In the first year of AI mobile phones, who can inspire more creativity will ignite more terminals

TechWeb

2024-05-15 16:28Posted on Hubei TechWeb official account

In mid-2023, Samsung officially proposed the concept of "AI large model mobile phone" and vigorously built up momentum in marketing. Since then, the phrase "the first year of AI mobile phones" has begun to be widely mentioned by mobile phone manufacturers. In fact, the first person to invent this term was Qualcomm President and CEO Cristiano Amon, who once said: 2024 will be the first year of global AI mobile phones, and generative AI is "crazy" into mobile phones.

The explosion of this concept has caused a number of AI mobile phone hardware concept stocks to soar, among which TD Optics and Furong Technology have risen by 10.03%. The skyrocketing hardware of AI mobile phones also means that the competition of mobile phone performance has entered the stage of volume AI technology from volume processor, volume screen, volume image, and volume memory. Of course, this also means that the competition for AI large model technology is intensifying.

1. Why is the "first year of AI mobile phones" so significant?

Since Qualcomm President Amon proposed that 2024 is the first year of global smartphones, many domestic mobile phone manufacturers have expressed their agreement. Among them, OPPO is particularly active, and its chief product officer Liu Zuohu said at the press conference that the company has established an AI center. Chen Mingyong, founder of OPPO, said in an internal letter on the first day of the new year: "2024 is the first year of AI mobile phones. He believes that in the next five years, the impact of AI on the mobile phone industry can be compared to the replacement of feature phones by smartphones.

As mentioned earlier, mobile phone manufacturers have competed in memory, screen, and photography in order to seize the market. However, in the case of slow iteration of hardware technology, it is more difficult for mobile phone manufacturers to occupy the minds of consumers through hardware upgrades. The advent of the wave of large models is a time when mobile phone manufacturers lack new stories and weak performance growth.

As a result, no mobile phone manufacturer wants to miss out on this new opportunity for AI. At the same time, the popularity of AI large models has also convinced mobile phone manufacturers that epoch-making innovation at the software level may come faster than at the hardware level.

It is foreseeable that "reshaping mobile phones with AI technology" will become a new competitive gateway for mobile phone manufacturers.

From the perspective of mobile phone manufacturers, the concept of "the first year of AI mobile phones" has at least three meanings:

First, the "first year of AI mobile phones" triggers the mobile phone upgrade cycle, and the market potential is huge.

In 2010, the mobile phone penetration rate in Chinese mainland was about 60%, and the smartphone penetration rate was about 10%, while the domestic mobile phone stock was 120 million that year. Four years later, both numbers are approaching 800 million. This shows that the popularity of mobile phones is happening at the same time as the arrival of the smart era. By analogy, after entering a new cycle of AI technology development, the popularization of AI mobile phones and AI large model technology may also occur at the same time.

In other words, the iterative innovation of AI large model technology will bring huge development space to the AI mobile phone market.

IDC predicts that by 2026, nearly 50% of terminal device processors in the Chinese market will have AI technology. At the same time, the domestic market will usher in a new round of replacement. According to IDC's inference, the shipment of next-generation AI mobile phones in the Chinese market will be 37 million units in 2024 and will reach 150 million units in 2027, with a market share of more than 50%.

Second, the concept of "the first year of AI mobile phones" is also a starting gun, and each company has entered a technological arms race.

With such huge growth potential, most mobile phone manufacturers are shouting "All in AI".

In terms of overseas markets, Samsung released a new generation of flagship smartphones Galaxy S24 series in California in January this year, touting as the first AI phone.

In terms of the domestic market, as early as October last year, when the general trend of AI mobile phones was emerging, Xiaomi, OPPO, vivo, etc. successively announced new systems. For example, Xiaomi has connected the smart assistant Xiao Ai to the self-developed large model MiLM, OPPO has created a new Xiaobu assistant based on the self-trained Andean large model, and vivo has released a self-developed general large model matrix.

To this day, mobile phone manufacturers continue to show their "AI muscles", such as Honor's Magic series and vivo's X100 series, all of which have shown their own characteristics in the application of AI technology. And their goal is quite the same: to become the head of the AI mobile phone industry, so that users think of themselves when they mention AI mobile phones.

Third, the opening of the "first year of AI mobile phones" means that the market has raced against time.

Samsung and Huawei are the two companies that are the earliest and fastest responding, and Samsung's first AI mobile phone sold more than one million units within 28 days of its launch, creating a record of the fastest transaction of one million units. According to incomplete statistics, the sales of the Huawei Mate 60 series have also exceeded one million, and experts expect the final sales to be at least 12 million units. AI phones have caused a huge shock to the market, and the new race has just begun.

2. AI mobile phones are fiercely competitive, why does Volcano Engine actively participate?

At a time when mobile phone manufacturers are competing fiercely, Volcano Engine has also joined the competition and launched the smart terminal large model alliance, introducing the first batch of alliance members including OPPO, vivo, Honor, Xiaomi, Samsung, and Asus.

Some people may ask, "Volcano Engine is not a mobile phone manufacturer, why should it actively participate?" ”

This starts with the essence of AI phone competition:

On the one hand, everyone is competing for the mental high ground of AI terminals. In this regard, various mobile phone manufacturers have developed different strategies around AI.

For example, Samsung has seized the minds of users with "in-depth insight into customer needs and improved user experience". Traditionally, smartphones have been limited to intelligent operating systems and large-screen displays, but Galaxy AI has played a new card – making phones more user-friendly. For example, with Galaxy AI, the big data model on the mobile phone side can realize localized information processing, and important information does not need to be uploaded to the cloud, which greatly improves the security of user data. For specific tasks, Galaxy AI can also call on cloud data to focus on outstanding problems. Powered by Galaxy AI, devices like the Samsung Galaxy S24 Ultra elevate the ultra-clear shooting experience, with the Galaxy S24 Ultra's "Quad Telephoto System" delivering crisp optical quality zoom at four different focal lengths: 2x, 3x, 5x, and 10x. It can be seen that Samsung Galaxy AI is taking the route of boutique service.

In contrast, ZTE chooses to reassure users in the two dimensions of "price and product". Compared with the Samsung S24 series, which is priced at a whopping 8,899 yuan, ZTE chooses to produce mobile phones that consumers can afford. Based on this concept, when competitors were still competing for the minds of AI mobile phone users at the price of 1999 yuan, Nubia took the lead in delegating AI functions to 100 yuan machines. In addition, in terms of AI product strategy, ZTE has chosen to gradually devolve from flagship models to mid-range and entry-level products.

On the other hand, vendors are competing for the diversification of AI applications.

At present, the application scenarios of AI on mobile phones are becoming more and more abundant, such as voice assistants, intelligent translation, health management, weather forecasting, etc., which are all key scenarios followed by mobile phone manufacturers, but there is still a long way to go before it can be successfully implemented.

The reason is that mobile phone manufacturers have rich industry experience, but lack professional large-scale model technical capabilities, and the volcano engine can complement mobile phone manufacturers.

Taking Samsung's Bixby voice assistant as an example, its goal is to make it more natural and convenient for users to interact with Samsung's devices and services, and provide personalized and intelligent services, but due to the effectiveness of the large model, Bixby cannot provide users with real-time, professional and diverse information.

Combined with the powerful semantic understanding capabilities of the volcano engine's self-developed bean bag model, Samsung's Bixby voice assistant can become more practical, providing users with rich graphic and video content such as travel planning, news and information, fitness and health.

Of course, the action of force is mutual. While empowering the Volcano Engine, it also needs enough application scenarios to test its AI capabilities, such as model inference training, timeliness, network stability, etc., so as to find deficiencies in practice and continuously improve AI capabilities, forming a positive cycle.

After all, the competition of AI mobile phones is ultimately a competition between innovation and ecology. Actively participating in the formation of the ecosystem, Volcano Engine has the opportunity to take advantage of this wave of AI mobile phones to build another "AI Android" on top of the Android ecosystem.

In other words, Volcano Engine wants its general-purpose model to become a must-have model for every AI phone manufacturer.

3. What is the confidence of Volcano Engine to be an AI ecosystem?

Not only Volcano Engine, but also many leading manufacturers are also making large models, and some models start earlier, have a higher volume, and have been developing and accumulating in many aspects for a longer time. So, how can the volcano engine build this ecology?

Many people may answer: because it is backed by the super financial resources of Byte Group, after all, the strategy of vigorously producing miracles is obvious to all.

Admittedly, group support is one of the reasons. But at the same time, we should see that the Volcano Engine has the ability to accurately gain insight into user pain points, the ultimate realization ability in technical performance, and the ultra-high efficiency in product iteration.

Recently, Volcano Engine released a family of bean package models, including bean bag general model pro and lite, bean bag role-playing model, bean bag speech synthesis model, and bean bag speech recognition model. Among them, PRO has the strongest comprehensive capabilities, supporting 128K window fine-tuning, and supporting 50+ byte services, such as reference Q&A, summary summary, creation, text classification, role-playing, etc. Lite is cost-effective, has low latency, and reduces latency by more than 50% in terms of ensuring basic quality conditions. This is a very high level of the current large-scale model track.

Relying on the bean bag model, the technical confidence of the volcano engine to support the AI mobile phone ecology can be summarized into four points:

Pictured: Tan Bei, President of Volcano Engine, announced the pricing of the bean bag model

The first and most striking point is that the pricing of the bean bag model is the lowest in the industry, and it should also be the lowest in history at home and abroad. Moreover, the bean bag model is not a little cheaper, it is a large discount on the previous industry price.

Tan Bei, president of Volcano Engine, believes that cost reduction is a key factor in driving the large model to fast forward to the "value creation stage".

There may be some people who don't quite understand the meaning behind this sentence. For example, after the popularity of GPT last year, there was a boom in large-scale model application entrepreneurship in China, but among the more than 30 startups I visited, none of them did C-end business. A CEO with a technical background once calculated such an account for the author, their main business is AI marketing writing, and every time their customers consume resources to write a 3,000-word manuscript, their own algorithm framework needs to call tens of thousands of tokens from various large models, and its cost is ten yuan. Because they are in the start-up stage, the quality of article generation is very unstable, and users can only generate a bunch of articles to pick from. The number of registered users is only a few hundred, but the resource consumption is more than 10,000 times a week, and the total cost of calls is more than 500,000 yuan. If they don't do B-end business, their investment will be spent quickly. But if you keep doing B-end business, the growth of this kind of startup will be very slow.

Although this is only a special case, it also reflects a very real problem: the high price of invoking large models has become a shackle to AI application innovation. It's not that entrepreneurs lack ingenuity, it's that their chances of validation are too expensive. If the price of large models is not lowered, AI applications will not really compete with each other.

At present, the pricing of the bean bag model is significantly lower than the market floor price: taking the pro-32k version of the bean bag general model as an example, the model inference output price is only 0.0008 yuan/1000 tokens, which is 99.3% cheaper than the model of the same specification on the market; The general model pro-128k long text version, the input price is 0.005 yuan/1000 Tokens, which is 95.8% cheaper than the model of the same specification; The general-purpose model lite can buy 1.25 million Tokens for one yuan, which is equivalent to the amount of text generated for 3 "Romance of the Three Kingdoms".

Tan Bei, President of Volcano, said that the price reduction of large models should not only provide low-cost lightweight versions, but also the main models and the most advanced models should be cheap enough to truly meet the needs of complex business scenarios of enterprises, fully verify the application value of large models, and catalyze innovation beyond existing products and organizational models.

The second point is that bean bags are the most invoked model. As we all know, the more calls a large language model has, the better the polishing effect. As of May 11, there are only three domestic large-scale model clients with more than 100 million downloads, of which bean bags rank first with 144.7 million, far exceeding the remaining second and third places. In the past year, the bean bag model has been internally polished and verified by the market, and is becoming one of the largest and most abundant application scenarios in China, with an average daily processing of 120 billion tokens of text and generating 30 million images.

The third point is to better understand the needs of users and be more practical. The more the model is used, the better it will be, the better it will be able to understand real user needs, and enable continuous iteration. It can also be seen from platforms such as Douyin, Xiaohongshu, and Toutiao that many users choose to use beanbags to solve practical problems.

Fourth, the scene is richer and more diverse. The bean bag model supports enough scenarios. On the C side, Doubao supports multiple businesses in the byte ecosystem, such as Doubao APP, Cat Box, Feishu, Tomato Novel, Hippo Aixue, Jianying, Douyin e-commerce, etc.; On the B side, the bean bag model has reached cooperation with Mengniu, China Merchants Bank, OPPO, Haidilao, Yum China, etc.

In addition, Volcano Ark has recently been upgraded to version 2.0.

Volcano Ark is a large model service platform released by Volcano Engine, which provides a full range of functions such as model training, inference, evaluation, and fine tuning, and focuses on supporting the large model ecosystem. In the face of security issues, Volcano Ark can also ensure the model security of model providers and the information security of model users through a stable and reliable security mutual trust solution, so that large models can serve all industries faster.

As we all know, enterprises often face problems such as model effect (MaaS service), application richness (corresponding to application service), inference scale, and supply (cloud service) when implementing models. The upgraded Volcano Ark 2.0 can provide a three-layer product architecture of platform capabilities, model combinations, and agents, and has one-stop model implementation capabilities, making it easier for enterprises to use.

The current Volcano Ark platform has better performance and the system carrying capacity has been greatly improved. For example, in terms of fine tuning, users can achieve better results with less investment; Another example is inference performance, all fine-tuned models on the Volcano Ark can achieve fast inference, and can reach the startup speed of 2000+ nodes in 5 minutes.

Moreover, Volcano Ark has now achieved a more complete and convenient landing of agents. For example, the professional version of the button (Coze.cn) released by Volcano Engine can realize zero-code development of various AI agents for both technical and non-technical personnel on the platform. At present, a wide range of developers have gathered on the button, and its ecology is very rich.

In addition, Volcano Ark also has more powerful large-scale model plug-ins, such as the network search plug-in, which is also the industry's first multi-modal networking component, and its real-time hotspot and intent recognition and retrieval capabilities are at the leading level in China; Another example is the content plug-in, which is selected from the intelligent content of the Douyin Group, covering 80% of Douyin's PV+ custom content; There's also the RAG knowledge base, which is more powerful and can reduce AI illusions to a great extent.

In the past two years, with the support of Volcano Ark, many companies have completed product innovation and upgrading, solved thorny problems, and greatly improved customer experience.

Taking the direction of terminal devices as an example, ASUS, as a world-renowned IT hardware manufacturer, developed the "Douding AI Assistant" based on consumers' demand for AI technology, and put it into use on the "ASUS A Bean" series of A Bean 14 2024 ADOL14VA and A Bean 14 Air M5406ADOL models in March this year. Powered by the technical foundation of the Volcano Engine, the Douding AI assistant has excellent conversation, finding, and creative capabilities. At the same time, with the help of the volcano engine model, the Douding AI assistant has become an intimate assistant for users with diversified functions.

Taking the securities industry as an example, the traditional securities customer service system has problems such as isolated product forms, insufficient generalization of intent recognition, and lack of multi-round conversation comprehension ability, resulting in poor customer experience and unable to form a one-stop wealth management assistant service capability.

Based on the Volcano Ark platform, Huatai Securities has built a new generation of wealth management assistant, which solves the problem that traditional technology cannot accurately identify intentions and cannot interact with customers in multiple rounds, and strengthens the interactive experience and one-stop service capabilities with users. Through the integration of large models, traditional models, and business transaction processes, the multiplication and multiplier effects of data elements are fully utilized, and more efficient business processes are realized.

4. AI mobile phones are a long-distance race, what is the most important thing?

When it comes to AI mobile phones, although the public recognizes the development direction, they will also worry about privacy and security. There are also many people who think that this concept is too advanced, questioning that it is new wine in old bottles, just like the original concepts of the metaverse and blockchain.

Experts predict that it will take at least 5 to 10 years for AI phones to fully reshape the mobile phone industry and make it fully accepted by society.

From the perspective of mobile phone manufacturers, this is undoubtedly a super long-distance race that has been delayed for a long time.

In order to stay at the forefront of the AI mobile phone wave, mobile phone manufacturers not only need large model manufacturers to maintain the most advanced AI technology, but also need them to maintain the most stable infrastructure supply. The key infrastructure for the training and distribution of large models is the cloud.

Volcano Engine is also the best cloud in the AI era.

The key challenges of cloud infrastructure focus on cluster management complexity, network performance, stability, etc. And the Volcano Engine can provide stable infrastructure capabilities.

In terms of ultra-large-scale computing power clusters, it supports Wanka cluster networking, 2K, 4K, 8K clusters, and Eflops computing resources. In terms of high-performance networks, it achieves one global network, 3.2Tbps RDMA network, flexible networking at Layer 2 and Layer 3, and self-developed BCC congestion algorithm. In terms of intelligent O&M, it can achieve the capabilities of active fault discovery, second-level and millisecond-level monitoring, and minute-level fault detection and discovery. In addition, Volcano Engine can also provide a multi-core, polymorphic AI infrastructure.

At present, many of the capabilities of Volcano Engine have been successfully verified in cooperation with ASUS, OPPO, vivo and other enterprises.

In response to the problems of model timeliness and poor intent recognition ability encountered by OPPO in the landing of large models, Volcano Engine has jointly developed with OPPO for AI mobile phone capabilities by virtue of various advanced large model capabilities within Bytes, and has launched various AI agent applications on new models such as OPPO Find X7 and OnePlus Ace in 2024, including [AI Interviewer], [AI English Teacher], lively and cute girl [Xiaoqing], and mature and stable [Wenfeng], which have been comprehensively improved AI competitiveness of OPPO phones in the market.

With the help of the Volcano Ark platform, vivo has also completed the online service creation, deployment and online inference of the Blue Heart 7B model, providing developers with a complete platform tool. Moreover, vivo relies on an open large model ecosystem to support the promotion and application of the Blue Heart 7B large model.

The development of the AI mobile phone market is destined to be a "long slope and thick snow".

In this 5 to 10-year long-distance race, whoever can maintain long-term cooperation with partners with leading technology and stable ecology will have a better chance of running in the first echelon of AI mobile phones.

View original image 308K

•