In this era of large models, all walks of life are actively embracing this wave of wind and trying to make the world full of AI. According to the latest publicly available data, as of March 2024, a total of 117 large models in China have successfully completed filing. These large models cover a wide range of fields and have sprung up in a wide range of industries.
However, for ordinary consumers, AI and large models are still floating in the air, and people mostly ask him questions and check information, which is still a bit far from being really used. It is said that AI is the fourth industrial revolution, so how does AI change work and life?
It's time to plug in the wings of AI to the production tool PC of migrant workers, and work efficiency can take off directly. As a result, more and more PCs are embracing AI. According to data from Counterpoint Research, the global PC market grew by about 3% in the first quarter of 2024 compared to the same period last year, mainly due to the strong momentum of AI PCs. It is expected that 45% of new laptops shipped throughout the year will be AI-capable.
Faced with the dazzling array of AI products on the market, I believe many users are as pensive as I am. So is the AI PC a real application or a fake requirement?
AI PCs are coming from the cloud to the side
In fact, the application of AI on the PC side is not a new topic. It is understood that the PC side is an early attempt carrier of AI intelligence. Microsoft implanted Cortana, an AI voice assistant, in Windows in 2015. In 2021, the release of Intel's 11th Gen Core processors has enabled AI technology to begin to appear in the daily use scenarios of PCs, realizing AI noise reduction, AI background blur, and AI radio functions on the PC side, significantly optimizing the use experience of remote office. In 2022, ChatGPT started a wave of AI large models, and AI application scenarios are becoming more and more abundant.
AI intelligence in PCs is driven by the synergy of hardware and software. On the one hand, the stable operation of AI applications requires sufficient computing power from hardware. On the other hand, the software layer can improve the user experience of AI applications. After continuous development in recent years, the hardware side and the software side are now maturing together, and it is expected that the AI intelligence process on the PC side will enter the fast lane
So what are the unique characteristics of AI PC that have evolved into a new species?
AI PC is a combination of AI models and PCs, bringing innovations in architecture design, interaction methods, content, and application ecology, in an attempt to deeply transform the PC industry.
It is important to know that the essence of AI PC is the collaboration between the cloud and the local side, using the big data processing power of the cloud to enrich the local PC usage scenarios, and relying on the cloud computing power to improve the local performance balance. Specifically, AI PCs can create local knowledge bases, run personal large models, support artificial intelligence calculations, and have stronger and more creative capabilities for natural interaction. Its most important feature is the use of a local knowledge base, which protects user privacy and meets the personalized needs of users.
Therefore, the industry generally believes that end-to-device AI will be the next stage of AI development. The PC side has stronger computing power than the mobile phone side, and it can also meet the needs of more large models, becoming the first smart device to be equipped with a large model on the device side.
In this regard, analysts also gave positive expectations. According to Canalys' forecast, AI-compatible PCs are expected to spread rapidly from 2025 onwards, with a penetration rate of about 37%, and it is expected that AI-compatible PCs will account for about 60% of all PC shipments in 2027, and AI PCs will become mainstream.
Hybrid AI models are the right way to open the device
To answer the question of how can AI reach and empower more industries, then hybrid AI is the inevitable path.
The reason for this is that public models are the catalyst and accelerator for the popularization of AI, but they have limitations in terms of network speed, cloud efficiency, cost considerations, and data privacy.
"Not every piece of information has to be in the public cloud, and a 'hybrid AI' environment is necessary. The most private data deserves the highest level of security and should not exist on the public cloud. Intel CEO Pat Gelsinger believes.
In the hybrid AI framework, the personal model and the enterprise model, as well as the personal agent and enterprise agent developed on the basis of it, can provide more personalized services and meet the needs of privacy protection, and realize coexistence and complementarity with the public model.
Personal agent is through large model compression technology, running on personal intelligent terminals or edge devices, receiving instructions in a natural interactive way, through personal travel records, shopping preferences, and other information stored on the device, better reasoning and action, it can even predict the next task according to your thinking pattern and behavior frequency, and take the initiative to make suggestions and find solutions independently.
At the same time, enterprises will also benefit from enterprise agents. It is scattered in multiple terminal equipment and infrastructure of the enterprise, and will be able to learn and reason about a large amount of enterprise data, support enterprise operation decision-making, improve management efficiency, and improve productivity while ensuring information security.
Qualcomm President and CEO Ammon also pointed out that hybrid AI can use AI on both the device side and the cloud, and when the cloud and the device use the same generative AI model, the device can bring a first-mover advantage to the cloud. End-side data can also improve the accuracy of AI applications, as it now has contextual information about the user.
Therefore, Yang Yuanqing, chairman and CEO of Lenovo Group, believes that in the future, individual models and enterprise models will coexist and complement each other with public models in the form of built-in personal agents and enterprise agents, so as to constitute hybrid artificial intelligence, which is the direction in which AI can be popularized and applied.
How to lay out the panoramic artificial intelligence from the "pocket to the cloud"? Yang Yuanqing recently announced at the Techworld conference how Lenovo AI PC realizes hybrid artificial intelligence.
First, the "intent understanding" technique based on a mixture of large and small models. Lenovo uses the cross-entropy loss minimization principle to optimally allocate the intent understanding task to both large and small models, thus balancing accuracy and complexity.
Second, the "heterogeneous computing" technology based on the hybrid scheduling of CPU, GPU, NPU. In the training and inference process of large models, the bottleneck often lies not in the computing power of the chip, but in the data transmission. Optimize both compute load and data transfer to minimize overall execution time.
Thirdly, the "intelligent question answering" technology based on the mixture of Model Fine-tuning (SFT) and Retrieval Enhancement (RAG).
Fourth, a hybrid "privacy security" technology based on hardware encryption and full-stack trusted architecture.
These four hybrid technologies also pose great challenges to enterprises, because they require the collaboration of device, edge, and cloud, software and hardware, and transmission and computing. To this end, Lenovo's new IT architecture, "Device-Edge-Cloud-Network-Intelligence", supports the implementation of hybrid technologies.
Hybrid AI will be the future. The "personal AI twin" and "enterprise-level AI twin" will coexist and complement public AI models, so that hybrid AI can truly be implemented and popularized.
AI PCs can bring about these changes
So what changes can this AI PC bring to people's lives?
Zhang Yaqin, an academician of the Chinese Academy of Engineering, said: "The way of thinking and acting of agents is closer to that of humans, and it is a real intelligent assistant for personal applications. For enterprise applications, agent technology has great potential to mobilize these complex systems with maximum efficiency, dramatically improving operational efficiency and productivity levels. ”
From this perspective, AI PCs will revolutionize productivity. As a hybrid of devices, edge computing, and cloud technologies, AI PCs will not only have powerful computing power and advanced AI technologies, but also bring innovative interactive and visual experiences to meet the needs of new generative AI workloads, while also providing users with a tailored experience
AI has the power to make it easier for human imagination to turn into reality. AI can help thousands of young people with movie dreams to cross the threshold of funding and directly present the great movie in their minds. The famous director Lu Chuan said that the most direct effect of AI-assisted film production is that it can greatly improve the visualization speed of creativity. Take my new film as an example, in the traditional way, it took two months from writing the script to the completion of production, but now it only takes two days. It greatly saves time and labor costs.
For manufacturing companies, the impact of natural disasters on their supply chain management is huge. If a typhoon is coming, enterprises can first get the accurate location and wind prediction of the typhoon through the public model, but when it comes to which specific customers' order delivery may be affected, it is time for the enterprise intelligent body to show its talents.
The agent can tell us which orders have not yet been produced and may be affected by the temporary closure of the factory, which orders have not yet been shipped and may be affected by the disruption of ground transportation, and which orders have not yet been scheduled and may be affected by material delays. This greatly improves the decision-making speed of enterprises to respond to crises on the premise of ensuring customer information security.
"The second half of artificial intelligence must be from technological breakthroughs to the stage of landing applications, which requires us to accumulate user feedback in practical applications, continuously improve and continue to innovate. Yang Yuanqing, chairman and CEO of Lenovo Group, said that the way to land is to cover the three pillars of smart devices, smart infrastructure, smart solutions and services.
For individual users, the most convenient way to obtain the experience of personal agents is to have personal agents built into their own computers, mobile phones and other personal computing devices.
For enterprise customers, the implementation of AI must rely on intelligent infrastructure. Enterprises need servers, networks, and storage devices to put massive amounts of data in place and make the most of them, and then realize the data intelligence of the enterprise through AI models and algorithms.
The design, deployment, management, and maintenance of hybrid intelligent infrastructure is more complex than that of conventional infrastructure, so more comprehensive and systematic service capabilities are required. For the complex data of different industries and different types of enterprises, how to learn and reason about it by cleaning, sorting, and then using appropriate algorithm models, so as to generate intelligence in all aspects of enterprise operation, such intelligent solutions will become a hot new IT application in all walks of life.
If you want to realize these three pillars, the whole industry still has a long way to go, but AI PC can allow consumers to intuitively feel the large model, and can also use it to exert their imagination to create a different imagination space, and everyone will know how to use it better.
It is believed that with the continuous iteration of the whole industry, these hybrid large models will move from "usable" to "easy to use", so as to truly make the world full of AI.