Xiang Ligang, founder of Feixiang.com: I believe that Zhou Hongyi's judgment on AGI is wrong

In response to the release of Sora, Internet tycoon Zhou Hongyi released his own article on AGI, judging that "Sora means that the realization of AGI will be shortened from 10 years to 1 year". I carefully read the article and believe that Mr. Zhou's judgment was wrong, and to a certain extent, Mr. Zhou did not fully understand what artificial intelligence is. I want to do a serious analysis, without emotion, just opinions.

In this article, Mr. Zhou expressed two views:

1. Sora is really not far from AGI, it is not a matter of 10 years or 20 years, it may be achieved soon in a year or two.

2. The AI gap between China and the U.S. may be widening.

In response to these two views, I will make an analysis:

1. Will humans achieve AGI in a year or two?

When it comes to AGI, many people may not understand what it is, AGI is general artificial intelligence, which is pre-trained by general large models to achieve some of the capabilities of artificial intelligence.

Another way for general artificial intelligence to correspond is dedicated artificial intelligence, that is, to realize that artificial intelligence is not pre-trained with a general large model, but with a specialized model for pre-training.

To make a more popular understanding, the general model is a god man, he knows everything, can ask him anything, and he may help you solve everything. A dedicated large model (SAI) is a person who has been specially trained to be trained in a certain area and has a certain ability to solve a specific problem.

In layman's terms, this is the difference between a generalist and an expert.

Is there a chance for artificial general intelligence (AGI) or a chance for specialized artificial intelligence (SAI)? These are two technical routes, and it is not clear who is stronger and who has more opportunities. Of course, artificial intelligence in the world first started with dedicated artificial intelligence, for example, the photos of our mobile phones today are not only taken, but calculated, the photo function is becoming more and more powerful, more and more beautiful, many beauties do not take pictures without beauty, this is how to do it, is the ability of dedicated artificial intelligence.

We gave the Internet company customer service after-sales service today, and at the beginning they were no longer people, they were all machines, and they would give you some questions, or according to your problems, they would provide basic answers step by step. What is this capability, the ability of dedicated artificial intelligence.

Today, our smart cars, walking on the road, use a smart chassis, which can adjust the chassis status according to your speed and the road conditions of your car, reaching 24 times per second. How is this done, or dedicated artificial intelligence. Not to mention the car's intelligent driving capabilities.

These special artificial intelligence is through different engines, special training on special data, such as our customer service, about the after-sales of a certain product, it has to process only a few thousand pieces of data, which can be trained better with more general hardware to solve all the problems of this service. It can solve problems simply and efficiently. Different product directions, models are also different, and the ability to process and generate data is also different.

The general large model is different, it is hoped that a model can be used to solve all aspects of the problems faced by human beings, which requires a powerful hardware platform, and the engine of the model will be more complex, and at the same time, it needs a huge amount of data for pre-training.

The company represented by OpenAI will start to release large models in 2022, from text to pictures to videos, on the one hand, there is rapid progress, but on the other hand, there are two extremely difficult problems for such models: one is a large amount of data training, and a lot of data is messy, useless, and uncleaned data, which greatly affects the efficiency of training. In fact, when we use ChatGPT to communicate today, it's not that the more you use it, the better the effect you can use, because for the same keyword, there will be a large number of people to enter data and conduct a training process, and the more you use it, the more nonsense it will be, and it is difficult to have a mechanism and method to ensure that the data fed to it is accurate.

When using ChatGPT to carry out work, we have no way to pursue more accuracy, so we can only reluctantly accept it, or get basic things and then use human intervention to modify them.

Sora has not been officially released to external users,You can see from the video that flows,Or splice some related videos,Of course, the splicing is quite smooth,This is on a large number of real content video platforms,See these wonderful splicing,It's also a special effect,It's just that there is more content like this,Everyone must be tired of it。 The most important thing is that if it is not human-computer communication, making dot maps and videos, can such an ability participate in social management, production practice and social operation, and can it become a valuable ability?

There is also a basic unsolvable problem of general artificial intelligence, that is, its security, because training with it is a process of data collection, and there have been Samsung technicians who want to write code with chatGPT before, which requires him to upload his own code before, revealing the company's major trade secrets.

As long as it is social management, social operation and manufacturing, there must be major security problems, these data must be kept confidential, and the data must be transmitted to the platform of a commercial company, I don't know how the company stores the data, who will analyze the data, and who will use the data. Which country would dare to open up like this?

My opinion is that human beings have not solved the high cost of general artificial intelligence, a large number of messy data interference, and have not solved the problem of data security, general artificial intelligence is just a toy, it is difficult to form a commercial application, and it is even more difficult to affect social life. ChatGPT has been released for more than a year, and there are many large models released in China, are people using it to improve the quality of life and help solve work problems? In the past, a company in the industry said that we don't want PR, and we used large models to write promotional drafts, and then I communicated with the leaders of this company in an event, and found that they had developed a large model and wanted to sell it to everyone, so they didn't need people to write propaganda drafts, in fact, they held meetings and gave speeches, and the drafts were still written by people.

Therefore, I think Mr. Zhou's judgment that AGI will appear in one or two years, is very unrealistic and too optimistic, I think this technology and industry are far from this extent, it is very likely that AGI will be cold in a year or two, and it may be forgotten by everyone in a few years.

2. Is the gap between China and the United States widening?

As for whether the gap between China and the United States is widening, I think Mr. Zhou has obviously confused AI with large models. This is something that I have always strongly opposed, and this is why I wanted to write this article, because Zhou is always an influential person, it is everyone, and such a message is to convey the wrong message to society.

What is artificial intelligence? This is when machines provide services to humans by imitating human intelligence and learning human behavior.

Why do human beings have wisdom? Human wisdom is that we first have a sensing system to understand the trust of the outside world, and this system is composed of eyes, ears, nose, tongue, mouth, skin, and nervous system, and receives information from the outside world. After receiving information from the outside world, we store this information, and some of us have a strong memory and can be admitted to a good university, and some people have a poor memory and cannot study. This is an important support for human wisdom. With information, we also need to transmit the information to our brain, which is processed by the brain, which makes some people react quickly and some people react slowly, which is the information transmission system. Humans receive information, process it, process it, and control our body organs to make movements, such as asking our hands to do something, letting our legs do something, and even making our necks turn and make our stomachs bulge.

The formation of human wisdom is a system of information sensing, information storage, information transmission, information processing, and terminal work.

Artificial intelligence is that if the machine wants to learn human wisdom and form intelligent capabilities, it must also form a human intelligence system, and information processing includes the ability of computing power, algorithms, and large models, and large models are not artificial intelligence, large models are only a part of artificial intelligence, and large models do not represent the level of artificial intelligence in the system of intelligent sensing, information storage, information transmission, information processing (computing power, algorithms, large models), and intelligent terminals of artificial intelligence.

In order to evaluate the level of artificial intelligence, it is necessary to integrate the five parts of the artificial intelligence system.

Today, the best representation of the level of artificial intelligence is of course not SORA, how valuable is it to generate a video with a text? The first field that represents the level of artificial intelligence is smart cars.

First of all, smart cars are the largest consumer goods and the largest industry of human mass consumer goods. In addition to real estate, one of the products that we spend the most money on is automobiles, and the world's economic power must be a big car, which used to be the United States, Japan, and Germany. Of course, today China is joining the ranks.

Secondly, smart cars show the most comprehensive capabilities of intelligent sensing, information storage, information transmission, information processing (computing power, algorithms, large models), and intelligent terminals. Today's powerful smart car must have these capabilities, our car is into dozens of sensors, from lidar, millimeter wave radar, ultrasonic radar to a variety of cameras, a large number of temperature, pressure sensors, this is the intelligent sensing ability, these capabilities help our car safer, more comfortable, more energy-saving, more convenient.

With sensors, this information will be stored in large quantities, and transmitted to the cloud at high speed, which is information intelligent storage and intelligent transmission, the stronger the storage and transmission capacity, the stronger the management ability, the higher the safety level, and the future intelligent driving must also need high-speed information transmission to assist.

For a large amount of information to be processed through computing power, algorithms, and large models, this does not need to be expanded, and everyone can understand it.

Finally, the information processed by calculation is converted into commands and transmitted to all parts of the car, intelligent chassis, intelligent lights, intelligent cockpit, intelligent motor, intelligent power management system, through these systems to achieve safety and comfort, driving intelligent cars will be more stable, safer and more comfortable, where do these capabilities come from, this is from artificial intelligence.

If you compare the artificial intelligence power of smart cars in China and the United States, there is only one Tesla in the United States, and there are others, but the impact is small, the plate is not big, the level of Tesla everyone knows, China's relatively powerful Huawei, Weilai, Xiaopeng, ideal, of course, there are also BYD, Geely, GAC, BAIC, SAIC, and Xiaomi, which is about to come in, in the field of smart cars, China's wolves can tear up any automobile company in the world.

In this regard, artificial intelligence technology and accumulation, my observation is not expanding, but gradually leading. If we want to catch up in this field, we still have to catch up with automotive chips faster, and there are no major problems in other aspects.

Mr. Zhou judged the ability to generate videos with text, saying that there is a gap between China and the United States, and I will not debate whether this gap in ability will affect the overall pattern, no. I'm not really concerned. But to say that the gap between the United States and China is widening, I would like to say a few words, from the perspective of the most representative smart cars, China is not worse than the United States, and even China is stronger.

3. Who is more judgmental about the industry, Zhou Hongyi or Xiang Ligang?

This question is very boring, the reason why there is this problem is that I know that this article is out, there are many cheap people who have heard that China is not bad, and they feel uncomfortable, I must say, Lao Zhou is not stronger than you Xiang Ligang? He is not more judgmental than you? So I will talk about this problem first, and block the mouths of the sluts.

Although Lao Zhou is not particularly familiar, he is still a person in the industry who has communicated, when our company was established, Lao Zhou also looked for me, and once planned to invest some money in me. My career has been much more successful than I was last week, I admit it.

But we must understand that when a person seizes the opportunity in his life and succeeds in his career, will he always see the general direction clearly? Actually, no, I implore Lao Zhou not to get angry after reading the following paragraph.

Lao Zhou saw an opportunity in the field of Internet security, seized the opportunity, and achieved a major leap in life. But Lao Zhou still has a problem, the concentration is not enough, not clearly see the Internet, mobile Internet, intelligent Internet development opportunities and trends, these old Zhou used the way is what is hot, what to chase, from browsers, search, smart homes, smart phones, cars to today's large models, and even recently I see that Lao Zhou still wants to be an Internet celebrity with goods, there are too many ideas, and there is no clear logic and determination to look at the industry.

They are all people in the industry, I shouldn't have said this, but this situation has determined that Lao Zhou also wants to seize the opportunity of the big model, whether it is the stock price or the degree of attention, he wants to seize some opportunities. Therefore, some views on the industry are based on the interests of their own enterprises.

For industry observation, I am confident, about artificial intelligence, I am rarely affected by commercial interests, spent decades tracking, observing, systematic research and methodology, at the same time I also make products, and industry-related enterprises have a lot of exchanges and communication, at present, there are not many people in China's state, in recent years, my judgment on many issues in the industry is more correct, 3G, 4G, 5G, the node of the chip industry, the development of mobile Internet, my monograph on the industry "5G Era" It is the world's No. 1 seller of 5G books, and has published a traditional Chinese version in Hong Kong, and an English version has been published by Macmillan, a world-renowned academic publishing house. In addition to being published in China, "5G Opportunities" is also published in Arabic by Compass Publishing in the United Arab Emirates. Last year, I also published "ChatGPT: Understanding the Technology and Business Logic Behind the AI Explosion", which is one of the earliest monographs on ChatGPT.

This year, I will publish "6G Changes", in addition to 6G, artificial intelligence and intelligent civilization will be an important part of it.

In my career, Lao Zhou is more successful than me, I have to learn from him and study the industry, I don't think I am worse than Lao Zhou.

Finally, to summarize: about artificial intelligence, I believe that for a long time, or dedicated artificial intelligence is dominant, the so-called general artificial intelligence (AGI) is very marginal, because of the cost, information interference, security, means that it cannot become the leading force of artificial intelligence. Not to mention one or two years, I don't think we can see large-scale commercial use in three years. It will not have an impact on the society of products like TikTok.

The most representative field of artificial intelligence should be smart cars, there will be no field, the level of artificial intelligence use can exceed smart cars, the impact on human society and economy is greater, I believe that China will not have a gap with the United States in the field of smart cars. Therefore, China's AI level will not be at odds with that of the United States.

We can also set a three-year period, the Spring Festival in 2027, and we will review it again.

To explain again, this article is not written to refute Mr. Zhou, but to convey the correct information to the society about artificial intelligence, which is our responsibility.