The emergence of AI models seems to have dropped a bomb on the mobile phone industry. Intelligent technology, which had previously stagnated, suddenly found a new breakthrough. Everything is suddenly moving towards us.
Many of the top technologists in the mobile phone industry are very excited: this is undoubtedly a technological revolution. For mobile phones, voice assistants, which previously seemed mediocre and less intelligent, can finally slowly move towards the role of "personal assistants", and more functions on the mobile side can also be supported by AI capabilities - although there is no obvious innovation in sight, but the future is promising.
In addition to the self-developed operating system based on the optimization of Android native applications, whether they can find a system development space suitable for the AI era outside the Android ecosystem, or whether they do not make an operating system and only serve as a middleware platform to support the interaction between the terminal and the system are the solutions that manufacturers are seeking.
At the hardware level, which determines the upper limit of software capabilities, image racing has also opened up new application space with the blessing of AI, and the folding screen market has begun to mature. Although the mobile phone industry has experienced two or three quarters of downturn in 2023, driven by these technologies, the mobile phone market is gradually moving towards a moderate recovery range. Where will the prelude to the technology race lead the mobile phone industry?
Doping: AIGC
Since its formal definition in 1956, the AI industry has gone through three phases and is now expected to move towards the AGI (Artificial General Intelligence) stage. As ChatGPT began to amaze everyone at the end of 2022, among the new machines released at the end of 2023, AI large model-related functions have come to the forefront.
AI large models have gradually moved from high-end models to mid-range price segments, and at present, manufacturers' actions are still relatively unified: text creation, sorting out material frameworks, one-step call services, passerby elimination, call assistants, etc. Of course, this is a phased exploration on the road to a more intelligent one.
Mr. Around, vice president of vivo and president of vivo AI Global Research Institute, believes that large models have greatly improved production efficiency and brought about the fourth major technological change. This is the beginning of a century of technological change.
"After decades of hesitation, this is the dawn of a great change in science and technology. At the end of 2022, the phenomenal ChatGPT-3.5 product was born, and people saw that artificial intelligence can finally conduct conversations and understand semantics and context like a teenager. He said that this means that large models can compress high-dimensional abstract human civilization into knowledge and information.
"AIGC is one of the technological innovations and even revolutions that I am most excited about since I made mobile phones. It will revolutionize the experience of mobile life. Liu Zuohu, senior vice president and chief product officer of OPPO, said in an interview that mobile phones will definitely become a super assistant for individuals in the future. "Because its knowledge goes beyond the knowledge of the individual, it can understand your language, observe you, learn Xi and understand your Xi habits like a human being, and give you the best help. "OPPO's end-side model has underlying cooperation with Qualcomm and MediaTek, and AIGC will give the mobile phone a new look in the future. It should be optimistic in the long term and cautious in the short term.
It is generally believed in the industry that large models with more than 50 billion parameters will have intelligent emergence, which is the premise of human thinking. However, in order to balance the different functional requirements of mobile phone use and future cloud collaboration, many manufacturers have released different specification matrices.
OPPO's self-trained Andean large model has three types of specifications: Tiny, Turbo, and Titan, which can cover models with a variety of different parameters from one billion to more than 100 billion, and can be intelligently scheduled according to different scenarios.
vivo's self-developed general large model matrix blue heart model also covers five large models at the level of billion, 10 billion, and 100 billion: the 1 billion level parameter model mainly covers the ability of localized text summary and summary, the 7 billion level model is a dual-purpose model for mobile phones, which can be used for semantic understanding and text creation, the 70 billion parameter model is the main model for cloud services, and the 130 billion and 175 billion blue heart models have logical reasoning and task orchestration capabilities.
Due to the different rhythms of new machines from different manufacturers, the high-end flagship X100 series released by vivo at the end of the year was the first to land the 7 billion parameter large language model on the terminal side, and the mid-range flagship S18 series released after that was also the first batch to be equipped with AI blue heart large models but did not specify the parameter specifications; OPPO released the mid-range flagship Reno 11 series at the end of the year, and its large model application is mainly reflected in image shooting and cutout capabilities. The Z60 Ultra series released by Nubia at the end of the year also mentioned the application of its 7 billion parameter Lingxi end-cloud fusion model, and the industry's first image AI model, which can also provide suggestions for shooting.
Although at present, it is not enough to regard the AI model as the replacement power at the current stage, it will have a significant impact on the mobile phone experience will be an inevitable trend.
Qi Yingnan, deputy director of research at Counterpoint, told the 21st Century Business Herald reporter that it is certain that in the long run, AI large models will be a revolutionary technology for mobile phones. According to the agency's forecast, the GenAI (generative AI) smartphone market share will account for 4% in 2023 and is expected to double in 2024, and by 2027, the GenAI smartphone market share will reach 40%, with shipments exceeding 500 million units.
"It is more significant that in terms of natural language interaction, it will be more natural and in line with human communication Xi in the future. While touchscreen interaction has become Xi, natural language is more efficient. He further analyzed that in the future, it is not excluded that the interaction mode of traditional APP may also change, and the service content can be directly called with the help of large models.
Qi Yingnan pointed out that in the OPPO Find N3 and vivo X100 series mobile phones, the voice assistant already has a strong ability to convert speech into text with a high accuracy rate, and the AI assistant can clearly understand the connotation of the instructions issued by the user. For example, if you give a specific command to the voice assistant to purchase a ticket, you can go directly to the purchase page. "This kind of advancement is not only the generative AI technology itself, but also the ability to be shared with the platform. ”
Stabilizer: Software Ecology
At present, the large model of mainstream mobile phone manufacturers is launched based on their self-developed operating system, so the adaptation progress of specific mobile phones depends on the adaptation speed of the whole machine manufacturer on the one hand, and on the other hand, it also depends on whether the hardware performance of the specific mobile phone is enough to support the operation of the large model.
Generally speaking, large models put forward a certain threshold for the AI computing power, memory NAND Flash and DRAM space of mobile phone SoCs, and usually the storage needs to start at 8GB, which will limit the popularity of AI large models on mobile phones to some extent, but it also depends on the manufacturer's compression of large model capabilities, such as the use of distillation, quantization and other technologies.
Of course, because the model continues to learn and Xi, there is still room for further improvement, "At present, there are some innovations in the use of functions, but it is hoped that the new functions in 2024 will redefine the mobile phone interactive experience, which is not yet reached." Qi Yingnan analyzed to reporters, in essence, it depends on the model parameters of the mobile phone itself and the adaptation of external software and hardware platform capabilities. "The parameters of the large model on the mobile phone side are relatively restricted, not as high as the PC, so some functions need to be realized through networking. It can bring differentiation in experience and competitiveness. ”
At present, mainstream mobile phone manufacturers mainly use Qualcomm and MediaTek, the two main chip platforms, both of which are in the exploration stage of specific large-scale model use cases, so the performance of each platform is similar in terms of experience. "The main constraint of the main chip platform is the upper limit of the mobile phone model capability, of course, if the manufacturer has a self-developed main chip, there will be a differentiated experience advantage. Qi Yingnan further analyzed that although Apple has not officially released the AI model, it will have more application advantages once it starts to do it, because it can decide to design use cases and optimize services according to its product cycle, without waiting for supplier support, so that the development efficiency is higher, the compatibility and energy efficiency ratio are also better, and it may even skip the exploration process of commercial products, and can be combined with a large-scale upgrade of the iOS system at one time.
This brings us to another layer of capability: the software ecosystem. Different from the independent ecosystem of iOS, the current ecosystem of the Android native system is very strong, and Google has requirements for the granularity of the reform of the Android system, so most mobile phone manufacturers adopt the route of self-developed operating systems based on Android, and cannot greatly innovate it.
Some vendors are also trying to think about exploring operating systems in the AI era. The self-developed operating system of Blue River created by vivo is written in Rust, which is not based on Android, nor is it compatible with the Android ecosystem, and has no intention of replacing the existing mobile phone operating system.
Zhong Xiaolei, a research analyst at Canalys, believes that Android manufacturers are working hard to build an ecosystem to compete with iOS. "Consumers are increasingly willing to switch camps for a better ecological experience, which has also led major brands to start developing self-developed operating systems and enhance their brand image by demonstrating their R&D capabilities. For example, the Huawei Mate 60 series provides a complete ecosystem experience through self-developed SoCs and HarmonyOS operating systems. Xiaomi Surging OS also improves the interconnection experience between multiple devices. The self-developed system is expected to open up multiple revenue growth curves for manufacturers other than smartphones, such as IoT and new energy vehicles. However, the key to the success of a self-developed system still depends on the ability to provide consumers with a unique and attractive enough experience. Therefore, manufacturers still need to continue to invest in collaborative research and development of software and hardware to defend their own user retention and profit levels in the face of future market changes. ”
Qi Yingnan pointed out that in the process of applying large models, the importance of security is more prominent, and at the same time, consumers do not want to accept more narrow information under the wave of digitalization, and it is also important to have the opportunity to make users feel more fair and diversified information push. "In general, in 2023, mobile phone manufacturers will release their large models one after another, and by 2024, they will be implemented in more mobile phone products, and it will take some time to expect disruptive innovation and changes. ”
Booster: Big hardware
Although the current era is called "software defines everything", hardware actually determines the upper limit of software capabilities, even if the complex hardware parameters are dizzying, but hardware stacking is obviously an important help in the mobile phone race.
As the most obvious and rigid demand for mobile phone experience, the improvement of imaging capabilities has been a topic that mobile phone manufacturers have been exploring over the years. From the early invocation of relevant performance in SoCs, to the targeted improvement of computing image performance by independent self-developed ISP/NPU chips, to the embracing of the era of large models, the road of mobile imaging has also gone up and down.
At the recent press conference, Ni Fei, chairman of Nubia Technology Co., Ltd., said: "Mobile phone innovation can bring about transformative experience improvement, one is the screen, the other is the image. We have continuously invested 5 generations to make under-screen (camera), just to bring a pure, ultimate full screen. ”
For example, the mid-range model vivo S18 Pro is equipped with the same 50 million Sony IMX920 bionic outsole sensor, the same 50 million ultra-wide-angle sensor, and the same professional telephoto portrait sensor as the X90s.
From the perspective of the overall technology trend, Qi Yingnan believes that the most obvious thing about mobile phones in 2023 is that the image capability has been greatly improved, and it is also very important to ensure the processing power in the video shooting process.
However, compared with the traditional straight-screen mobile phones that appear to be weak in innovation, the stable iteration of the folding screen matrix of a horizontal fold + a vertical fold by mainstream manufacturers has given birth to a growth momentum that is different from the silent market.
According to statistics from CINNO Research, a third-party organization, in the third quarter of 2023, the sales of folding screen mobile phones in the Chinese market will reach 1.98 million units, an increase of 175% year-on-year and 70% quarter-on-quarter, and it has been the twelfth consecutive quarter of positive year-on-year growth. In the future, with the further maturity of folding screen product technology, further reduction of cost, and more application adaptation, especially after the development of applications that can give full play to the unique advantages of folding screen hardware, folding screen products are expected to occupy more market share.
"I think that in 2023, the folding screen mobile phone category will stand up and become a mature product that is difficult to be picky. Qi Yingnan analyzed that in terms of reliability, holding experience, battery life, image, screen performance, etc., the flagship folding screen mobile phones released during the year have matured, and the price is basically stable, and the pace of product upgrades is also stable.
"There will no longer be a situation where there is a significant generation difference between the next generation and the previous generation of folding phones as before. The generation gap obviously means that the technology is still in the process of progressing, and I think the hardware side of the folding screen has completed the process of rapid iteration. After becoming mature and stable, you can have more energy to focus on the software ecosystem, such as improving the mobile office experience. Qi Yingnan continued.
Zhong Xiaolei said that brands are starting to enhance the competitiveness of high-end product lines, and focus on improving product hardware design and software UI. "Under this trend, we believe that the competition in the global folding screen market will continue to escalate. ”
Looking back, in addition to the software and hardware competition that highlights the technology and device stacking capabilities, price competition is also an important point of interest in the mobile phone industry in 2023. Some manufacturers are also emphasizing that they are willing to sacrifice hardware profit margins in order to seek greater competitive advantages, and under this condition, the revenue support of Internet services is more important.
However, as the supply and demand of mobile phones from the whole machine to the supply chain are balanced, the supply of components has started to rise in price, how to calm the new pressure on the supply chain, and how to find more suitable AI native applications in the era of AI large models, will become a new point of competition in the coming year.
For more information, please download the 21 Finance APP