In the first three quarters of this year, global PC shipments witnessed a long-awaited consecutive year-on-year growth, with the second quarter recording the fastest growth rate at 3%. Mainstream PC manufacturers generally achieved higher business growth rates than in previous years, with Lenovo, for example, reporting a revenue growth rate of 24% in the second quarter. These data undoubtedly challenge the long-held "mainstream narrative": in today's deeply developed mobile internet era, PCs will eventually fade under the impact of smartphones and tablets. However, the fact we see now is that PCs are rising, and the core driving force behind this rise is AI.

According to estimates by consulting firm Canalys, AI PCs have accounted for 19% of global PC shipments this year and are expected to grow to 60% by 2027. Some PC manufacturers are even more optimistic, predicting that AI PCs will account for 80% of the global PC market by 2027 at the latest. By then, most mainstream PCs on the market will possess certain edge AI capabilities, transforming PCs from a dormant hardware platform into a core platform for using AI applications, especially productivity applications. Such dramatic changes were unimaginable just a year or two ago; how quickly the world is changing!

Therefore, we can understand why Quark launched its PC application in August this year and invested significant resources in it. The Quark PC application achieved success within just three months of its launch. According to PConline data, since the beginning of this year, Quark PC has ranked first in domestic PC AI application downloads, almost equivalent to the combined total of the second and third places. Moreover, among Quark PC users, those under 25 years old account for more than 50%, making it currently the most popular AI application on PC among young people.

In the past two days, I have been avidly trying out the Quark PC application. It is currently the AI application I use most on my computer, excluding mobile devices (not counting mobile usage). From the perspective of an ordinary user, my experience with it can be summarized in three main aspects:

It offers a comprehensive range of functions, including AI search, AI writing, AI document summarization, and more, seamlessly integrated with commonly used cloud storage and scanning tools to form an all-encompassing AI application. Quark claims to provide a "one-stop AI service," and after using it, I found that this claim is a firm commitment rather than a gimmicky hype.

It covers all scenarios and provides a smooth user experience. For example, while writing this article in Word, I can select a paragraph of text at any time and summon Quark to perform AI search, translation, or polishing through text highlighting. Quark can also be summoned using shortcuts in most applications and web pages. It feels like Quark is not just a standalone application but an integral part of the PC system.

It seamlessly connects the PC and mobile platforms. Under the same account, all user records and data are seamlessly synchronized! For working professionals who need to work anytime, anywhere, synchronization of data between PC and mobile platforms can be a nightmare, wasting considerable time and causing numerous issues. The feature of cross-platform data synchronization is a timely assistance, and it is easy to imagine how welcomed it will be by working professionals!

With these characteristics, Quark is well-positioned to become the "first stop" for young users seeking AI applications, providing them with comprehensive, low-threshold, and all-encompassing generative AI services. Just now, I conducted a test using the Quark PC application: assuming someone with no knowledge of the technical principles of generative AI needs to summarize a knowledge document on this topic and then create a PPT that can be shared with others, can Quark help them do it? And how long will it take?

The conclusion is that not only can it be done, but it can be done efficiently. I first typed "technical principles of large models" into the search box and specified the "academic search" function. After summarizing 34 academic papers from authoritative institutions such as CNKI, VIP, and Wanfang, Quark provided an academic response of about 1,200 words, with the first paragraph stating, "The core technology of large models is the Transformer architecture." The following four paragraphs explained how large models based on the Transformer architecture are pre-trained and fine-tuned, how they possess reasoning abilities, and how they are optimized at different levels.

Great, but let's assume I'm a tech novice and my next question would be, "What is the Transformer architecture?" This time, Quark's academic search summarized 27 papers and provided a response of about 900 words, clearly stating that the Transformer architecture consists of an encoder and a decoder, with the self-attention mechanism as its core component. Quark also correctly pointed out that the Transformer architecture was proposed by scientists at Google in a 2017 paper titled "Attention Is All You Need." This response cited a total of 11 academic papers, each of which can be viewed by clicking on the annotations for comparison with the original text – so you don't have to worry about the AI "talking nonsense" or fabricating arguments, as you might with other AI applications!

If you find the thousand-word "academic response" too complex, Quark also generates a more concise "academic outline," distilling complex long texts into a series of single-sentence points organized into multiple levels. For example, the characteristics of the Transformer architecture are divided into multiple points such as "Overview," "Self-Attention Mechanism," "Encoder-Decoder Architecture," with the "Self-Attention Mechanism" further outlining its implementation methods and technical characteristics. If I am completely unfamiliar with the topic, this outline will be a good starting point for learning; if I am already familiar with the topic, it can guide me to review and reconsider, as well as find more high-quality references.

Next, I generated a PPT of about 15 pages from this outline with one click. In fact, the system allows me to generate PPTs of various lengths, up to over 40 pages; styles, colors, and templates can also be customized as desired, as Quark Docs already boasts powerful PPT template capabilities. Below is one page of the PPT I generated. Of course, I can also generate a PPT based on a Word document I have written without relying on Quark's academic search results – a feature I am sure to use in the future.

The waiting time for the above steps is negligible: a single academic search takes only a few seconds to a few dozen seconds to return results, and generating a PPT takes a bit longer but generally does not exceed one minute. If someone uses Quark for learning and document organization, they can maximize their time spent on learning itself. As the saying goes, "A sharp knife never blunts the woodcutter's work," and here, the knife is already sharpened, ready for chopping wood.

I can imagine the tremendous productivity value Quark PC offers for young users – from previewing and reviewing course materials, class presentations, and literature review organization for college students to weekly and monthly reports, speechwriting, and internal and external information analysis for working professionals. When repetitive tasks are handled by AI, people can focus their limited energy on areas requiring creativity and judgment. Therefore, I can fully understand why Quark PC has accumulated so many users within just three months of its launch; I believe this is just the beginning.

The above discussions focus on the application itself. Let's take a higher perspective and look at more macro and strategic factors. To be honest, while the concept of AI PCs is currently hot, it is still very immature and can even be considered a "trendy buzzword" to some extent. Major PC manufacturers have simply integrated graphics cards with local inference capabilities into PCs but have not provided a software application environment to utilize this inference capability (with Apple perhaps being the exception). Moreover, let's not forget that AI PCs only accounted for 19% of PC shipments this year and likely only a single-digit percentage of the vast existing PC stock. So, the question arises: faced with a complex hardware environment and an almost non-existent software environment, how can users, especially non-AI professionals, efficiently use AI applications on PCs?

My view is clear: under current conditions, most users prefer a "user-friendly" and "one-stop" comprehensive application to meet their generative AI needs on PCs. This application must be "system-level," convenient and easy to use, integrating multiple commonly used functions; it must possess a powerful base large model as the foundation for providing high-quality AI services; it should be cross-platform to adapt to users' information synchronization needs; and it must be compatible with all mainstream PC hardware and operating system environments. More importantly, before the full popularization of edge AI inference capabilities, most of the inference demands of this application should be handled on the server side to smoothly and equally address the inference needs of all users, which places high demands on its AI computing power reserves and technical infrastructure capabilities!

Obviously, no one is more suitable for this role than Quark. Its generative AI capabilities are built on self-developed large models, including the Quark large model launched over a year ago and the "Lingzhi" learning large model introduced in November this year, which is optimized for learning scenarios. In the AI era, the importance of technical infrastructure cannot be overstated!

What is most noteworthy is Quark's product innovation capability. Over the years, Quark has evolved from a search tool into a comprehensive productivity tool encompassing search, documentation, cloud storage, and more functions; entering the era of generative AI, it has quickly transformed into an all-encompassing "AI all-rounder." In this process, it has learned to focus on user needs, pragmatically addressing user pain points, especially those of young users! This is Quark's product culture and the fundamental reason for its strong foothold among young people. Even more than two years ago, before the era of generative AI arrived, more than half of Quark APP's users were already young people under 25 years old.

This is precisely the foundation for Quark's transformation into an "AI all-rounder": I still remember how enthusiastic my post-95 and post-00 friends were in my social circle when ChatGPT emerged, actively trying to use it to transform their workflow, with some even asking AI to help write their weekly reports in the first month! In contrast, the "middle-aged" post-80s generation (including myself, who considers myself young at heart) were much slower to adapt. Many of them became core users of Quark APP's generative AI features; when they need to use PCs for learning and work, they naturally think of Quark, which they have been using for a long time on mobile devices. This is a significant reason for the rapid rise of Quark PC and something that competitors find hard to replicate.

The application of AI on PCs is only just beginning, and it may take more than three years before we can see the full picture of this blue ocean. Throughout its development, Quark has repeatedly proven itself as an "undefined" application product without any baggage or restrictions, making it highly suitable for exploring this blue ocean. I am eager to see if Quark will venture into edge inference or even collaborate with hardware manufacturers to launch new solutions after the full popularization of AI PCs – anything is possible. That will be a brand-new era where AI application services based on both cloud and local computing power will reach depths and efficiencies unimaginable to us today, just as the mobile internet of today was unimaginable to people over two decades ago. Of course, that is a topic for another discussion. Just thinking about it is exciting!