Author: Liu Wenxuan.
In the second half of this year, large models have become the focus of widespread discussion, from the "100-model war" to the "industry model", "device-side model", and even smartphones have participated.
The large model can help the users of the terminal equipment to improve the productivity, for the productivity, what kind of terminal equipment can give full play to the role of the large model? The answer given by Lenovo is - AI PC.
On December 7, at the first AI PC Industry Innovation Forum held by Lenovo, Lenovo Group and IDC jointly released the first "AI PC Industry (China)**hereinafter referred to as "*".
** pointed out that in the future, each individual can have their own AI PC and run their own "personal large model". The personal model can inherit the powerful capabilities of the public model, and can provide personalized and exclusive services for individuals, so as to meet the needs of users in various aspects. The key to model termination is to compress large models to a scale suitable for the terminal for local inference and real-time response. For this reason, technologies such as model distillation and compression become crucial.
Why AI PC?
Personal computers (PCs) have powerful computing and storage capabilities, rich interaction methods, and a wide range of application scenarios, and have full-modal human-computer natural interaction conditions, making it an ideal platform suitable for carrying large models. AI PCs will become indispensable personal AI assistants for individuals, families, and enterprises, providing users with personalized services in general scenarios, providing instant and reliable service responses, lower cost of using large models, and trusted and secure personal data and privacy guarantees.
Provide personalized services in general scenarios:
AI PC can provide personalized services such as personalized creation services, personal secretary services, and equipment butler services for work, study, Xi, and life.
Based on the customized design of terminal manufacturers, scenario-based function presets, and continuous exploration of user needs, with the support of a rich model and application ecosystem, AI PCs have the capabilities of personalized creation, secretarial services, and equipment management, which can reflect a variety of unique values in work, study, Xi, and life and entertainment scenarios.
Immediate, reliable service response:
In the 2023 IDC survey on users' experience of using the AIGC platform, "slow response speed" and "long feedback time" are the main negative feedback from users. AI PC is mainly based on local inference, supplemented by edge and cloud inference, which can intelligently and reasonably allocate tasks between hybrid computing power and hybrid models, effectively reducing response time.
AI PCs can still function without an Internet connection, and can do creative work for users at any time and from anywhere, so that users are no longer constrained by network conditions. Local large models are usually more easily integrated and fully utilized with local knowledge and data, which can effectively avoid the generation of "illusions". And according to the user's style preferences, it can accurately generate works that meet the needs of users. This accurate, reliable service is a unique advantage of AI PCs.
Lower cost of using large models:
AI PC is mainly based on local inference, supplemented by cloud public inference, and users can enjoy free local inference services throughout the life cycle after purchasing AI PCs at one time, plus limited cloud subscriptions, which can significantly reduce the cost of individual users using AI large model services.
For application vendors, moving some processing from the cloud to the endpoint can greatly reduce the pressure on cloud infrastructure and reduce operational expenses. Indie app developers and app developers will also rely on AI PCs to explore and build apps more cost-effectively and efficiently.
Trusted, secure personal data and privacy guarantees:
As the foundation of security, the personalized local knowledge base of AI PC has a secure space dedicated to storing user-specific types of files and data, ensuring that the privacy and confidential information of individual users and enterprise users can be securely isolated locally and can only be called in a trusted environment.
The AI PC itself can effectively identify and manage user privacy data through personal agents, and complete most of the data processing and model computing locally through localized models and sufficient AI computing power, and only public requests that do not involve user privacy need to call cloud services.
In order to realize these user values, the AI PC will not just be a hardware device, but a hybrid that includes AI models and applications, as well as hardware devices. At the same time, driven by AI PC, the PC industry ecosystem will undergo subversive changes, forming a new AI PC ecosystem: people-oriented, terminal-led, and AI-native.
As a device that carries local hybrid AI computing power, AI PC also covers a variety of product forms, including AI laptops, AI tablets, AI desktops and other forms, among which AI desktops include not only traditional desktop computers, workstations, all-in-one machines (AIO) and other forms, but also edge devices that closely coordinate with terminals, home AI hosts, and enterprise AI edge hosts. In addition, with the development of AI PCs in the future, more innovative forms will emerge.
In the era of AI PCs, terminal devices, personal models, and personal agents have become an inseparable combination, working together to bring users an intelligent experience. The innovation of AI PCs needs to be driven by the satisfaction of user needs and the improvement of experience, and the combination of "device + personal model + personal agent" needs to be integrated and innovated.
In the new industry landscape, terminal manufacturers, as the core hub of communication between users and the entire industry, as well as the promoter and integrator of the transformation of computing power, model and application ecology, are the best deliverers to complete this integrated innovation. On AI PCs, personal agents will serve as an integrated entrance for users to AI functions, and terminal manufacturers will become the creators and maintainers of this entrance experience with decades of research and accumulation of user interaction and experience Xi.
Five features make AI PC the preferred terminal for AI inclusiveness
Fok Jinjie, President of IDC China, said that AI PC is not only a brand-new product, it represents that the entire industry is about to enter a new era.
Fok Jinjie, President of IDC China.
Dr. Wang Jiping, Vice President of IDC China, interpreted the "**" released this time. He pointed out that the AI PC integrates a hybrid AI computing unit on the hardware, and can run a "personal large model" locally, create a personalized local knowledge base, and achieve personalized and natural interaction. It not only assumes the functions of the original productivity tool and content consumption carrier, but also upgrades the function from the tool attribute to the assistant attribute, becoming a personal "second brain" and a personal AI assistant like a twin with the user.
Jiping Wang, Vice President of IDC China.
AI PC will create four core values for users: First, personalized services in general scenarios, providing personalized creation, personal secretary, and equipment butler services for general scenarios such as work, Xi, and life. Second, the service response is immediate, reliable and can be used even when offline. Third, the cost of using large models is lower, with local free inference as the mainstay, supplemented by cloud public inference, and you can enjoy free local inference services throughout the life cycle with a one-time purchase of an AI PC. Fourth, trusted and secure personal data and privacy protection, personalized local knowledge base, local inference mechanism for private data, and Device Manager service bring users the ultimate security experience.
Wang Jiping concluded that a real AI PC product should have five characteristics:
Personal agents for natural language interactions:Multimodal natural language interaction UI, personal agent based on local large model;
Embedded personal model in the terminal:A large model that focuses on local and supplemented by edge and cloud; Personalized local knowledge base;
The terminal is equipped with local hybrid AI computing power as standardCPU&NPU&GPU local hybrid computing architecture, personal terminal and home host, enterprise edge host collaborative computing;
Connect to the open AI application ecosystemAI-native applications and AI-enabled applications can be scheduled by agents and adapted to hybrid AI computing platforms.
Device-level personal data & privacy protection:Local privacy inference > non-sensitive tasks call cloud large models, hardware-level security chip protection & personal data encryption desensitization transmission.
However, Wang Jiping believes that it will take a while to evolve to a state where the five characteristics are complete. He divided the development of AI PC into two stages: AI ready and AI on.
AI Ready Stage,The product has basic local hybrid AI computing power, which can provide basic guarantee for software and service innovation of AI PCs, and start experience innovation. In the initial stage, the main performance is the upgrade of the chip computing architecture, and the CPU with the integrated NPU computing unit has been introduced to the market one after another, with a higher energy efficiency ratio to achieve the improvement of computing speed, and has higher stability and reliability in the operation process. The hybrid AI computing power of terminal devices enables AI software to process data and tasks faster, which is especially important for AI applications that require real-time responses, such as speech recognition and image processing.
In 2023, AI PCs in the AI Ready stage will be launched one after another. This type of AI PC has a certain amount of AI acceleration computing power in hardware, but it does not yet have the complete characteristics of AI PC.
ai on stage,The product has complete core features of AI PC, and provides an epoch-making AI innovation experience in core scenarios, becoming everyone's personal AI assistant. With the innovation of core technologies, the optimization of product experience, and the prosperity of the AI application ecosystem, AI PCs can serve a wider range of general scenarios, and can realize device-edge collaborative computing, cross-device interconnection capabilities, and even fine-tuning and training of individual large models in the edge private domain environment based on personal data and usage history.
Starting in 2024, AI PCs that meet the standards of the AI ON stage, that is, have complete core characteristics of AI PCs, will enter the market one after another, which will further drive the rapid growth of the market and inject new vitality into the development of the entire industry.
Personalized, build a personal model
Since the beginning of this year, generative AI technology based on large models has made major breakthroughs and developed rapidly. On the other hand, the so-called large model is a public large model pre-trained based on a public database, which provides basic inference services that do not involve personal privacy data, and is not an exclusive AI for individuals.
Ablikim Ablimiti (hereinafter referred to as "Amu"), Vice President of Lenovo Group and Vice President of Strategy and Business Development in China, reviewed the history of PC development for decades at the forum.
Ablikim Ablimiti, Vice President of Lenovo Group and Vice President of Strategy and Business Development in China.
He introduced that the first upgrade experienced by the PC was in the late 70s to early 80s of the last century. At that time, the introduction of large-scale integrated circuits and new CPU general computing architectures greatly reduced the cost of computers, and at the same time, they also began to be miniaturized, and more intuitive graphics interaction methods also allowed ordinary people to get started quickly, lowering the threshold for use. At the same time, in China, Lenovo and other terminal manufacturers have been established, and launched personal computer products, quickly forming a global industrial chain of personal computers, the era of personal computers has officially arrived, and PCs have truly become veritable personal computers.
The second upgrade of the PC At the turn of the century, the development of network communication technology and mobile computing technology allowed the Internet to establish a connection for every computer. During this period, Lenovo's "Tianxi" Internet computer was born, which truly changed the relationship between PCs and users through "one-click Internet access", and at the same time won the world's top PC in the context of global development.
Amu pointed out that at present, the PC is ushering in the third major upgrade. "Hybrid AI is increasingly evolving into a more definite trend for AI to serve everyone better and more exclusively in the future," he said. PCs have shown excellent performance and performance in all aspects, and are expected to once again become the go-to terminal for AI inclusion to everyone. PC will take on the historic mission of bringing users an epoch-making new AI experience, so that AI can truly become everyone's personal AI twin. ”
"For the sake of data security and privacy protection, as well as to respond to user needs with higher efficiency and lower cost, people not only want to get the powerful general services of the public model, but also hope that AI can truly understand themselves, provide exclusive personal services, and fully protect personal data and privacy security," Amu said. He believes that the hybrid deployment of public and personal models to meet the needs of users is becoming more and more an industry consensus, and hybrid artificial intelligence is increasingly becoming the development trend of AI in the future to better and exclusively serve everyone.
In order for everyone to have their own AI, it is necessary to build a personal model. The establishment, deployment and application of a large personal model is not an easy task, and it requires terminals that can carry multi-modal natural language interaction, can carry complex general scenario functions, have powerful local intelligent computing power, and can provide sufficiently large, credible and secure personal data storage and protection functions.
Amu said that Lenovo, as a leader in global PC and Chinese PC, has completed the preliminary research and insight of China's AI PC with IDC, which has been deeply involved in the IT industry for many years, and took the opportunity of today's innovation forum to officially release it, and rely on this platform to communicate with all walks of life in the industry to accelerate the path and method of AIPC development.
Fang Jing, chief analyst of Minsheng** electronics industry.
Fang Jing, chief analyst of Minsheng** electronics industry, believes that the current PC market has ended destocking and has started to recover, "The surging computing power in the first half of the year is the supply side, and the current intensive landing of AI terminals is the demand side." Among many end products, AI PCs can perfectly solve the needs of local large-scale model applications, privacy protection, and customized solutions. Manufacturers represented by Lenovo will occupy a core position in the new AI ecosystem and are expected to lead this wave. ”
Industrial transformation and building a new ecosystem for the AI PC industry
AI PC will open up a new AI PC ecosystem that is completely different from the past: human-oriented, terminal-oriented, and AI-native. AI PCs are about to bring disruptive changes, and the PC industry ecology will shift from "application-oriented" to "people-oriented", from application-driven to intent-driven, and users will become the drivers and creators of industry ecological innovation.
In the new AI PC era, terminal devices, personal large models, and personal agents have become an inseparable combination to bring intelligent experience to users. The innovation of AI PC requires the integration and innovation of "device + personal model + personal agent". As the best deliverer of integrated innovation, terminal vendors will become the core hub for communication between users and the entire industry, and will be responsible for promoting the transformation and integration of computing power, models, and application ecosystems. The terminal manufacturer has advanced from a product provider to an ecological organizer.
The traditional PC industry ecosystem is based on the operating system, where users operate directly in the system interface and manage and apply a variety of applications. In the new AI PC ecosystem, the iteration of AI PC technology and experience cannot be achieved alone through application and system updates, or the upgrade of cloud model technology, nor can it rely solely on the update and iteration of computing power products.
Large model vendors generally use the "plug-in" method to invoke the modularization capabilities of other applications. The plug-in of applications and the integration of large models means that third-party applications can become part of the capabilities of large models and provide services to users. Only by combining the large model with the ecological plug-in function can "omnipotent" be realized.
** pointed out that with the continuous development of the AI application ecosystem, the proportion of AI native applications around public and local large models will continue to rise. AI native applications are developed with large models at their core, and various AI tools and frameworks are used in the development process to implement AI functions and features. The personal agent will replace the operating system as the user's integrated entrance to AI functions, and with the support of the large model and application ecosystem, it will complete the tasks proposed by the user.
An open application ecosystem should include AI-native applications, AI-enabled applications, and public models. AI-native applications refer to those applications that are built directly on AI algorithms with AI capabilities as the core from the beginning of infrastructure and design, and will become more and more common in the era of large models. AI-empowered applications refer to traditional applications that are also plugged and enhanced into functional capabilities that can be called up or called by large models.
More and more enterprises are joining the open ecosystem of AI PCs, forming a prosperous ecosystem with multiple layers of open users, terminal vendors, models, applications, and computing power. The user's vision will change, and the new relationship between the model and the application will change the underlying logic of the user's needs. User needs are the key element to promote the development of the entire ecosystem, and terminal manufacturers integrate industrial resources for users based on scenario requirements to provide a hybrid delivery experience that integrates software and hardware. Model manufacturers and chip manufacturers will also have closer ties with terminal manufacturers.
** pointed out that in the era of AI PC, the user's voice will be improved, and the user's initiative will increasingly affect the evolution of products and the development of the industrial ecology. The development of AI technology will further lower the threshold for program development, and users can directly "develop" applications using AI PCs through the guiding idea of problem + solution, or they can make their own applications and software with the help of AI large models.
With the increase in the number of user-developed mass innovation applications on the device side and in the cloud, a new form and operation mode of the community will also be formed. Users can share their AI-created apps on the community for other users** and uses; Users can comment and communicate with UGA on the community, post paid needs, and obtain income.
Based on the personal model, the agent calls various applications through plug-ins, and the user's evaluation and usage behavior will become the key basis for the distribution of the agent's application. Users only need to give feedback on the results of their experience to influence the app's call priority and how often it reaches users. The personal agent can select and switch the service provider behind the application at any time according to the user's behavioral feedback, forming a form of user-driven application innovation.
Started the first year of AI PC
At this forum, Lenovo Group also officially released and launched the "AI PC Pioneer Action". The action aims to co-prosper the ecosystem and accelerate the early implementation of AI PCs, including two aspects: the "AI PC Ecological Developer Program" launched for AI ecosystem developers; The "AI PC Content Creation Contest" was launched for AI content creators.
Wang Chuandong, Vice President of Lenovo Group and Chief Marketing Officer of Lenovo China.
Wang Chuandong, Vice President of Lenovo Group and Chief Marketing Officer of Lenovo China, announced the official release and launch of the Lenovo AI PC Pioneer Action with the theme of "Lenovo AI PC Together". The action includes two aspects: the "AI PC Ecosystem Developer Program" for AI ecosystem developers; The "AI PC Content Creation Contest" was launched for AI content creators. Wang Chuandong said that Lenovo's AI PC Pioneer Action aims to co-prosper the ecosystem, accelerate the early implementation of AI PC, and make AI PC benefit everyone.
Wang Chuandong introduced that Lenovo will also hold the 2023 Lenovo Tianxi Ecological Partner Conference on December 26, at which time it will release an AI applet platform based on the Tianxi ecosystem, and welcome more AI PC developers and partners to participate in the construction of the AI ecosystem and win the future together. At the same time, Lenovo will also cooperate with AI PC ecosystem partners to hold an AI PC innovative application competition, so that more developers can share the dividends of the AI PC era. In the "AI PC Content Creativity Competition", Lenovo will invite AI bloggers, AI graphic designers, and AI content creators who are full of imagination and creativity to imagine the bright future of AI PCs.
Regarding the market prospect of AI PC, Gao Fei, editor-in-chief of Zhiding Technology, said that there is a widely circulated saying in the industry that large models will make all software do it again. However, in fact, large models will also change the form of hardware. Just like the history of the development of mobile Internet and smart phones, there will be a process of mutual achievement. That is, new hardware forms create new scenarios, and new scenarios in turn put forward higher requirements for hardware. The computing power supply of AI PCs is just right for users to be both "smart enough" and "reliable enough" for large models on the device side.
Fang Jing said that driven by leading manufacturers such as Lenovo, all links of the industrial chain will become "AI PC concept stocks", such as processor packaging manufacturers with built-in NPU units - Tongfu Microelectronics, Changdian Technology, etc., and heat dissipation faucets - Siquan New Materials, Zhongshi Technology, Feirongda, etc.; Electromagnetic shielding faucet - Longyang Electronics; Structural parts faucet - Everbright Tongchuang, Spring and Autumn Electronics, etc.
In addition to the change in the identity of the terminal manufacturer, AI technology (large model) manufacturers have been introduced for the first time in the new ecosystem of the AI PC industry. Among the many large-scale model manufacturers, the understanding of the large-scale model by the terminal equipment manufacturer will play a key role. Dr. Fan Jianping, vice president of Lenovo Group and head of the artificial intelligence laboratory of Lenovo Research Institute, said that in order to provide truly credible and personalized services, Lenovo proposed the concept of "cloud public large model + enterprise private large model + end-side personal large model" hybrid intelligence (hybrid AI).
Fan Jianping, vice president of Lenovo Group and head of the artificial intelligence laboratory of Lenovo Research Institute.
In order to achieve the goal of terminal manufacturers to build a hybrid intelligent model, large model manufacturers need to provide public large model technology that can be built into lightweight local large models, which can be optimized and adapted for terminals of different forms and performances, and may also launch personalized fine-tuning and self-fine-tuning services for users.
While large model manufacturers bring users lightweight and miniaturized large models suitable for AI PCs, they also bring infinite imagination to the application of AI PCs. In the future, the proportion of AI-native applications around public and local models will continue to rise. Many traditional applications will also be empowered by large models through cloud calls and other methods to improve the efficiency and intelligence of task completion. In most cases, being called by an individual agent in the form of a plug-in will be the primary way to launch and use the application, rather than revolving around the operating system framework.
Tongyi Qianwen has taken the lead in the industry to realize full-size and full-modal open source. Zhou Chang, head of Alibaba-Tongyi Lab and Tongyi Qianwen model, said that since its release in April this year, the Tongyi Qianwen model has continued to iterate, and has successively open-sourced 4 large language models with different parameter sizes and two multimodal large models, which have been widely praised by the global open source community. Both developers and end users are looking forward to the prospect of end-side deployment of Tongyi Qianwen, "Lenovo is a global PC leader, and we will strengthen cooperation with Lenovo to jointly explore and promote more scenarios in the ecological field to implement large models." ”
Zhou Chang, head of Alibaba-Tongyi Lab and Tongyi Qianwen model.
At the forum, Chen Bo, assistant president of Kingsoft Office, further demonstrated the WPS AI of the intelligent office assistant developed by Kingsoft based on large language models. He said that he looks forward to building a new ecosystem with Lenovo to realize the combination of WPS AI and AI PC to ensure the security and confidentiality of enterprise and personal data assets.
Chen Bo, assistant president of Kingsoft Office.
Zhou Feng, Executive Vice President of NetEase and CEO of NetEase Youdao, introduced the first vertical model in the field of education in China - "Ziyue" education model and related applications, and said: "AI PC will bring huge development opportunities to the industry and ecology. ”
Zhou Feng, Executive Vice President of NetEase and CEO of NetEase Youdao.
Like application vendors, computing power vendors have existed in the PC industry ecosystem before, but their roles are also changing under the catalysis of large models. In terms of building local intelligent computing power, it is one of the most mature solutions to provide local computing power with a CPU + GPU + NPU heterogeneous architecture solution. Heterogeneous hybrid computing power not only ensures the feasibility of terminal model inference, but also rapidly improves computing power, enabling terminal devices to carry larger and larger AI models and perform more and more complex tasks.
At the meeting, Gao Yu, general manager of technology at Intel China, said: "AI workloads come in many shapes and sizes. So, there is no one-size-fits-all solution from the hardware. Therefore, based on years of Xi and market experience, we proposed the concept of XPU, including GPU NPU CPU. He also said that Lenovo is a strategic partner of Intel, and the two companies are already advancing the development and innovation of AI experiences based on the upcoming Meteor Lake processors. ”
Gao Yu, general manager of technology at Intel China.
Amu said: "The maturity of the AI PC industry requires a process, and industry participants such as machine manufacturers, models, applications, and computing power need to work together with users to create a new AI ecosystem in China. ”
IDC**, AI PCs (excluding AI tablets) in the Chinese PC market will increase rapidly in the next few years, reaching 85% by 2027, becoming the mainstream of the PC market. Among them, the wave of replacement in the consumer market will accelerate, and the proportion of new AI PC assembly will be slightly higher than that of the overall market; The proportion of AI PCs assembled in the SMB market will increase faster than in the consumer market, reaching 88% by 2027. Among large enterprises, the proportion of new AI PCs assembled will reach 74% by 2027, among which large enterprises in IT, Internet, finance, and professional services will take the lead in introducing AI PCs.
IDC data shows that the current crowd structure of the PC consumer market is conservative, and the total proportion of thrifty herd users and smart and autonomous users who are sensitive to the best and pay attention to practical functions is close to 6%. With the advent of AI PCs, the demographic structure of the IDC PC consumer market will change. Taste-savvy users who are sensitive to new technologies and willing to try new technologies will be the first early adopters of AI PCs. After that, more and more stable and decent users who pay attention to the quality of life and work efficiency will replace their old phones under the influence of tasters, become the mainstream users of AI PCs and drive the high-end population structure of the PC consumer market. After 2025, with the popularization of AI PCs, more and more savvy autonomous users who are sensitive to changes in technical parameters and ** and pay attention to performance and functions will become the main consumers of AI PCs.
Driven by AI PCs, the penetration rate of personal computers in Chinese consumer markets will break through the bottleneck period and achieve continuous growth. In the next few years, more and more families will choose AI PCs as the first terminal for home office, Xi, entertainment, and creation because of their unique advantages of multi-modal interaction, full-scene application, strong computing power, and secure storage.
Wang Chuandong predicts that 2024 will become the first year of China's AI PC. Based on the new IT technology framework of "device-edge-cloud-network-intelligence", Lenovo Group has formed full-stack intelligent capabilities such as AI-embedded intelligent terminals, AI-oriented infrastructure, and AI-native solution services, laying a solid foundation for fully supporting the deployment of large models in terminals, and making the most adequate preparations for AI to benefit every person and enterprise.