A quarrel? Professor Wang Yu, director of the Department of Electronics at Tsing

A quarrel? Professor Wang Yu, director of the Department of Electronics at Tsing

Many people are watching to see who will come out on top in the battle of AI.

Remember when Chat GPT first emerged, some speculated that it might be just another "flash in the pan" for AI.

However, those who understand the cycles of technological explosions will know that AI will definitely not stop here. In the early morning of the day before yesterday, Open AI released its latest masterpiece - GPT-4o. With its groundbreaking intelligent interaction capabilities, it has completely overturned our perception of AI voice assistants. This is not only a technological leap but also a significant step in the history of human-computer interaction.

At last night's just-concluded Google I/O conference, Google, with its new Gemini AI large model and other products, attempted to regain the initiative in the AI race, "roaring" out AI 121 times in two hours.

Now, AI has become the center of technological discussions.

Many people are watching to see who will come out on top in the battle of AI. What should AI entrepreneurs do? What dangers and opportunities lie beneath the technological explosion?

Today, at the "AI Creation Era - 2024 Jiazi Gravity X New Directions in the Technology Industry" conference, Zhu Xiaohu, Managing Partner of Gobi Partners, Fu Sheng, Chairman and CEO of Cheetah Mobile and Chairman of Orion Star, Li Zhifei, Founder and CEO of Mobvoi, Professor and Department Head of the Department of Electronic Engineering at Tsinghua University, National Natural Science Foundation Outstanding Young Scientist, IEEE Fellow, and Initiator of Wuwen Xinqiong, Wang Yu, and Zhang Jianzhong, Founder and CEO of Moore Threads. Five guests with distinctive characteristics from investors, industry, and experts gathered together to discuss the current status of artificial intelligence and core competitiveness, and to explore the future direction of China's AI.What are the overall trends in AI for the year 2024?

Advertisement

"More serene" is the keyword given by Li Zhifei. Throughout 2023, new concepts, new terminology, and new knowledge emerged continuously. For instance, in March, Li Zhifei, an industry insider, had not yet heard of the term AI Agent, but by April and May, everyone was discussing Agents. At the same time, the industry kept introducing new large models, and the discussions around new knowledge were very lively. Looking at this year, two exciting developments are Sora and humanoid robots. The industry feels more serene.

"Intense competition" is the keyword that Fu Sheng believes in. In 2023, AI training models appeared time and again, with each iteration boasting higher and higher metrics. Not long ago, the release of GPT-4o was considered very impressive by many. However, instead of releasing GPT-5 or GPT 4.5 to make a significant leap in large model performance, Open AI began to focus on application, engineering, and cost. Fu Sheng believes that the lack of continuous improvement in large model performance is due to a bottleneck in algorithm updates.

"Infinite possibilities" is the keyword written by Wang Yu. He stated that an increasing number of young people in universities are beginning to experiment with AI. In April, Tsinghua University established a School of Artificial Intelligence. Around AI, there are two aspects: the AI core, which includes the evolution of algorithms with many companies upgrading; and AI plus, where various industries are starting to move, seeking to find the leverage points for AI within their sectors.

"Imagination" is the keyword for Zhang Jianzhong. He said, "Poverty limits imagination." Most people are poor, and for startups, raising hundreds of millions or even billions may seem like a lot, but it is still not enough to support the construction of a computing power center. Even Open AI, which is very wealthy, also lacks computing resources. Therefore, the key lies in solving the critical issue of companies being unable to experiment and iterate due to a lack of resources caused by poverty.

"Commercialization quality" is the keyword for Zhu Xiaohu, as an investor. He believes that in domestic entrepreneurship, one should not overly pursue technical issues because technology evolves very quickly. The key is whether the company can achieve commercialization and deliver products to customers. Many "disposable" AI products are precisely because they have not met commercialization requirements, and customers will not continue to use them after the first login. Therefore, achieving commercialization requirements is very important in AI entrepreneurship.

Future AI Schools: Technological Faith or Market FaithZhang Jianzhong believes that the development of AI relies on both industry and technology. The industry takes the lead; without an industry, any technological innovation cannot be commercialized. Pioneering industries are likely to be where AI brings about the latest transformations and accelerates innovation. AI has great commercial value in many industries.

He cites the example of early facial recognition technology, which had an accuracy of about 60%, but after deep learning, it could reach an accuracy of 90%. Later, after commercialization, facial recognition became even more accurate than human recognition. Twins that the human eye cannot distinguish, AI can do so.

Currently, it is difficult to conduct research and development in China without considering commercial returns, so the industry must take the lead. However, there are innate conditions for the industry to take the lead, requiring interdisciplinary and cross-professional talents. If AI is to develop in a certain field, it requires talents who understand both professional knowledge and AI knowledge. Behind the industry's lead, it is still talent that drives the development of the industry.

"Long-term technology, short-term market" is the answer given by Fu Sheng. Looking at technology first, technology is changing the productivity, structure, and efficiency of society. The impact of AI on industry models certainly exists, so Fu Sheng is optimistic about technology. However, in the short term, it is important to note that technology is not a linear development process. Technology often brings a wave of market applications after a breakthrough, and technology is phased. When it is truly implemented, do not believe that technology will be the same every year as it was the previous year.

In the 1980s, during the period when the cost of robot technology was higher than the cost of manual labor, turning the automated industry into a flexible production of manual assembly actually greatly improved production efficiency. Therefore, entrepreneurs must closely integrate what they look at in 1-2 years with the market.

In addition, Fu Sheng believes that under the condition of limited resources, it is more possible to innovate according to the needs of the market. The meaning of "resource trap" is that sometimes there is an over-reliance on the explosive power brought by technology itself, and an indiscriminate investment in costs, which can lead to the collapse of industry bubbles. Therefore, AI entrepreneurs should pay more comprehensive attention to applications and how the market can get returns from the market.

Li Zhifei believes that technology and pragmatism can make companies develop better in the trend of AI. For technology entrepreneurs, the belief in technology is an instinct hidden in the genes, but many technology entrepreneurs will have misunderstandings in the early stage: too impractical. They evade real problems under the banner of ideals, such as making many products and always believing that the use of cutting-edge technology, if no one uses it, is the user's problem. In fact, it is more important for entrepreneurs to pay more attention to business, face competition, and understand whether the technology meets user needs.

Wang Yu also gives two paths of technology and business. In universities, the obsession with technology is naturally present. What he shares more is about business. The industry is always waiting for the breakthrough of technology, looking forward to a batch of commercial progress brought about. Therefore, "the transformation of scientific and technological achievements in colleges and universities" is much concerned in the industry, but "social information input" is mentioned less. The key lies in whether university professors and researchers can realize what is lacking in the world at an early stage. Research a systematic technical solution, and solve the problem theoretically. Now, from the communication between universities and business, and then back to technology, the closed loop of this route is not fast enough.In enterprises, it is already quite challenging to conduct research for 1-3 years. However, for universities and research institutions, they can undertake research for 5-10 years. Therefore, if some technical challenges are placed within universities to form a more favorable ecosystem, it still requires collective effort within the industry.

What is Jensen Huang most concerned about at this moment?

Wang Yu indicated that for large models, the emergence of an ecosystem has reduced the need from 2000 operators to possibly only 200 operators, thus affecting the demand for GPUs. Originally, GPUs had excellent performance, and everyone was discussing the demand for GPUs. Gradually, the market could move towards not using CUDA and only providing hardware. With the use of plugins, the underlying chip requirements might not be as explicit.

Li Zhifei stated that there are many differences between China and the United States, from entrepreneurship, choices, to paths taken. For NVIDIA, the largest AI revenue, 80% still comes from various giants, so the biggest concern should be that the giants do it themselves. For instance, Meta, Microsoft, Google are all developing their own chips. When the applications and prices of models all converge, large companies will create hardware that performs similarly, or even cheaper and controllable by themselves. Essentially, whether the investment in large models can match user demands is what determines the sustainability of algorithm charging.

Zhang Jianzhong suggested that Jensen Huang should be worried about what the next AI application will be.

A chicken or the egg dilemma: whether it is the acceleration of Transform in GPUs that is better, or whether there are GPUs that enable Transform. The ecosystem is crucial; everything requires an ecosystem. He believes that after GPUs, Transform can develop better; at the same time, after Transform, it also promotes the better improvement of GPU architecture. Now, Transform has undergone various changes, so what's next?

In the process of large model language acceleration, every minor algorithmic change is a new improvement brought about by the technical architecture. Requirements for communication and scaling are the best places to improve GPUs. Therefore, if someone asks whether GPUs can adapt to the next generation of future technologies, the answer is: OF course. In addition, Zhang Jianzhong stated that the iteration speed of GPU products is not slower than that of large models; instead, the iteration of GPU architecture is also accelerating.

Jensen Huang might believe that GPUs will never fail, and will not be replaced by others, but Fu Sheng holds a different view. He stated that the entire AI industry is moving towards the direction of pruning parameters. Google released the first large model equipped with a mobile phone locally, demonstrating Google's on-the-spot recognition of scam calls. That is, by installing an anti-fraud app on the phone, it does not consume any server-side GPU. If this path is successful by 10%, then 10% of the computing power demand will decrease. Currently, both Google and Apple are working on small parameter models.

On the other hand, Open AI is also streamlining the architecture of Chat GPT 4, optimizing inference, and saving more GPUs. Therefore, as the arms race progresses to the present, a significant trend is to use less computing power, lower costs, and provide better services.Please provide the text you would like translated into English.

Comment