Hui Yuan is the founder and CEO of Xiao-I, which has been a leading artificial intelligence (AI) business for more than 20 years. In the recent past, AI has made some seemingly big leaps, with large language models (LLMs) like ChatGPT shaking up the status quo. By now, every industry is looking to benefit from this technology, and the global AI industry is already projected to reach nearly $2 trillion by 2030.
In this interview, Yuan talks about the recent IPO of Xiao-I, the rise of LLMs and how it came about, as well as the state of the AI industry as a whole. Additionally, he dives into the history of AI development, the deployment of Xiao-I's own LLM, called Hua Zang, and the future prospects of the AI industry.
Mr. Yuan, you established Xiao-I more than 20 years ago in 2001. Could you share your initial vision for the company at that time and elaborate on how it has evolved over the past two decades?
Hui Yuan: About twenty years ago, we aspired to establish an innovative enterprise that utilizes advanced technologies to create a better world. Back then, the realm of AI was an almost blank canvas, overshadowed by decades of disappointment following two failed attempts at artificial intelligence.
But we believed in the technology's potential, so we developed Xiao-I, the world's first business chatbot. We released it in 2004 and, from that moment on, it encompassed all the dreams, possibilities, and purpose we had long sought after. We aimed to serve and benefit humanity and the world through artificial intelligence, which it enabled us to do.
Over the past two decades, our initial intention has never wavered. But we have followed many visions, encountering countless challenges, disappointments, and successes in the process.
Just three years after the release of our Xiao-I chatbot, we became the exclusive partner of Microsoft's global 700 million-user platform. In 2008, we were the only artificial intelligence company in the world. We covered more than 200 countries and had over 100 million users globally. Whether it was an Italian-speaking chatbot in Italy or a French-speaking chatbot in France, they were all using the development toolkit provided by Xiao-I.
Back then, the world had a different view of us. We were not just seen as China's Google; we were the world's Xiao-I. However, the path of innovation was extraordinarily arduous for us. Although people always wanted to develop and utilize universal LLMs, it wasn't until now that these aspirations truly began to unfold.
While we faced countless challenges during our innovation journey, they ultimately led us to the wealth of core technologies we now possess. After years of accumulating data amidst the explosion of LLMs in the era of artificial intelligence, we are now able to wholeheartedly fulfill our initial intention. This is the foundation of our existence.
AI is divided into three components: perceptual, cognitive, and action intelligence. While the sensory and physical aspects are undoubtedly important, the most crucial part is cognition - the brain, so to speak. In this particular aspect, we have developed robust technologies and contributed to the formulation of nine domestic and international standards.
Xiao-I recently completed a highly successful IPO, raising $39 million with the backing of Alibaba Group Holding. What aspects of your company have made it particularly appealing to these investors in your opinion?
Yuan: Innovation is not about simply using or developing a new technology. Since a technology only creates value when it actually improves a product or service for users and companies, you need data to figure out how to best utilize it. Over the past 20 years, our clients and users have provided us with application scenarios, data, and a natural AI battlefield through their extensive use of AI products provided by Xiao-I, so we are very grateful for their continued support.
Today, there is an urgent need for speczialized LLMs, and Xiao-I has accumulated and developed profound algorithms, data, and application scenarios to offer just that. This is how we can now empower the entire global industry through the two pillars of technology and business with our Xiao-I LLM.
Xiao-I focuses on cognitive intelligence and operates in the specific field of business AI. It might seem like all the other companies in the sector are our competitors, but I don't think it actually makes sense to compete with other companies. Every company has a distinct DNA, so even if another company ventures into technologies related to cognitive AI, it will be completely different from Xiao-I. We also hope that our technology can create commercial value immediately upon deployment.
Our goal is to integrate AI technology into businesses' operational processes through techniques such as NLP, deep semantic interaction, speech recognition, and machine learning. In addition to intelligent customer service, the Xiao-I cloud offers various data integration platforms and intelligent bots.
With this holistic approach, we are gradually building a comprehensive intelligent system covering all bases, from service to management. Currently, we are enabling AI advancement by already covering over a dozen industry sectors, including finance, government affairs, retail, and education.
Additionally, just about every person and business around the world wants to enhance productivity, and technology companies provide ways to achieve that. We want to become a global and international technology powerhouse in the next 10 years. So, we are looking to share our advanced technologies, products, and solutions with the world. With people from different regions and countries around the world having a strong desire for technology that can improve their lives, we chose to list Xiao-I on the NASDAQ to maximize our impact.
The release of ChatGPT took both consumers and AI experts by surprise and created significant buzz in the industry. In your opinion, how impactful is this product for the AI industry as a whole?
Yuan: The technology behind cognitive intelligence actually evolves from generation to generation. First, it went from rule-based computing in the early years to expert systems. In the last five years, it changed to deep learning. Now, we are utilizing LLMs, and some of the core technologies being employed are rapidly iterating.
A robust AI system is not created overnight. In fact, arriving at the point we are at took 60 years of long-term investment and R&D in AI from multiple parties. It's an iterative process that depends on new insights and developments. All the currently used theoretical frameworks are built on those developed before them.
Additionally, a wide range of new integrations have been accomplished in the last 10 to 20 years, from data bandwidth and chips to computing power. From the outside, it might have seemed like ChatGPT came out of nowhere. However, in reality, it only came out now because all of the developments in the sector leading up to its release facilitated its creation. It is a process of gradual accumulation and fruition, and the necessary resources and elements are now more available, integrated, and concentrated. That being said, its release still shocked people due to its cognitive intelligence.
However, ChatGPT is not the end-all-be-all of AI, and it certainly isn't the AI singularity. ChatGPT is actually a conglomerate of "magnificent endeavors" that harnessed all available resources and stacked them on the existing foundation. It is similar to Siri or Xiao-I in the early years, leaving plenty of room for growth and future development. In that sense, not much has changed in the development of cognitive intelligence in the past six decades.
It might be enticing to think that stacking all the current knowledge together would leapfrog the current progress and pave the way for the future of AI. However, this is evidently not the case, as we have already witnessed the emergence of ChatGPT3.5 and ChatGPT4, with their staggering number of parameters at 175 billion and 100 trillion, respectively.
While the resources invested between the versions are incomparable in scale, the ultimate outcome fell short of the anticipated effect. All of this implies that artificial intelligence will evolve generation by generation, which will lead to a golden decade for AI in the future. However, there are still significant obstacles to overcome at the moment.
Xiao-I is preparing to launch its own LLM tailored specifically to the Chinese market. What motivated the development of this product, and what sets it apart from ChatGPT and other existing large language models in the market?
Yuan: The industry believes that the development of LLMs is an important trend in AI that can improve the accuracy, generalization, and efficiency of AI. This, in turn, drives the development of new AI applications. Currently, there are three critical challenges for LLMs:
The first challenge is the controllability. For example, OpenAI and ChatGPT are grappling with transforming from uncontrollable to controllable with respect to the ethical, moral, and ideological dimensions. To facilitate widespread adoption, the Hua Zang LLM that Xiao-I developed is controllable in all aspects, including its algorithms, ethics, legal values, and display content.
Without a controllable and secure large-scale model, it remains impossible for companies to deploy in commercial applications. History has already demonstrated this twice.
The second challenge is customizability. It's not enough for a model to possess the core capability of an LLM. It needs to meet user expectations. After all, what value does it hold if it doesn't generate the output the user desires?
However, since every user has different expectations, customization needs to be possible. Consequently, Hua Zang is fully customizable to fit the unique needs of each user. This allows our LLM to serve as an on-demand foundation for every scenario, every enterprise, and every government.
The third challenge is adoptability. The cost of LLMs today is exorbitant, particularly in terms of computational costs. OpenAI initially invested $1 billion, and now it requires additional tens of billions of dollars of investment to keep going. No one can sustain such expenses.
So, for the successful commercialization of LLMs, it is imperative to transform their development and implementation from being difficult to being feasible. To achieve this, Hua Zang has low training and computing costs. With Xiao-I solving these three core problems, we will usher in China's LLM era.
Who are the primary users of Xiao-I's large language model, and what are the key applications or purposes for which they will predominantly utilize it?
Yuan: With its flexible and customizable nature, the Xiao-I Hua Zang LLM can support businesses across a wide range of industries, such as finance, public services, medical services, construction, and tourism. Our LLM can connect to dynamic or static data by interfacing with its users' private systems. Additionally, it can also generate richer and more customizable business scenarios by interfacing with third parties.
At the same time, Hua Zang can be used to customize content, components, and business scenarios. These scenarios include aspects such as digital marketing, data insight, production, risk control, and management. As a result, we can offer a perfectly tailored product that fits the demands of our clients.
Looking ahead, how do you envision the development of cognitive AI in the coming years? What can users and investors anticipate from Xiao-I and other AI companies in terms of future advancements?
Yuan: The ultimate goal of AI is the singularity that human beings expect. Thus, from my perspective, neither LLMs nor deep learning technologies are the future of artificial intelligence. Because the underlying theories and frameworks of all our current applications have been accumulated over the past 60 years.
So, in the next five to 10 years, one of the biggest challenges for humanity and the development of AI is how we break through this theoretical framework of AI. To reach a future singularity, we need to subvert what was laid out over the past 60 years and create a whole new system.
The first stage of AI is to empower thousands of industries, which we are currently experiencing. We are now in the era of capitalizing on the dividends of artificial intelligence, and the next 10 years will be the golden decade of AI. All walks of life will witness groundbreaking changes and benefit from the huge value brought to us by AI. This is the culmination of 60 years of development and research accumulating, akin to the eruption of an underwater volcano.
At this point, digitization and AI are considered the primary forces driving the transformation of the entire global economy. In the long run, digital transformation will be the core of new infrastructure developments, and AI technology will become an essential part of public infrastructure, like water, electricity, and the Internet. Since Xiao-I believes in the long-term potential of AI, it will continue to focus on developing new technologies, such as advanced AI contact centers, AI smart city concepts, and AI architecture designs.
