity and saturation. The changes in new flagship phones are getting smaller and smaller each year, and the user replacement cycle has also been greatly lengthened.

Especially this year, people in the industry know that the supply chain costs have risen significantly this year, which will inevitably severely squeeze the profit margins of mobile phone manufacturers. This is why all mobile phone manufacturers are paying more and more attention to the construction of service ecosystems, because services can increase user stickiness.

More importantly, it brings new imagination space for income growth.

In the AI field, the big model field, which was full of passion last year, has also been a bit anxious this year, and urgently needs a more frequent entry scenario. Most of the current big model applications exist in the form of web pages and apps, or provide API interfaces for developers. But for real ordinary users, it is still not very convenient to make full use of big models, and the forms are not rich enough. Moreover, big models that run completely in the cloud are also difficult to use users’ personalized data for deeper functional integration.

ChatGPT is the best example of this. Since its launch in November 2022 and its rapid explosive user growth, ChatGPT has suffered from user fatigue last summer, with traffic decreasing instead of increasing. Now, ChatGPT has about 180 million users. Considering that it acquired 100 million users in just 2 months, and this number has only increased by 80 million in 14 months, we can feel the huge friction it has encountered in growth.

ChatGPT | Source: Visual China

Similar problems actually plague all large model applications. First of all, the user acquisition cost is now extremely high. At present, the comprehensive user acquisition cost of some large model native applications in China, including the user download cost and the computing power consumption of serving retained users within a week, is likely to have reached 12-15 yuan per person, and this cost seems to be getting higher and higher. The magical effect of the large model in the early stage can easily attract users to try it out, but how to connect with specific needs to retain users and convert them into paid subscriptions to bring in revenue is a much more difficult and important problem.

Therefore, when we discuss the combination of mobile phones and AI, the first thing we need to make clear is that there is actually no competitive relationship between mobile phone manufacturers and large-scale model technology companies for “ecological niches”, and neither side can “choke the other’s neck” in terms of technology.

Both parties have their own business problems to face. Only under the premise of open cooperation, can they drive improvements in user experience and create market growth to achieve a win-win situation.

02 Diligence + Duty is required

Of the two leading companies in the industry, Samsung has chosen to directly access Google’s Gemini big model, with a relatively low level of participation in development, which can almost be understood as “shelling”. Apple, according to current rumors, is developing its own big model in full stack, although there are also rumors that Apple will cooperate with Baidu’s Wenxin Yiyan in China, but this is most likely to solve compliance issues.

As the technology industry becomes more and more convinced of the Scaling Law, the capabilities of large models will continue to rise in the future, even leading to AGI. This will be an increasingly desirable, but also extremely expensive goal.

Even if you have a strong desire to achieve AGI, you still need to consider whether the company has the ability to continue to invest almost indefinitely. Obviously, Samsung and Apple have different choices in this matter, and even Apple, the world’s number one, may not necessarily walk this path alone in the end.

In this era, mobile phone companies need to be diligent and responsible. The so-called responsibility is to weigh whether the profit margin of their own hardware can seek a top-level general-purpose large model of “full stack self-development” and be unique in the world; and they should not expect to subvert the original mobile Internet ecosystem through “smart assistants” just because the hardware is in the user’s position. Thinking too much will do more harm than good.

Diligence means that with the advancement of AI technology, we should actively consider different usage scenarios and different needs, and solve complex semantic and image problems in an effective way in pursuit of continuous improvement in efficiency and experience. We should even consider using locally deployed models with relatively small parameter scales on the terminal side to solve simple and immediate user instructions. The effective and balanced progress in capabilities, experience, and cost that these users need is exactly where mobile phone manufacturers should play a role.

I think OPPO at this stage is more in line with the definition of “hardworking + responsible”. For example, I have observed that it does not overemphasize “full stack self-developed” and does not have the idea of using mobile assistants to occupy a position. Instead, it is open to cooperation and specializes in solving user problems. In fact, mobile phone manufacturers such as OPPO already have long-term experience in AI development in mobile phone scenarios. From voice recognition to image processing, they already have a full understanding of user needs. Now the emergence of large model technology actually provides new tools to solve user problems. In such a battlefield, companies that focus on products and user experience will definitely find their own position and new goals.

As far as I know, OPPO started its AI layout related to large models as early as 2020. The Find X7 released at the beginning of this year, as the first generation of flagship phones with AI functions, also fired the first shot of “AI mobile phones”. During the Spring Festival, OPPO pushed updates to AI functions to many existing models and tens of millions of users. Up to now, OnePlus has completed AI function coverage on thousand-yuan phones. OPPO’s large model development process actually reflects the unique thinking of Chinese mobile phone manufacturers on this issue and their own methodology: committed to becoming a contributor and popularizer of AI mobile phones, so that AI can go from being tasted to being commonly used, and accelerate the arrival of the AI mobile phone era.

OPPO releases Find X7 mobile phone | Source: Visual China

In addition, let me tell you a point behind the scenes that many people have not yet seen. OPPO’s AI technology team completed its integration last year and is no longer a group of scattered functional teams. This has built a very necessary new architecture for the development of technical capabilities in the next AI 2.0 era.

An analysis shows that, on the one hand, OPPO invested in self-research, trained the Andes large model independently, and carried out three different scale classifications during deployment to pursue higher efficiency; on the other hand, OPPO did not excessively pursue large parameters and large computing power. In the call of super-large models, it reserved an open ecosystem and growth space for upstream large-model companies. It can be said that OPPO, as a “pioneer”, set an example for other manufacturers and took the lead in building a model for the AI capability architecture of mobile phones in the new era.

This development strategy has actually begun to translate into results. Since ColorOS 14 last year had built-in AI big model function, OPPO has achieved the deployment and installation of AI big model function on the most models. The function pushed during the Spring Festival is said to have received very good user feedback and increased usage.

03 Find a detour based on user needs

The future of AI-Native hardware is not limited to mobile phones.

There are many startups currently trying various new hardware attempts. For example, some teams are trying to make hardware with full voice interaction, or touch devices similar to “game consoles” but without third-party apps… It seems that the future of combining large models and hardware has unlimited room for imagination.

Today, when we say that “mobile phones are the best carriers of AI”, we are not saying that they are the only possibility for AI to be implemented on the C-end. It is just saying that the multimodal processing capabilities currently demonstrated by large AI models involve pictures, videos, voices, and texts, and the only hardware that can conveniently implement these multimodal input and outputs is smartphones.

On the model side, large model applications require both sufficiently strong networking capabilities and certain local training and computing capabilities, and the ability to achieve secure desensitization when privacy is involved. Smartphones are still the only ones that can achieve these two points.

On smartphones, big models have the broadest application space. Whether it is a voice assistant based on big models, combining the image generation capabilities of big models with cameras and photo albums, or performing in-machine data searches based on big models, etc., they can bring huge changes to the user experience.

OPPO mentioned in the AI mobile phone white paper released in February that their ultimate goal of embracing AI is to enable machines to actively understand users’ habits and needs, making user interactions intuitive. For example, taking photos and editing photos no longer require using multiple apps, and searching for information no longer requires crossing multiple platforms, making the process of using mobile phones simpler and more convenient.

Just like OPPO has equipped Find X7 with many AI big model capabilities, the most popular one is the AI removal function in the album. Behind this is OPPO’s thinking about user needs. In the past few years, mobile phone manufacturers have been rolling lenses, rolling large bottoms, and rolling algorithms to allow users to “take better photos”, but the real needs of most users are “simple and perfect recording of life”. Pure technical capabilities can be piled up, but the insight and cognition of user needs is OPPO’s opportunity to overtake on the curve.

OPPO’s AI erasure function in the Find X7 photo album | Source: OPPO official website

The features launched so far are just the first step in combining AI with mobile phones. As users use it more and more and become more proficient, more new AI needs will be discovered little by little.

In fact, this will bring some new opportunities to mobile phone manufacturers. As AI services penetrate into the mobile phone experience, consumers’ decision-making model for purchasing mobile phones will gradually shift from a “one-time consumption model” to a “service-based subscription model.” The development and sales of smartphones will no longer be based on a “model” cycle, but will become more and more coherent, and the competition between mobile phone manufacturers will become more and more like that of Internet companies.

Business problems will always return to user needs in the end. Looking back at history, every time a need is better met, it will bring an opportunity to “overtake on the curve”. OPPO has set a good example among domestic mobile phone companies. The big model technology wave just happened to bring an opportunity for leapfrogging user experience. This is also an opportunity for OPPO to surpass Samsung and Apple.

The battlefield of mobile phones has gradually become quiet after the end of the mobile Internet era. However, the advancement of large-scale model technology and the emergence of AI mobile phones, a new corner, will give us the opportunity to see exciting stories unfold again.