Source: Caixin.
Author: Huang Xinyi

Alibaba Group CEO and Alibaba Cloud Intelligence Group Chairman and CEO Wu Yongming delivered a keynote speech at the 2024 Yunqi Conference. He believes that in the past 22 months, the development of AI has been faster than any other period in history, but we are still in the early stages of AGI transformation. The greatest imagination of generative AI is not to create one or two new super apps on the mobile screen, but to take over the digital world and change the physical world.

Wu Yongming predicts that the next stage of advanced models will require billions or hundreds of billions of dollars in investment. And robotics will be the next industry to undergo a major transformation. In the future, all movable objects will become intelligent robots. Almost all software and hardware in the future will have reasoning capabilities, and their computing cores will adopt a computing model where the GPU AI computing power is the main component and the CPU traditional computing is the auxiliary component.

Wu Yongming revealed that in the emerging market for additional computing power, over 50% of the new demand is driven by AI, and the demand for AI computing power has become mainstream. This trend will continue to expand. In the past year, Alibaba Cloud has invested heavily in new AI computing power, but it is still far from meeting the strong demand from customers.

The following is the main content of the speech:

In the just concluded summer, Alibaba Cloud fully supported the historic breakthrough of the Paris Olympic Games, with cloud computing surpassing satellites for the first time and becoming the main broadcasting method for the Olympics. AI was also widely used for the first time in the Olympics. Today, the focus of the Yunqi Conference is also on AI. I will mainly share three points of content:

First, in the past 22 months, the speed of AI development has exceeded any historical period, but we are still in the early stage of the AGI revolution.

Large-scale model technology has iterated rapidly, and technological availability has greatly improved. Large-scale models now have multimodal capabilities for text, speech, and vision and can begin to execute complex commands. Last year, large models could only help programmers write simple code. Today, they can directly understand requirements and complete complex programming tasks. Last year, the mathematical abilities of large models were only at the level of high school students. Today, they have reached the level of international Olympiad gold medals and are approaching the level of Ph.D. in subjects such as physics, chemistry, and biology.

At the same time, the inference cost has decreased exponentially, far exceeding Moore's Law. Over the past year, the price of calls to the Thousand Q&A API on Alibaba Cloud has decreased by 97% on the BaiLian platform, and the cost of calling a million tokens has dropped to as low as 0.5 RMB. The cost of inference is a key issue for application explosion, and Alibaba Cloud will continue to work on reducing the cost.

The open-source ecosystem is thriving. In June of this year, the Thousand Q&A released Qwen2 as open-source software and quickly rose to the top of Huggingface's global open-source model ranking. On Huggingface, the native and derivative models of Qwen are close to 0.05 million, ranking second globally. The Magic Community of Alibaba Cloud has over 0.01 million models and has served over 6.9 million developers.

All of this is just the beginning. To achieve true AGI, the next generation of models needs to have a larger-scale, more universal, and more generalized knowledge base, while also possessing more complex and hierarchical logic reasoning capabilities. The investment threshold for advanced models worldwide will reach the level of tens of billions or hundreds of billions of US dollars. AI has a clear path to creative ability and helping solve complex problems for humans, and it has also opened up the possibility of extensive applications of AI in various industry scenarios.

Secondly, the greatest imagination of AI is not on the mobile screen, but in taking over the digital world and changing the physical world.

Many industry insiders have been thinking about the biggest application of AI, perhaps always thinking about innovative super apps of the AI era on smartphones. But we believe the greatest imagination of AI is definitely not on the mobile screen. The greatest imagination of AI is in infiltrating the digital world, taking over the digital world, and changing the physical world. This is the greatest imagination of AI.

We should not only look at the future from the perspective of mobile internet. The biggest imagination of generative AI is not to create one or two new super apps on the smartphone screen, but to take over the digital world and change the physical world.

In the past thirty years, the essence of the internet wave has been connectivity. The internet has connected people, information, businesses, and factories, and has improved the world's collaboration efficiency and created enormous value, changing people's lifestyles. But generative AI creates new value through the supply of productivity, thus creating greater intrinsic value for the world, which is an overall improvement in the world's productivity level. This value creation could be ten or even dozens of times higher than the value created by mobile internet connectivity.

We believe that generative AI will gradually infiltrate the digital world and take over most things in the physical world with AI capabilities. This will create a new generation of AI-capable products that are connected to the AI-driven digital world in a synergistic way.

For a long time, the focus of AI has mainly been on simulating human perceptual abilities, such as natural language understanding, speech recognition, and image recognition. However, the rise of generative AI has brought a qualitative leap, as AI is no longer limited to perception but demonstrates the power of thinking, reasoning, and creativity for the first time.

Generative AI has given the world a unified language - Token. It can be any text, code, image, video, sound, or the thoughts of humans throughout history. AI models can understand various aspects of the real world by tokenizing physical world data, such as human walking, running, driving vehicles, using tools, painting, composing, writing, expressing, teaching, and programming skills, or even starting and running a company. After understanding, AI can imitate humans to perform tasks in the physical world, which will bring about a new industrial revolution.

We can see that the automotive industry is undergoing such a transformation. Previous autonomous driving technologies relied on humans to write algorithmic rules, with hundreds of thousands of lines of code, yet they still could not cover all driving scenarios. With the adoption of "end-to-end" large model training, AI models directly learn from massive human driving visual data, enabling cars to have driving capabilities surpassing most drivers.

The robotics industry will be the next one to undergo tremendous changes. In the future, any movable object will become an intelligent robot. It could be a robotic arm in a factory, a crane on a construction site, a warehouse worker, a firefighter at a fire scene, or even domestic companions like a pet dog, nanny, or assistant.

In the future, there will be many robots in factories, producing robots under the command of AI large models. Just as every urban household now has one or two cars, in the future, each household may have two or three robots to help improve efficiency in daily life.

It can be seen that the AI-driven digital world is connecting the physical world with AI capabilities, which will greatly enhance the productivity of the whole world and have a revolutionary impact on the efficiency of the physical world's operations.

Thirdly, AI computing is evolving rapidly and becoming the dominant computing system.

Whether it's edge computing or cloud computing, this is a very obvious trend. The generative AI will bring fundamental changes to computing architecture in the digital and physical world. In the past few decades, the CPU-dominated computing system is accelerating its transition to the GPU-dominated AI computing system. In the future, almost all software and hardware will have inference capabilities, and their computing cores will become a computing mode where GPU AI computing power is the main component and CPU traditional computing is the auxiliary.

We can see that in the market for additional computing power, over 50% of the new demand is generated by AI-driven applications, and AI computing power demand has taken the mainstream position. This trend will continue to expand. In the past year, Alibaba Cloud has invested heavily in building a large amount of AI computing power, but it still cannot meet the strong demand from customers.

Today, all customers, developers, and CTOs we interact with are almost all reconstructing their products with AI. A large amount of new demand is being driven by GPU computing power, and a large number of legacy applications are also being rewritten with GPU. AI computing is accelerating its penetration in industries such as automotive, biomedical, industrial simulation, weather prediction, education, enterprise software, mobile apps, and gaming. In every industry, an invisible new industrial revolution is quietly evolving.

In all industries, there is a need for infrastructure with stronger performance, larger scale, and more adaptability to AI requirements.

Alibaba Cloud is investing in AI technology research and infrastructure construction with unprecedented intensity. Our single network cluster has expanded to the level of hundreds of thousands of GPUs, and we are rebuilding an advanced AI infrastructure for the future in every aspect, including chips, servers, networks, storage, cooling, power supply, and data centers.

Based on historical experience, people often overestimate new technological revolutions in the short term and underestimate them in the long term. This is because in the early stages of new technology applications, the penetration rate is relatively low, and people have no experience of such events. Most people's instincts will generate doubt, which is normal. However, new technological revolutions will grow amidst people's doubts, causing many people to miss out while hesitating.

Standing at the beginning of the AI era wave, I feel extremely excited. Today, we have invited entrepreneurs and scientists from the fields of big models, autonomous driving, and siasun robot&automation. They are running to reshape our world with AI, looking forward to their wonderful sharing.

Editor/Jeffy