AI Pin is a crazy experiment, and AI PC is the renaissance of personal computing

Original source: Silicon Star People

Image source: Generated by Unbounded AI

It’s time to revisit Bill Gates’ vision of 48 years ago – “a computer in every home”. This sentence should be upgraded to “let every family have an AI PC”, and this should be a new starting point for human beings to create more possibilities.

The most exciting “Game of Thrones” in the world today is the game and pull between OpenAI’s board of directors and Sam Altman, CEO who was ousted by lightning and made a dramatic return. OpenAI, the world’s top startup that used ChatGPT to detonate the generative AI revolution, has been countered in the form of human struggle, which is enough to reflect people’s more direct, deep and even instinctive worries about the future of “super AI”.

These concerns come from the omnipotence of artificial general intelligence (AGI), from the excessive optimism and extreme laissez-faire of some intelligent human beings for the development of “super AI”, from the invasion of AI on the boundaries of human life and privacy, and from the awakening and efforts of human beings themselves - by creating new “hardware”, “super AI” is locked in a physical cage, endowed with more powerful computing resources and computing power, and at the same time restraining it from human work, learning, The excessive acquisition of life and private data allows it to better serve human creation and progress, rather than overriding it.

This means that OpenAI and Microsoft are not the only options for people to get to the inevitable kingdom of AI, and few hope so. This means that the consciousness and explosion of “hardware innovation” means that personal computing devices are about to enter an era of conspiracy and dance with generative artificial intelligence after experiencing two generational leaps between personal computers (PCs) and smartphones.

What humans need, is it a hard plug-in for ChatGPT?

The market has already given a more “radical” answer: Sam Altman, the former and returning CEO of OpenAI, participated in the investment of a company called Humane to launch an “epoch-making” AI hardware - AI Pin. Dubbed the “iPhone of the AI era” by many media outlets, this hardware has no screen and display, the content is projected onto the hand with a laser, and simple touch and voice are the only way it can interact with people. To be precise, it’s a “shell” of batteries, laser projections, mobile networks, cameras, and microphones that hang on people’s chests. And its real “soul” comes from OpenAI’s $24 a month GPT service.

AI Pins

In other words, it is completely a “hard plug-in” of ChatGPT hanging on people’s chests, and people can only use it smoothly if they give all their information to ChatGPT and the OpenAI behind it by default, upload it to the cloud (Microsoft’s Azure), and by default this information and data can be used by OpenAI to train data that can serve others. And because of this, it is not destined to become an epoch-making piece of hardware that has swept the world like the iPhone, and it embodies the arrogance of AI in protecting human privacy and dignity, and it is enough to ignore the vision of human beings creating the future world through AI instead of using AI as a “crutch”.

The epoch-making hardware that people really need in the future and embraces “super AI” should be a “productivity tool” that supports the imagination and possibilities of human beings to create the future, and a new smart device that fully protects the privacy of human data and stimulates and restricts AI capabilities. It doesn’t necessarily follow the evolutionary path of personal computing devices from PCs to smartphones to future devices with more pocket-sized minis. Compared with the generative AI we are embracing, the evolution from PCs to smartphones in the past 50 years can only be classified into the same era - the digital improvement of productivity is to achieve the real “primitive accumulation” in the PC era, and the real value of smartphones is to share the fruits of human digital productivity in the past 15 years to more people for consumption.

In this sense, the intelligent improvement of productivity under the name of “GenAI” (generative artificial intelligence) is the beginning of a new era of information technology. And the primitive accumulation of human beings in the early stage of intelligent productivity improvement will allow human beings to “recover” a more powerful new hardware device in a sense - it is intelligent in soul, and may not be so radical in body, and even a little retro - in a sense, you can see it as the renaissance of some “ancient hardware”, that is, the renaissance of PC.

In other words: the real “AIGC” (AI-generated content) will first happen on the “AI PC” (AI personal computer) on a large scale. The birth of “AI PC” was first catalyzed by “AIGC”, and it has become a fact that it is happening.

On October 23, 2023, at the Lenovo Innovation Conference (TechWorld), Lenovo, the overlord in the PC field for many years, has demonstrated the prototype of the “AI PC”, unveiling the possibility of a new round of hardware revolution. And upstream chip manufacturers are also competing for strength.

The process of AI PC has been started one after another.

In October 2023, at Qualcomm’s Snapdragon Summit, a new mobile SoC platform and PC SoC platform were announced, both of which are capable of running generative AI models with up to 10 billion and 13 billion parameters on the “device side”. Intel is also expected to release a new Core processor at the end of 2023, which will improve the power of AI computing by adding a neural network acceleration unit (NPU) for the first time. Two weeks ago, Intel CEO Pat Gilsinger set a clear goal of shipping more than 100 million AI PCs by 2025.

The end-side large model is a rigid need

When it comes to AI models, many people must think of “computing power, storage, and network”. Far beyond the performance requirements of ordinary software programs, it has been very close to cloud computing in its early development.

As the person who uses GPUs to promote the development of the entire AI industry, Jensen Huang, the leader of the leather coat, keeps repeating “Buy more!Save more!” every time a new product is released on stage. As GPU chips become larger and larger, the memory chips equipped with them are getting larger and larger, the network communication capabilities between chips are constantly increasing, and the scale of large models is also increasing.

The huge investment in exchange for the growth of the model scale has made the large AI model can only be placed in the cloud in the training stage, but with the passage of time, the growth of the large model capacity in the cloud has begun to slow down (GPT4 has nearly 1,000 times the model size has increased compared with GPT3).

This marks that, like the previous waves of AI, large AI models with deep learning as the core have begun to touch the “ceiling” of their capabilities. Once the growth of the industry’s most powerful model capability stagnates, the whole industry will move to the stage of practical application.

However, for now, the AI large model ecosystem led by OpenAI is still making slow progress in promoting the landing of applications for the majority of users. The key to this phenomenon lies in the “unwillingness” of users to put their own demand data and experience data into the public model, which may be for business value, or because of the importance and concern of personal and organizational privacy.

In April 2023, there was a news in South Korea: Samsung employees inadvertently leaked the company’s top-secret data while using ChatGPT to handle work. The specific process is that the engineers of Samsung’s semiconductor division chose to use ChatGPT to fix code problems, and as a result, some employees threw the source code of the new mobile phone program and some internal meeting minutes data to ChatGPT, which is equivalent to trade secrets falling into the hands of OpenAI.

Samsung’s solution was to retrain a large model and deploy it in a private cloud.

In China, there was a similar news not long ago - the artificial intelligence office application “WPS AI” under Kingsoft Office has opened public beta, and the AI function has also been opened to all users for experience. But soon, some users found that their “Privacy Policy” directly clearly stated: “We will use the documents and materials you upload on your own initiative as the basic materials for AI training after desensitization.”

Although Kingsoft Office later apologized and promised not to use user data for training, many users publicly said that they did not dare to use it again.

The only way to solve this contradiction is to provide users with a solution to privatize the capabilities of the large model.

Yang Yuanqing, Chairman and CEO of Lenovo Group

On November 21, 2023, at the “Caijing Annual Conference 2024: Forecasting and Strategy”, Yang Yuanqing, chairman and CEO of Lenovo Group, expressed his opinion, that is, "At present, the user scale of large models is still relatively small, and most large models are trained on public clouds with strong computing power. In the future, with the expansion of the scale of users, whether it is for the sake of data security and privacy protection, or for the consideration of higher efficiency and lower cost to respond to user needs, the computing load of large models will gradually sink from the cloud to the edge and device side, and more and more artificial intelligence inference tasks will be carried out at the edge and device side, which makes personal large models more necessary and possible. ”

To put it bluntly, it is to let everyone have enough computing power and resources to run the inference link of AI large models as much as possible. In this way, the local device can input local data and calculate the corresponding large model results immediately.

The device that can take on this role can only be the PC - the new type of PC of the future, that is, the new species of AIGC-driven - AI PC.

Whether it’s a smartphone in your pocket, a laptop in your backpack, or a desktop at home, it’s the result of decades of careful work across the semiconductor and IT industries. The application of AI may change the way humans interact with these PCs, and even partially change their form, but it will not directly eliminate these critical computing nodes.

AI PC: The Renaissance Moment of the PC

Looking back at the history of computer development visible to the naked eye, the emergence of the PC (personal computer) not only brought choices for personal use of computers, but also laid the “unspoken rules” of the computer industry to benefit mankind.

In the wave of Internet and communication technology that followed, consumer electronic devices, including smartphones with increasingly sophisticated functions, were aimed at the greatest number of consumers, allowing everyone to fully enjoy the right to access information anytime, anywhere.

This wave of AI large model technology is coming, and it is still PC that can shoulder the burden of benefiting mankind - at least at the beginning of the stage, and the time of this stage will be accompanied by the real popularization of AIGC, not one or two years.

No contrast, no harm.

As we mentioned earlier, the “new hardware” AI Pin reflects the ambition of OpenAI’s “creator” to create a ChatGPT twin hardware: if it wants to work properly, it must remain connected to the Internet and constantly upload microphone and camera data on the user side. No matter how you look at it, it doesn’t meet the demands of today’s users for mobile devices such as strong productivity, fast response, and privacy protection. The only explanation is that Humane wants to circumvent the two more mature ecological platforms of iOS and Android, and directly build OpenAI into a “new” platform.

Compared with PCs, AI Pin is highly dependent on OpenAI’s online services, but the interaction methods are extremely limited, there is no open ecosystem, and there is no way to even operate independently in a state of privacy protection.

This is the background to the birth of “AI PC”, and it is also the problem that this old and young “new species” has to solve from the beginning.

At the annual meeting of Caijing, Yang Yuanqing defined the five characteristics that AI PCs need to have:

• AI PCs are capable of running large individual models that are compressed and optimized for performance.
• AI PCs have stronger computing power and can support heterogeneous computing, including CPUs, GPUs, and NPUs.
• AI PCs have larger storage, which can hold the data of the whole life cycle of individuals and form a personal knowledge base, providing fuel for the learning, training, inference, and optimization of personal large models.
• AI PC has a smoother natural language interaction, and can even interact with it with voice and gestures.
• AI PCs have more reliable security and privacy protection.

This is the most complete and clear definition of AI PC that we can see so far, and it is not difficult to smell the smell of “both want and want”. It is “radical” enough in the pursuit of heterogeneous computing and natural language interaction, “conservative” in the protection of security and privacy to meet human needs, and in terms of compression, performance optimization, and more powerful storage to support the construction of personal knowledge bases, it is very technologically neutral and humanistic.

In his speech, Yang Yuanqing gave a more precise positioning for AI PC: “It can not only serve as the entrance to the public large model, but also run a personalized private large model independently, it can grasp the most comprehensive personal data and information, and can strictly guard your secrets.” Only you can wake it up, use it, and at the same time, only it understands you best, far better than the public model. ”

It’s not easy to do that. Let’s take a look at some of the current practices:

In order to allow everyone to have their own personal large model, Lenovo has set the AI PC to be able to call the local + cloud hybrid large model and personalized knowledge base, truly combining the value of public data and the value of personal data. In this way, users can not only use the huge knowledge base of the public model, but also can confidently store personal data for local computing, fully adapting to personal needs and scenarios.

In terms of natural language interaction, Lenovo emphasizes the need to actively retrain local data in the interaction process to gradually improve the intelligent and personalized performance in the interaction process. Provide users with AI capabilities that can grow sustainably, rather than just starting out.

Another key is the open ecosystem, in the past, PC products, core computing chips have clear industry standards, according to the developer tools to write software to ensure the operation effect. In the era of inclusive AI models, application developers and model service providers have put forward higher requirements for basic terminal equipment, and terminal manufacturers must also participate in the operation of the core ecosystem.

Finally, privacy and security, AI PCs have a special privacy and data protection security system compared with public large models, which has a better guarantee for the core security requirements of individuals and enterprises.

Of course, the AI PC wasn’t built overnight, it has an evolutionary path that was once primitive, but gradually took shape, and eventually embraced the wave of generative AI.

In 2018, Lenovo has already proposed the concept of “smart PC”. Subsequently, the AI element was gradually expanded to the full range of PC product lines such as Lenovo Yoga, Legion, ThinkBook, and commercial ThinkPad. Although its “intelligence” is not today’s “intelligence”, it goes without saying that the trigger for the full rise of AI PCs is AI large models, but it is inconceivable to launch an epoch-making AI PC prototype in less than a year without AI exploring the future boundaries and possibilities of PCs.

The “water makers” of AI computing power, Nvidia, Intel, AMD and Microsoft, all have deep cooperative relationships with Lenovo. In the process of building AI PCs in the future, it is indispensable to fully cover a variety of systems and architectures such as Windows, Android, x86 and ARM, and fully meet the needs of various scenarios and users. And these layouts and foreshadowing are gradually reflected in the financial data.

In Lenovo’s financial report for the second quarter of 2024, the group’s revenue increased quarter-on-quarter for two consecutive quarters, reaching 104.4 billion yuan, and the gross profit margin increased to 17.5% year-on-year, a record high in the second quarter. Non-PC business accounted for more than 4 percent, an increase of nearly 3 percentage points year-on-year.

The momentum for the recovery comes from the recovery of the current consumer electronics market. But Lenovo has higher expectations for 2024, especially the upcoming explosion of AI terminals. According to IDC data, AI terminals will account for 41% of the Chinese market in 2023, and this proportion will reach 85% by 2027. At the earnings conference, Yang Yuanqing also mentioned that AI PC will further promote the profit improvement of Lenovo’s main business in the future.

Obviously, Lenovo, which deployed AI PCs early and took the ability of inclusive AI large models as one of its goals, is expected to play an extremely important role in the popularization of global AI terminals in the future.

It’s not just a company’s ambition. It’s time to revisit Bill Gates’ vision of 48 years ago – “a computer in every home”. This sentence should be upgraded to “let every family have an AI PC”, and this should be a new starting point for human beings to create more possibilities.

This is also the rebirth of the ancient species of PC in the name of AI PC, and it is a renaissance.