A large model without application is worthless? Here are the best quotes from the AI ​​tycoons at the “China Science and Technology Spring Festival Gala” | WAIC 2024

When the summer breeze blew by the Huangpu River, the annual World Artificial Intelligence Conference (WAIC) kicked off in Shanghai.

Even though the conference entered its second day, the excitement at the scene remained undiminished. The enthusiasm of the participants seemed to be ignited by the summer sunshine. There was a buzz of people inside and outside the venue, and their faces were filled with curiosity and desire for AI technology.

Here, there are not only the most cutting-edge AI products, but also sparkling ideas bursting out in the collision.

For every attendee present, whether they are technology enthusiasts who stopped and stared in front of the booth, or leaders who grasp the pulse of the industry, they all participated in this debate about the future of AI in their own way.

A seemingly ordinary point of view may be pregnant with the next seed that may change the world.

From the rapid development of AI technology, to its implementation in all walks of life, to the insights surrounding AI security, the path forward for the AI ​​industry has gradually become clear in the debate, and it has also shed light on issues such as how humans view AI and how to coexist with AI. Clear your mind.

Some golden sentences are as follows:

  • Wang Jian, founder of Alibaba Cloud: The potential of GPT has not been fully explored today.
  • Wall-facing Intelligence CEO Li Dahai: The client-side model has a unique ecological niche. It is closer to users and can better meet the data privacy needs of individuals and corporate units.
  • Zhang Peng, CEO of Zhipu AI: Large models bring a brand new ability, which we call human-like cognitive ability.
  • Baidu CEO Robin Li: Without applications, just a basic model, whether open source or closed source, is worthless.
  • Zhou Bowen, Director of Shanghai AI Laboratory: Now, only 1% (of resources) in the world is invested in alignment or security considerations.
  • Schwarzman, Chairman of Blackstone Group: We cannot be like an ostrich, burying our heads in the sand and not paying attention to the risks of external technological development.

The potential of GPT has not yet been fully explored

Wang Jian, academician of the Chinese Academy of Engineering and founder of Alibaba Cloud:
I sometimes say that artificial intelligence has a very long past but a very short history.

How long will it last? So long that you can actually go back over a hundred years ago. But when will it be short? It's as short as GPT.

The potential of GPT has actually not been fully explored today.

MiniMax founder and CEO Yan Junjie I think the core issue is how to reduce the error rate of large models from 30% or 40% to 3% or 4%. Even reducing it by an order of magnitude, this can turn AI from a tool that assists humans into an individual that can complete work independently.

When the error rate of the model can become single digits, that is, when it can approach the best human level in human-defined tests, it becomes more natural for it to produce greater benefits in the real economy.

In addition, this also means that large models can complete tasks independently.

Li Dahai, co-founder and CEO of Face Wall Intelligence:
Creating the world's leading lightweight, high-performance large models and compressing "knowledge density" has become a new important direction for future competition in large models.

The client-side model has a unique ecological niche. It is closer to users and can better meet the data privacy needs of individuals and corporate units. It is the key for large models to enter thousands of households and industries.

Zhang Pingan, Huawei Managing Director and Huawei Cloud CEO:
China's AI development is inseparable from innovation in computing infrastructure. This AI innovation path includes releasing the demand for AI computing power from end-side hardware to the cloud.

Just like the camera function of Huawei mobile phones can already present pictures realistically and with rich details, but if the picture processing is placed in cloud computing, the pictures will be more three-dimensional and clear, like a SLR, (like the picture shown) Ant fuzz is clearly visible.

In the cloud, a sustainable AI computing power foundation will be built through collaborative innovation on the cloud-network end-core architecture, including three aspects: core-end computing power migration to the cloud, AI-oriented network architecture upgrades, and cloud infrastructure system architecture innovation.

Zhang Peng, CEO of Zhipu AI:
AI doesn’t just happen today. It has happened for many years, and many problems have been solved in the previous generation.

You can see why some AI methods in the past, such as face recognition, can already surpass human levels at the index level. Why do you think this is not the ultimate answer to our AI?

The large model brings a brand new ability, which we call human-like cognitive ability, which is our vision of Wisdom AI. Let the machine think like a human, instead of making the machine become a machine. , a tool, but allows machines to think like humans.

We think it is more important to think about this ability and the improvement in efficiency it brings.

The killer application in the AI ​​era will be at least three years away

Baidu Chairman and CEO Robin Li:
We must avoid falling into the "super app trap" and believe that an APP with a DAU of 1 billion must appear to be successful. This is the thinking logic of the mobile era.

In fact, not necessarily. In the AI ​​​​era, "super capable" applications are probably more important than "super applications" that only look at DAU. As long as they can bring great gains to the industry and application scenarios, the overall value will be much greater than that of the mobile Internet. .

Many people use open source models to modify them, thinking that this will better serve personalized applications. However, the unique model created in this way cannot benefit from the basic model, and there is no way to share computing power with others.

When you are in a fiercely competitive market environment, you need to make your business more efficient and cost-effective than your peers. At this time, the commercial closed-source model is the most effective.

Without applications, just basic models, whether open source or closed source, are worthless.

So I have been saying since the second half of last year that everyone should stop focusing on models and focus on applications. But I see that our media still focuses mainly on the basic model, running scores and rankings all day long. So and so has surpassed GPT-4 again, OpenAI has released Sora, GPT-4o has come out again, and so on.

Today’s stunning release, tomorrow’s epic update, but I have to ask: Where is the application? Who benefits from this? In fact, applications are not far away from us. Applications based on basic models have begun to gradually penetrate into all walks of life and various fields.

MiniMax founder and CEO Yan Junjie:
As the largest application in the AI ​​era, I think it still has a lot of opportunities, but (currently) this APP is definitely not that big yet.

I think it may take three years (maybe) before there will be something popular, but it doesn’t matter. Once you can do it for the first time, and then your ability becomes stronger, your resources increase, and your technical capabilities become better, it is possible. Do.

At the same time, this thing is coming step by step, and then I think (Killer App) will be at least three years away.

Wang Jian, academician of the Chinese Academy of Engineering and founder of Alibaba Cloud:
First, as long as it is a new technology, new big companies will definitely appear. If a new technology comes out and no new big companies emerge, then there is a question mark as to whether it is a disruptive technology.

But I believe that in this era there will be new big companies coming out. Just like GE back then, this is the first logical way to look at this matter.

Second, but at the same time, a very interesting thing is that there must be big companies that are reborn from the fire. Just like what everyone saw about Apple a few days ago. I think Apple’s announcement is very interesting. I personally don’t think it’s using AI to serve C-side customers.

In my opinion, artificial intelligence technology has actually restructured Apple. So this reconstruction of Apple starts from reconstructing its operating system.

When we say that AI is a revolutionary force, you will find that AI will have an impact on every department. It is difficult for everyone in all departments to embrace AI, which is difficult in many large companies.

I think the difference between small businesses and large companies is fundamentally. Large companies will definitely feel that AI is a revolution in tools. Small businesses will definitely find this a revolutionary tool.

I think if large companies realize that AI is a revolutionary tool, then this change will come.

Jing Xiandong, Chairman and CEO of Ant Group:
The industry generally believes that when general large models are implemented in rigorous industries, they face three "capability shortcomings": relative lack of domain knowledge, difficulty in complex decision-making, and dialogue and interaction does not equal effective collaboration.

In order to solve these problems, professional agents are an effective way for general large models to be implemented in rigorous industries. Through the deep connection of professional agents, Al will bring intergenerational upgrades of services just like the Internet.

The intelligent user experience in the future will not rely solely on one large model, but will require in-depth collaboration across the entire industry, requiring the participation of many professional agents, each performing their own duties.

Yang Jie, Chairman of China Mobile:
First, the new generation of information technology represented by AI has become an important engine for the development of new productivity.
Second, further promoting AI has become an important way to cultivate new productive forces.
Third, using AI+ to promote the development of new productivity has become an important task of the era for the information and communications industry.

Just like our Internet+ era in the past decade or so, computers have not replaced the human brain, but people who are better at using computers will stand out. I think the same is true in the AI+ era.

Xu Li, Chairman and CEO of SenseTime Technology:
My view has always been that although our industry is very hot, including chat applications brought by GPT and video applications brought by Sora, it has not yet reached its "super moment" because it has not really entered the market. Vertical applications in an industry have caused widespread changes.

But in the past two days, I suddenly felt a change of mind. Because a retired teacher from my middle school kept asking me in the group how to use artificial intelligence to write copy, generate blessing pictures, and send them to his retirement group, etc.

I suddenly thought that super moments and applications actually achieve each other. Only the cognitive changes brought about by super moments can finally promote application. Looking back, if we can use it as a support, then our current moment is a "super moment." Therefore, application is the key to "super moments".

Xu Bin, general manager of the National and Local Humanoid Robot Innovation Center:
AI will have great empowerment and changes in our technology industry, cars, mobile phones, PC terminals, etc.

But this process will definitely produce many new business formats. In my opinion, the most core, most critical, or most typical business format will be humanoid robots.

I feel that especially after ChatGPT-4o comes out, the related technology will far exceed our judgment.

Our preliminary judgment now is that in 3 to 5 years, it can be applied in some factory smart manufacturing sites, and in 5 to 10 years, it may be implemented in some complex open environments such as home services.

Ke Ruiwen, Chairman of China Telecom:
First, artificial intelligence requires new digital information infrastructure to provide strong support.
Second, the development of artificial intelligence requires not only the technological research and development of large models, but also the application of models.
Third, the development of artificial intelligence requires joint efforts to build an open ecosystem and jointly promote governance.

Xie Ling, founder and CEO of Yufeng Future, the protagonist of our low-altitude economy is new aircraft that are intelligent, electric and unmanned.

Our topic today, artificial intelligence, mainly replaces the human brain. We also need machines to replace human eyes, noses, ears, hands and feet, and we need a machine to acquire and execute information.

Intelligent aircraft are the same as robots. It is a terminal. The development of artificial intelligence will inevitably bring more development to our low-altitude economy.

Zhang Peng, CEO of Zhipu AI:
We now have a very important point to break through the multi-modality of large models. Why multimodality? This is because when a real person solves a problem in the real world, the information he needs and inputs is itself multi-modal.

In addition to natural language, there are also vision, hearing, touch, and common sense. All of these need to be integrated to solve many common problems in the real world, and they are not even complex problems.

For example, we want it to help us sweep the floor, cook, and wash clothes. In fact, don’t underestimate these tasks. The information they need to input is very multi-modal, so breakthroughs in these capabilities will bring AI to all, and AI will have greater possibilities.

Just turn the original pyramid structure, which means you have a big base, a lot of investment, but a very small return, into an inverted pyramid structure, so that you can truly amplify its value.

AI security and performance development are out of balance, explore the AI ​​45° balance law

Turing Award winner Yao Qizhi
AI risks come from three aspects: First, the extension and expansion of network risks. Now, we feel that managing data security is already very difficult, and it will be 100 times more difficult with the emergence of AI.

The second is social risks that are not realized. For example, AI is very powerful and can be used in many ways, so the possibility of subverting the current social structure in a short period of time exists. For example, someone mentioned that AI may cause massive unemployment in the future.

The third is the most interesting level, survival or existential risks. We have faced this before. When trains or steam engines were invented, some people had such worries.

As a computer scientist, I have seen the most profound problem. On the one hand, we have to control the AI ​​well. After all, this is what we designed. On the other hand, we do not want it to be destroyed by us. This trade-off is very difficult. As Turing said, this is unpredictable, and it is impossible to predict what the machine will do after it has enough computing power.

Wang Jian, academician of the Chinese Academy of Engineering and founder of Alibaba Cloud:
I am a hopeless technological optimist.

Therefore, I believe that humans will definitely solve any problems that arise during the development of technology.

Zhou Bowen, director of Shanghai Artificial Intelligence Laboratory:
At present, from the perspective of algorithm research, talent intensity, business driving force, and even investment in computing power, our investment in AI security lags far behind our investment in AI performance. Right now, only 1% of the world's (resources) are devoted to alignment or security considerations.

The fundamental reason for concerns about these risks is that our current AI development is unbalanced.

To develop trustworthy AGI (artificial general intelligence), we must take into account both safety and performance. Therefore, we need to find a technical system that prioritizes AI safety but can ensure the long-term development of AI capabilities. We call such a technical ideological system the AI-45° Balance Law , in the long term, development should generally follow the 45-degree balance between safety and performance.

The so-called balance means that it can fluctuate in the short term, but it cannot be lower than 45° for a long time, as we are now; nor can it be higher than 45° for a long time, which will hinder the development of technology and industrial applications.

Schwarzman, Chairman and CEO of Blackstone Group:
We cannot be like an ostrich, burying our heads in the sand and not paying attention to the risks of technological development outside.

There is a saying that there are no brave old men in the investment world, which means that you should always be alert to risks. When AI truly develops, some investments may become in vain.

Energy issues should be viewed dynamically in the time and space dimensions

Wang Jian, academician of the Chinese Academy of Engineering and founder of Alibaba Cloud:
New energy will come out, new forms of computing power will come out, new oil will be discovered, and changes will occur.

So I think all problems must be solved in the dynamic process. You would never think that the computing power mentioned in ten years’ time may not be the same computing power as the computing power mentioned today. The electricity we talk about ten years from now may not be the same concept as the electricity we talk about today.

So I think this still needs to be solved in a dynamic process, and we must not solve the problem ten years from now in the current situation. We need to use the situation ten years from now to solve the problems ten years from now.

In addition to the time scale I just talked about, I still want to talk about space being also important, right?

Even space is important. So I want to give you a basic figure, which is that China's annual power generation is more than the United States, Japan, and Russia combined. So I say China is the least worried about not having electricity. We have other places that can do just fine. At least we use a lot more electricity in other places.

So it must be related to space and time. Whether you can talk about it in the space you left is a question.

Let me emphasize again that China's annual power generation exceeds the combined power generation of the United States, Japan, and Russia. So I think we're probably the region that least needs to worry about this at this stage.

The most important thing in the short term is actually the advancement of AI technology

Zhang Peng, CEO of Zhipu AI:
The decline in the price of large models is driven by technology. Because the technology itself is getting better and better, the cost is getting lower and lower, and the price continues to fall. But it’s not good to go too far. It’s the value transfer I just talked about.

The real value should be that level by level, everyone continuously adds their own value to the same thing, and then amplifies and expands this value. We provide better and high-quality services to everyone, and hope that everyone can use this service to create greater value. Accordingly, this part of the value we create should be passed back in reverse.

Just like our BP algorithm (error back propagation), everyone gets their own part of the value. This is a normal and reasonable market value chain.

How do we use the new generation of generative AI technology and large model technology to empower the real economy? We believe that it must solve this problem from this direction, that is, you need to build a more general and basic capability, and use this Use general and basic abilities to solve multiple problems, and then divide the sum of the benefits by your investment cost.

MiniMax founder and CEO Yan Junjie:
The value of AI must not lie in selling this technology, but in how much value technology changes bring to users. I think that whether it is technology itself, products or business, it should develop in the direction of value.

Xie Ling, founder and CEO of Yufeng Future, we hope that the large AI model will have a ubiquitous and universal infrastructure similar to the satellite positioning or cellular communications we see today.

MiniMax founder and CEO Yan Junjie:
Frankly speaking, I think most (domestic) companies have not differentiated themselves yet. They are all about the same, perhaps with similar model levels and similar products, and then they will start "price competition."

I think this thing is not a bad thing. In fact, it forces everyone to do better technological innovation.

I think whether technology is online or not depends on whether you are a qualified company. If the technology is not good enough, maybe he should be eliminated. This is the first one; but good technology is useless, because now there are AI filters. For example, if you look a year from now, it must be Maybe not exactly on a technical level, maybe looking at the company from a commercialization perspective.

Therefore, I think the most important thing in the short term is actually the advancement of AI technology.

# Welcome to follow the official WeChat public account of aifaner: aifaner (WeChat ID: ifanr). More exciting content will be provided to you as soon as possible.

Ai Faner | Original link · View comments · Sina Weibo