Where Are the Entrepreneurial Challenges and Investment Opportunities in AIGC?
-
At the end of August, the 'Model Debate: Large Model Industry Ecosystem Forum: Boiling Capital, Riding the AGI Wave' hosted by Future Technology Power was held in Shanghai. We discussed the iPhone moment of large models, the explosion of vertical large models, the industrial application of AIGC, and computing architecture design with over 20 industry leaders, academics, and investors from companies like Cheetah Mobile, Analysys, Inspur Information, NetEase Fuxi, IDEA Research Institute, and Emotibot. The event also attracted over 400 industry professionals and media representatives. As a globally oriented technology content platform and innovation connector, we remain committed to closely following the development of China's large model industry, firmly believing that China's AI sector will ultimately carve out its own path of innovation and breakthroughs.
Image credit: Generated by AI, image licensed by Midjourney.
Below is the transcript of the roundtable discussion 'Entrepreneurial Challenges and Investment Opportunities in AIGC':
Discussion Topic: Entrepreneurial Challenges and Investment Opportunities in AIGC
Participants:
- Wang Zhaoyang, Head of Content Center, Future Technology Power
- Bai Zeren, Vice President, Linear Capital
Wang Zhiwu, Founder & CEO of Yuanjing Technology
Zhong Guangqing, Co-founder and CPO of Fabarta
Qiu Yiwu, Founder of Zaowu Cloud
Wang Zhaoyang: Could each of you share your experiences within this wave over the past 8-9 months based on your companies? If we were to divide this into stages, what phase do you think we're in now?
Bai Zeren: Linear Capital is an early-stage investment firm based in Shanghai. Since our establishment in 2014, we've focused on data intelligence and cutting-edge technology investments, particularly AI applications in vertical sectors like industry, construction, and even agriculture. We've invested in many AI-related companies and have always maintained close attention on AI developments.
When ChatGPT emerged around late December last year, we were pleasantly surprised. One of our portfolio companies, Mindverse Universe, developed a cognitive model aiming to provide more intelligent and human-like AI services. At that time, our investment team - all with technical backgrounds - engaged in discussions with their founders, primarily exploring ChatGPT's capabilities and future potential out of technical curiosity.
We recognized early on that ChatGPT represented something fundamentally different from traditional AI. At the beginning of this year, we decided to focus our investment strategy in this direction, having reviewed over 200 projects from both domestic and overseas Chinese teams. Over these past eight months, we've witnessed continuous monthly innovations in AI. While eight months is relatively short, if we must define stages, the period before Chinese New Year marked one phase when most viewed ChatGPT primarily as an AI chatbot.
ChatGPT has been widely accepted for its generative, reasoning, and execution capabilities. With the emergence of more APIs and frameworks, it can now handle more complex tasks. In the entrepreneurial circle, the initial focus was on building large models, but now the trend is shifting towards applying these models in vertical industries to address efficiency and cost issues. However, the current stage is still relatively early. Having worked in AI at Tencent from 2017 to 2019, I see this as the early phase of the second wave of AI, following the deep neural network era.
Wang Zhaoyang: The past eight months have been quite dynamic, but it's still hard to categorize things clearly. For example, applications are evolving one after another. I'm curious about the industry's or entrepreneurs' perspectives. Has your understanding of the technology changed, and is it in sync with investors' views, or are there trends you see that haven't been fully realized domestically?
Bai Zeren: For the most part, our understanding is aligned. We gather insights from active discussions with entrepreneurs and our own investment experience. We've invested in many AI applications for industries, focusing on the value they bring. Based on this and our understanding of AI's current capabilities, we're optimistic about AI's potential in vertical sectors like industrial and healthcare fields.
However, there are still few related projects in the market. We believe the adoption of vertical applications will be gradual, involving market education and industry awareness of how AI can solve real problems. Unlike the hype around Web3 and the metaverse, this wave of AI is more sustainable and long-term.
Wang Zhaoyang: There's no excessive excitement; we understand that some things take time to integrate into industries.
Bai Zeran: Indeed, we've been observing AI for quite some time. Looking at traditional AI today, we feel both excited and maintain a rational perspective about this wave of innovation.
Wang Zhaoyang: Could Mr. Wang share some key milestones or financing events at this stage? This would help everyone understand what specific companies are working on.
Wang Zhiwu: Thank you for the invitation. We specialize in virtual digital humans and have received strategic investment from Tianyu Digital Technology. I've been in this industry since 2016, starting relatively early with large models. We've gradually categorized the development of virtual humans into several stages.
There are several stages: The first is the static virtual stage, where virtual entities exist as static images and text - an early form of virtual representation that we've evolved into more intelligent forms.
The second stage features video-presented virtual humans with relatively low precision that require real human drivers and video presentation. However, this year, after working with all major models and training vertical knowledge, we've found virtual humans can enter the third stage - AI virtual humans.
AI virtual humans represent an early implementation direction. We've made many attempts in both virtual human and digital human tracks, implementing them in vertical industries. Our current virtual humans are mainly applied in live commerce, healthcare, finance, education, and other fields. Additionally, we've developed concepts like smart receptionists, intelligent customer service, and intelligent clients.
I believe virtual humans may take the form of digital employees, digital labor, or digital companions in the future. We are particularly optimistic about their B2B applications, but in the long run, they will inevitably move toward the C2C market, which we are currently exploring. Finding ways to integrate virtual humans with the consumer market is our most important mission for the second half of this year—identifying practical applications for the C2C sector.
The first scenario we are exploring is elderly companionship. Aging is indeed a significant issue, and we aim to integrate virtual humans into this industry. We summarize virtual humans with five key attributes: memory, a soul (their expressions and communication abilities must closely resemble a human's), an engaging personality beyond just an attractive appearance, rich emotional expression, multi-sensory interaction (not just text but also image, video, and voice chat), and hyper-realism. These are the five areas we are focusing on in the second half of the year.
Wang Zhaoyang: The second stage emphasizes virtuality, while the third stage emphasizes humanity. Our roundtable discussion is closely tied to business. I’m curious whether the current applications rely more on mature second-stage technology or if breakthroughs in technology are driving new opportunities. Are these scenarios emerging because of technological advancements?
Wang Zhiwu: It’s actually driven by technology. Previously, we used LP technology for AI virtual humans, but they appeared stiff and ineffective in conversation, mostly used for entertainment. However, with the advent of large models, virtual humans can now behave more like real people. Technology has pushed us forward—without large models, second-stage virtual humans might still dominate.
Wang Zhaoyang: Next, let’s hear from Mr. Zhong about the infrastructure side. Initially, people thought of chatbots, but it turns out there’s much more beneath the surface. Companies involved in infrastructure are gaining attention, and investment trends are shifting in this direction.
Zhong Guangqing: Fabarta is an AI infrastructure company. The emergence of large models has formed an industrial chain with various roles. At the source are companies providing large models, while developers are at the terminal. In between, various ecosystem players focus on industry-specific models or helping large models integrate into business operations. Regardless of the application form, large models require infrastructure support.
For example, topics like vectors and multimodal capabilities are widely discussed. Fabarta offers a "one-core, two-wings" product system. The core is a multimodal intelligent engine, serving as foundational infrastructure for large model data support. When deploying large models, practical issues like optimizing inference capabilities arise. Using open-source models directly can lead to high GPU resource consumption, which we have optimized.
The "two wings" refer to data and AI. From an ecosystem perspective, our data-side products help users manage and inventory data assets intelligently, enabling data flow within enterprises. With data, we connect to the "AI wing," aiming to help clients achieve data circulation and form a closed loop, allowing them to effectively apply large models. As the era of large models arrives, such infrastructure support is essential for enterprises to implement and utilize these models.
Wang Zhaoyang: Could you share insights into current customer demands?
Zhong Guangqing: Since we primarily serve B2B clients, our customers are mainly from industries like finance and manufacturing. Demand is strong, as businesses seek to transform operations using large models. Through discussions with IT and data management teams, we see a focus on leveraging data effectively and integrating it with business. Market opportunities fall into two categories: content generation (e.g., intelligent assistants) and "decision intelligence," a core concern for enterprises beyond text or image generation.
Currently, large models still face some issues in practical applications, especially as many Chinese companies cannot access foreign models like GPT-3.5 or GPT-4. We have tested several large language models and found a significant performance gap compared to GPT-3.5, including in terms of emergent intelligence. When a model's capabilities are insufficient, such gaps become evident. However, customer demand remains strong, and we are now entering deeper waters. In the past, companies created many demos, but these were simplistic. When it comes to actual enterprise implementation, the requirements are much higher—there can be no errors, especially when enterprises demand higher precision. We are addressing these challenges by leveraging our team's core strengths, including our foundation in graph intelligence, to further integrate large models. Currently, there is active discussion in both industry and academia, and we are actively participating.
Wang Zhaoyang: Professor Qiu, Zaowuyun and your background stand out on this stage. As a serial entrepreneur, what are your reflections or insights about your company over the past year?
Qiu Yiwu: We are an angel project. Last year, I described us as an industrial version of Cool+. After graduating in 2013, I’ve been continuously entrepreneurial. My first company was Yunzao, where I explored how internet technologies could transform manufacturing. Over the past decade, before founding Zaowuyun, I worked in smart hardware, covering design, R&D, branding, sales, and even running factories. The motivation behind Zaowuyun was a collaboration with Zhejiang University professors. Zhejiang University’s design program is unique globally, housed within the computer science department. In the 1970s, President Pan worked on AIGC; in the 1990s, it was computer-aided design. Now, we’ve entered the era of AI-assisted product innovation.
Zaowuyun was founded with a sense of mission—to transform Zhejiang University’s technological and design innovation capabilities into a broader platform for manufacturing and branding enterprises, empowering them with product innovation capabilities. The idea originated in 2018 when new AI technologies emerged. Several professors called me, noting how this generation of AI could revolutionize design. We spent two years developing the first demo. Initially, we used GAN, but the results were too poor for enterprise-level applications. We realized that accelerating product development inevitably involves 3D, so we focused on 3D solutions first. When diffusion models emerged last year, they enabled illustration generation. This year, the realism of visual models has astonished everyone.
Since March, we have fully embraced AI. At Zaowu Cloud, we are truly leveraging AI's capabilities to bridge the gap from C to M (Consumer to Manufacturer). I've noticed an interesting phenomenon: industries that were not previously industrialized are now being AI-ized earlier. For example, in handmade crafts, jewelry, accessories, and home furnishings, as long as AI can generate creative designs, craftsmen can produce the items based on those designs, achieving a semi-standardized industrial process. It's like skipping the train era and directly entering the high-speed era. We hope that through this approach, more consumer brands or individuals in the next era can leverage our technological capabilities and platform to create better products.
Wang Zhaoyang: I’ve noticed two common points among everyone here. First, before this wave of technology arrived, each of our companies or startups was already deeply embedded in the business environment with some existing technologies. Now, this new technology has arrived. Second, each of us has long-term experience in what we’re doing, but we’ve suddenly realized that the needs and demands of those we serve have become more advanced and aggressive than we anticipated. These two points seem contradictory. So, the second question to discuss is: From your respective perspectives, especially after the slight cooling down over the past month or two, how different is this wave of technology from the technological routes you’ve previously followed or are currently exploring, from a business-centric viewpoint?
Qiu Yiwu: Let me give an example. Our background is primarily in graphics, which is also a strong area at Zhejiang University. When GPT first emerged, I instinctively felt it wasn’t very relevant to our work in design. When it debuted last December, I thought it was just a slightly smarter version of Xiaomi’s voice assistant, with no significant impact. But over the past six months, we’ve come to realize that it’s a foundational, general-purpose model that can connect many elements. Digital humans, for instance, serve as carriers—they incorporate design knowledge, marketing knowledge, and engineering knowledge. After some reflection, we believe our future design platform will inevitably combine the capabilities of language models with visual design models. This is what this generation of general-purpose models brings to us.
Previously, working with NLP (Natural Language Processing) felt like a distant goal. But with these tools now available, the barrier to entry is much lower, allowing us to better harness their potential. In the future, combining these elements will create different "small universes" tailored to solve specific business scenarios. Regardless of the tools used, the goal is to meet the demand. For example, in our case, when a design company proposes solutions to clients, it’s not enough to just provide a few images. We also need design strategies, explanations, user personas, and design descriptions—all presented in a multimodal way. The advantage of this generation of AI is that it levels the playing field by simplifying many foundational processes.
Wang Zhaoyang: The general-purpose nature of these tools not only lowers the usability threshold, making them accessible, but also introduces entirely new possibilities—tools that didn’t exist before are now available. Mr. Zhong, please share your thoughts as well.
Zhong Guangqing: From a commercial perspective, I believe large models or large language models (LLMs) differ significantly from past technologies. We can evaluate the impact of a technology on the business ecosystem from two angles. First, from the standpoint of the business ecosystem or industrial chain, we assess whether the introduction of this technology brings new changes. The second angle is whether it disrupts traditional business models or enables capabilities that were previously unattainable.
From the business ecosystem perspective, we can already see the formation of a relatively complete industrial chain around large models. Although this chain is still evolving, it has already taken shape. From the viewpoint of traditional business and transformative development, many industry experts emphasize that numerous application areas are ripe for transformation through large models.
One familiar example is intelligent customer service. In 2015, I developed an intelligent customer service system for a company and encountered a problem: users questioned why the system provided the same response regardless of the query. This made the system seem ineffective, as it couldn't offer personalized or precise answers. While this issue has largely been resolved today, other challenges remain.
Another unmet need at the time was in auditing services. Many large companies providing auditing services wanted to implement intelligent auditing. Auditing is a highly demanding task, requiring the review of corporate contracts and documents against various rules. With the advent of large models, generating such analyses is no longer the biggest challenge. I believe LLMs excel in these tasks due to two core capabilities.
The first core capability is their exceptional generalization ability. The second is their outstanding natural language generation. From a linguistic perspective, large language models can replace much of the work done in the past, as language is a fundamental tool for human interaction, enabling LLMs to tackle diverse tasks. Based on this assessment, I believe current large language models are indeed distinct from many traditional technologies. Of course, they also face numerous challenges as we delve deeper into their applications.
Wang Zhaoyang: Everyone is discussing different directions and characteristics of technology. From a commercial perspective, considering both enterprises and clients, I feel that this technology, compared to traditional AI technologies, is inherently closer to the client side and better meets their needs. From the perspective of technology providers, solving problems that couldn't be addressed with current technologies—these are the most basic needs that this technology can now fulfill. I believe this is more beneficial for commercialization because it directly addresses the most fundamental perceptions and needs of clients.
Zhong Guangqing: In the past, many enterprises weren't even mature in IT technology. Now, with big data, business operations have evolved. Enterprises can mine insights from operational data to support decision-making. Previously, solving such problems required lengthy development processes. Today, with relatively standardized data, as long as the data quality is good, new business models can be quickly generated, allowing for trial and error and iteration. Most importantly, this transformation can grow alongside the business, which is a major concern for many enterprises given the rapid pace of business development.
Wang Zhiwu: I have deep feelings about this. Before interacting with clients, especially large B2B clients, we used to rely on large teams—sometimes up to 200 people—to handle projects, including development and internal support. It was exhausting. After each project, we conducted reviews and identified key issues. One major challenge was accurately capturing client needs. For example, our designers once spent two weeks on a draft, only for the client to say it wasn’t what they wanted. When AI-generated images emerged, we tried it—generating four options in 10 seconds. The client immediately chose one and said, 'This is exactly what I want.' That moment was eye-opening.
Another instance was when we used ChatGPT to draft a proposal on the spot. The client approved it immediately, and our business team found themselves doubling as strategists. Additionally, our developers, who previously spent days or even weeks writing scripts or modular code—often encountering bugs—can now generate code in seconds using ChatGPT.
Our internal team is growing rapidly. Before coming to Shanghai, we were discussing our muscle system solutions, and I was amazed by how realistic the muscles looked. Our AI has replaceable capabilities at every stage. Last Saturday, we assigned a new task to the team: starting in September, everyone will be evaluated on their AI proficiency. Each team member must be familiar with AI tools. When serving external clients, we ensure they recognize our professionalism, and internally, we use AI to boost efficiency in process development.
Designers and developers are already facing significant job displacement. The skills you’re mastering now may soon become obsolete. I’ve been pondering a question: What form should the future take when we combine AI with virtual humans? The answer lies in a companion-like work assistant that helps handle tedious tasks. With AI capabilities, these virtual assistants can operate 24/7, efficiently managing numerous tasks in specialized fields. We’re already using such AI to reduce costs and improve efficiency, both internally and in client interactions.
AI and large models are permeating every industry. For virtual human companies like ours, this presents both a challenge and a greater opportunity. We aim to explore these opportunities by integrating AI-driven virtual humans into various vertical applications, making them truly helpful. Previously, virtual humans were identity-based, like Tianyu, China’s first cultural export virtual idol, who gained significant influence but had high production costs. Now, AI enables us to lower these costs and shift from identity-based virtual humans to service-oriented ones. These new virtual assistants lack personal attributes but excel in their roles, handling tasks efficiently. This transition is already underway, and we’ll soon see many new forms and scenarios emerge.
Wang Zhaoyang: This reminds us that we haven’t paid enough attention to this technology. From a commercial perspective—focusing on efficiency, cost, and client needs—the appeal of this technology is undeniable. However, it’s easy to overlook its immediate impact, creating a human illusion about its timeline. Whether it’s short-term or long-term, the answer is clear. Let’s also hear from Bai Zeren, who has previously shared his desire to dive deep into this technological revolution.
Baizeren: Originally, in NLP, we dealt with many subtasks, considering a more human-like language model as the Turing problem. But today, suddenly, everyone is using this capability without even mentioning the Turing test, as if we've naturally crossed that threshold. This is the most significant difference I've observed. Regarding the commercial aspect, several guests have already shared their frontline experiences. I'd like to offer a slightly more macro perspective. I want to discuss my thoughts from the angle of productivity and production relations. Many say this wave of AI represents the fourth industrial revolution because it meets the primary condition: it creates new production factors.
From the first industrial revolution with mechanization, to the second with electricity, the third with industrialization, digitization, and informatization, and now the fourth with general machine intelligence—machines now demonstrate scalable production capabilities. General intelligence today is sufficiently capable of handling many complex tasks. Previously, we viewed AI more in terms of the 'AI+' concept when it came to implementation, using AI's capabilities—then referred to as small models—to solve vertical industry problems. We had to collect data, label it, build models, and go through lengthy processes, often starting with tasks that weren't perfect but could surpass humans in simple tasks.
Today's AI starts with powerful capabilities. We enhance model performance by aligning them with domain experts to solve sufficiently complex professional problems. When such knowledge-intensive tasks, once considered scarce, can be addressed by AI, it inevitably brings significant industrial opportunities. This wave is very different. For example, take the hardware industrial design scenario: appearance design, functional design, simulation design, and modeling. The existence of so many steps elongates the entropy-increasing process, greatly reducing efficiency in meeting demand-side needs. AI can streamline this process, bringing it closer to the demand side, greatly unleashing demand-side potential, and possibly even altering industrial structures. This is my brief macro perspective on the most significant difference brought by this wave of AI.
Wang Zhaoyang: Your entire logic still reflects linear characteristics. You understand this industry and its technological foundations. One question: Under the logic you just mentioned, how has your own linear approach and past investment strategies been impacted? Have you made any adjustments to adapt?
Baizeren: I think there are both changes and constants. The constant is still based on the essence of business. As for the changes, if we're talking about impact, I believe the biggest impact is this wave of AI applications. From a technical perspective, people can quickly create previously unimaginable products through integration with GPT. This may bring two issues: first, everyone's starting line is now leveled. The previous small-model-based technology may no longer seem as important, allowing companies with strong data accumulation to move faster. This is the first issue. Second, certain directions may become highly competitive (red oceans), which might be the biggest impact for us.
But returning to the essence of business, there are still significant opportunities here. Although the technical barriers to entry are lower, as Zhong Guangqing mentioned, what needs to be addressed are its uncontrollability, compliance, and even underlying innovations, including computational efficiency and cost reduction. There are many engineering optimization issues, followed by productizing the technology. Currently, after seeing the effects of using AI capabilities, it becomes a product. Today, we still see many entrepreneurial teams applying AI to their vertical scenarios. We need to evaluate one thing: in my AI application project, is 80% of my capability derived from large models, or does the large model's capability create strong barriers in our work? Do we have a deeper understanding of the entire vertical workflow? Can we establish effective products in this work and form a closed loop? For certain experts in this workflow, such as in industrial design scenarios where different experts handle complex problems like simulation, it's not just about relying on large model capabilities. We must return to the essence of business—understanding the scenario and genuinely solving customer problems while continuously improving problem-solving efficiency.
Wang Zhaoyang: Very interesting. It's both a powerful technical capability and a threshold that pulls many people back, making the probability of entering a red ocean very short. So when people consider this an important business opportunity, it becomes a red ocean. Entering this stage, past accumulations still matter—it's not that past accumulations don't count. This is a very interesting phenomenon brought about by technological accumulation.
Bai Zeran: China's traditional industries are actually very large in scale. With the emergence of AI today, these industries have an opportunity to leapfrog development. They are eager for cost reduction and efficiency improvements, but currently, we see few entrepreneurs who truly understand what these industries are doing. However, the issues in these vertical industries, as mentioned earlier, could present significant opportunities if someone is willing to use AI to address long-chain problems and tap into the potential of demand-side needs. I strongly encourage everyone to explore vertical fields.
Wang Zhaoyang: I’d like to ask a question. There seems to be a noticeable gap when it comes to virtual humans. People often associate virtual humans with consumer-facing applications, but in reality, when we look at virtual human companies, their revenue or business models are still largely B2B-based. Is this a normal phenomenon or just a short-term trend? How will this develop in the future?
Wang Zhiwu: Currently, the entire virtual human industry faces high barriers in terms of production, technology, and application. These costs make it difficult for the consumer market to enter. After the open-sourcing of MMD models, some enthusiasts began sharing fan-made works on platforms like Bilibili or other communities, fostering collaborative creation—this was their initial vision.
Secondly, there’s the operational cost. Simply releasing virtual content without a structured plan makes it hard to build a sustainable system. Even if some traffic is generated, monetization remains challenging. For consumer users, it’s difficult to create real commercial scenarios or integrate with brands, which prioritize long-term value, design concepts, and cultural alignment. However, with recent advancements in open-source models and 3D technology, virtual humans are on the verge of entering the consumer market. What’s missing is the right opportunity.
The B2B market, on the other hand, has the financial resources and willingness to pay. Customized B2B demands will always exist, with even higher requirements for personalization. Serving these clients allows for the realization of design concepts. What we lack is a way to lower the barriers—such as industrialized or streamlined products—to make virtual humans accessible to everyone. Imagine virtual humans appearing on every desktop as companions, assistants, or other roles. This requires extensive technological integration, combining various technologies into a cohesive system.
We have developed a comprehensive training system with our own vertical models, which we trained ourselves, along with some open-source models that we fine-tuned. These are specialized vertical domain models. After training these vertical models with sufficient computational resources, our next step is to target specific consumer groups or vertical markets—not everyone will have access to them, as many lack the capability. We focus on creating knowledge-trained vertical models for certain user segments, with consumer applications being our technical priority this year. While results may not emerge this year, many companies are exploring this space, and we believe tangible outcomes will appear by next year. Consumer applications will likely start as free offerings, requiring significant preparation, which presents both pressure and challenges for us. Currently, our focus remains primarily on B2B.
Wang Zhaoyang: The consumer market is widely recognized as a major opportunity. We're not just waiting for it but actively preparing—whether in terms of scenarios or technological integration.
Wang Zhaoyang: Regarding technical questions for Mr. Zhong, the technology itself isn't entirely new but rather an integration of existing technologies. If we categorize by technical domains, there's graph computing, databases, etc. Recently, we've been examining graph computing as a concept—it combines with various technologies, making it hard to define clear boundaries for large models as purely technical. Could you share your insights? The core technologies we're seeing didn't emerge solely because of large models. These technologies are now being packaged as graph intelligence concepts or commercialized in new ways. What changes has the so-called 'large model' technology brought? Are there new opportunities, or are traditional technologies finding new applications?
Zhong Guangqing: Let me briefly explain graph technology. When we mention 'graph,' many people mistake it for images, but it actually refers to relationship network data. Why is this technology so popular? Because it enables deep data mining through network relationships. By analyzing network connections, we can derive group characteristics from individual attributes.
Graph technology covers many fields. Starting from graph technology, we've developed a graph intelligence engine that includes graph databases and graph computing engines. The graph computing engine uses iterative algorithms—like those depicted in the recent movie 'No More Bets,' where tracking transactions beyond the fifth layer becomes difficult. But we can monitor such data in real-time. We combine neural networks with graphs because graphs and AI are inherently related—graphs existed long before large models. After large models emerged, people began exploring their integration with graphs, as the primary issue with large models currently is their lack of interpretability, whereas graphs have well-defined network structures.
In the context of graph technology, knowledge graphs are more widely recognized—storing vast amounts of knowledge in a graph structure and solving problems by integrating knowledge graphs with large models. Combining graph technology with large models can significantly enhance the capabilities of these models. For example, graphs enable large models to better understand semantics and perform logical reasoning based on factual data. Beyond this, we are also working with multimodal data types. With the emergence of large model algorithms, there is a growing need for unified vector engine platforms. We can integrate deterministic graphs with fuzzy vectors to address this need.
In certain business scenarios, we have already implemented this approach in practice. These are some of the changes brought about by large models. From an entrepreneurial perspective, it’s essential to observe the transformations in your field of focus. In our domain, the advent of large models is accelerating the flow of enterprise data. Previously, businesses struggled with dynamic data intelligence, making tasks like generating BI reports cumbersome. While this issue has been addressed abroad for years, large models now allow us to quickly generate key metrics.
Another consideration is the assessment of AI applications and deployment methods. In the past, AI deployment was time-consuming and costly, often encountering issues that hindered timely responses to business needs. However, with the availability of advanced large models, we can now seamlessly link data with business operations, integrating data and models to maximize their utility in business contexts. To adapt to these changes, we have designed a 'one core, two wings' product matrix. For entrepreneurs, the key is to analyze industry-specific changes and determine the necessary adaptations—whether by riding the wave of innovation or outperforming previous efforts.
Wang Zhaoyang: This technology has revitalized data to some extent, indicating that there is still much to be uncovered or repurposed in data. This opportunity requires interpreting data relationships.
Zhong Guangqing: Indeed, data has been recognized as a production factor at the national level. While data governance is a compliance requirement, it was traditionally managed manually. However, if enterprises can effectively utilize data to create value, data can circulate efficiently. For instance, manufacturing industries, unlike banks, are not subject to stringent regulatory requirements but are actively managing and leveraging their data. In our work, we not only rely on multimodal intelligence but also aim to activate and utilize raw data.
Wang Zhaoyang: What you just said was quite interesting. Industries without high levels of industrialization are actually more willing to adopt new technologies or find it easier to transform. This is somewhat similar to the logic of China's mobile internet, which skipped the PC era and jumped directly to the internet. Regarding the company you're currently running, it seems to have some ambitious goals, such as designing platforms for this era. Is Zaowu Cloud aiming to become a company of the next era or at that scale?
Qiu Yiwu: Back in the PC era, software tools were the norm. Now, in the cloud era, the inherent structural properties of the cloud bring new dimensions to content. We’ve seen a lot of discussion about the AI era as well, where tools like Midjourney are being used for creative searches. However, many enterprises find that Midjourney’s solutions don’t quite fit because they can’t balance creativity with time constraints. This led us to rethink what the tools or platforms of the AI era should look like in China.
First, it’s quite challenging in the B2C space. Even in vertical industries, Midjourney can’t compete with us. For this industry, we design things that require models. Enterprises want options—they can choose between what we create with our models and what Midjourney produces, depending on what works best for them. As mentioned by other leaders earlier, in business scenarios, it’s not just about solving a single point problem. In China, we’re likely to see the emergence of industrial design platforms. Kujiale, for example, has created a collaborative network for the industry. This is the kind of software platform with Chinese characteristics, or the SaaS model, that will thrive. In specific industries, we’ll see platforms that combine AI capabilities, design tools, and solutions tailored to real-world applications.
Wang Zhaoyang: One final question, and this is the last segment of today’s discussion. Let’s step back a bit. We’ve talked a lot about the specifics of business and technology today, but there’s still a lot of room for imagination. The previous segment, where designers shared their insights, had that imaginative flavor. Setting aside your professional role, what do you envision for this technology in the more distant future? How might it impact your life, career, or company in its ultimate form?
Qiu Yiwu: I've been contemplating what remains unchanged. From a product perspective, I've been studying history—for instance, a purple clay teapot from 200 years ago is nearly identical to one today or even 200 years from now. What elements stay constant? Take Su embroidery, idioms, or symbolic meanings—these share commonalities across different cultures and arts. Auspicious motifs in jewelry, for example, are universally appreciated. Amidst endless variables, I focus on identifying immutable factors. We've trained over 200 Chinese craftsmanship techniques into models usable globally—these may become enduring constants despite future changes.
Zhong Guangqing: This morning I read a paper by Turing Award winner Bengio's team titled Consciousness in Artificial Intelligence. It argues current AI lacks consciousness, but achieving it is possible. If AI truly attains consciousness, humanity must learn to coexist with it, as depicted in sci-fi scenarios. However, humans possess greater wisdom—we can undoubtedly steer AI to better serve humanity.
Wang Zhiwu: I envision virtual futures where technology significantly replaces human labor. Jokingly, I told a friend I want to 'be useless'—to lounge while virtual beings handle work via eye-contact communication. Imagine two virtual assistants seamlessly collaborating, freeing humans for leisure. This utopian vision hopes virtual entities will genuinely reduce life's burdens.
Bai Zeren: Within R&D processes, I ponder how today's intellectual scarcity forces breakthroughs in fields like cancer treatment or material science to rely on few brilliant minds. Given AI's current capabilities, could it propel humanity past stubborn challenges—helping us transition from a Type I civilization to the next stage by solving intractable problems?