Shulou(Shulou.com)11/24 Report--

The upsurge aroused by ChatGPT is spreading rapidly among Chinese technology companies. Everyone is looking forward to this: who will be the first ChatGPT in China?

But in Xiao Bing CEO Li Di's view, compared with the rush to do the Chinese version of ChatGPT and show off the "muscle" in a panic, what is really worth thinking about is what changes the rise of the new paradigm of large model technology, represented by ChatGPT, will bring. And what is AIGC's real business model?

Xiao Bing is a rare AIGC all-track player in China. In the vertical technology scene of AIGC, Xiao Bing team has explored the subdivided fields such as audio generation, text generation, image generation, video generation and digital human, and the commercial achievements have been widely used in finance, culture, textile, tourism and other fields.

Li Di himself is not only a practitioner of AIGC track, but also used to be the management of Bing. From 2013 to 2020, he served as the general manager of Microsoft's Bing search engine Asia.

He believes that in the field of artificial intelligence, China does not have much disadvantage in terms of talent, algorithm, computing power, or in terms of market and data, and the leading countries in the world. The difference is that Chinese science and technology companies are slightly eager for quick success and quick profit in the polishing of technological innovation. "OpenAI made ChatGPT based on eight years of accumulation, but domestic companies only want to rely on a few months of sprint."

Before ChatGPT came out of the circle, Li Di, who had been benched on the AI road for many years, believed that the decisive factor in whether this could be done was a state of mind that could withstand loneliness. "all innovation is very good at the beginning, but it is difficult to optimize and implement it later. The hardest test lies in concentration and determination, and it lies in whether or not AI First has been achieved."

In his view, the big model itself triggered a new round of innovation, in which the ChatGPT is only the first product to enter the public eye. Li Di said, "this is just the beginning."

First, "to be a Chinese version of ChatGPT is not our pursuit"trend": what kind of change and impact does the popularity of ChatGPT bring to Xiao Bing?

Li Di: every few years, the AI industry will fall into a technical bottleneck. Before the emergence of ChatGPT, everyone in the industry felt more depressed. We kept training for deep learning, but we always felt that the effect had improved slowly.

Suddenly, there is such a company called Open AI, fully committed, believing in the big model of "violence produces miracles", spending a lot of money and a lot of time, when no one knows if it can be done, Open AI spent several years wading through the jungle, and everyone saw that the original road was open, and the technical ceiling of the whole industry was further broken.

It can be predicted that in the next few years, everyone will make all kinds of innovations around this technology. All relevant companies, as long as they have the strength and ability, will make innovative breakthroughs in the way of large models, and we all benefit from the fact that some people take the lead to explore new directions and break through the ceiling of technology. it has brought the vigorous development of the whole industry in the next few years.

"Tide": there are so many AI companies in China, including Xiaobing, why didn't we produce ChatGPT?

Li Di: because the open domain dialogue itself is very risky. In 2016, Microsoft actually built a robot that could modify its responses in real time through user feedback. That year, we posted the result on Twitter, and 24 hours later, it was taught to be a racist, and Microsoft's CEO came out to apologize and then took it off the shelves. Even today, this risk is troublesome for ChatGPT.

Second, when the product is launched, you have to predict in advance where the product may be used. Today, I can say very clearly that not only Google but also Xiao Bing has a similar proposal for ChatGPT, but I dare not release it.

Now ChatGPT is in the United States, and a very large group of users are students. Nine out of 10 American students already know about ChatGPT because it can help students with their homework, which is an important driver of ChatGPT's popularity. This matter involves AI ethics. You know very well that it will be like this, so do you want to launch such an influential to C product? For AI, the quality of the conversation is not a problem, the problem is how to make it last for long, can continue to exist.

"Tide": compared with foreign countries, do we have a big gap in large model technology?

Li Di: there won't be much difference. In the field of artificial intelligence, China does not have much disadvantage with the leading countries in the world in terms of talent, algorithm, computing power, or in terms of market and data.

Although ChatGPT is very large, it is actually a single point of technological breakthrough in the large model, and it has not yet formed a large ecosystem. Of course, Open AI also noticed this, so it began to invest in all kinds of ecology, injecting capital into its upstream and downstream companies, hoping to break through this concept. But this actually reflects a problem, that is, the single point breakthrough itself does not have too strong barriers, there is time to take the lead, but this lead is not impossible to catch up.

Our disadvantage is that the polishing of technological innovation will be a little bit eager for quick success and quick profit, such as artificial intelligence, especially large models, which takes many years to accumulate, constantly polish, and sit on the bench for a long time. OpenAI made ChatGPT, relying on eight years of accumulation, but domestic companies only want to rely on a few months of sprint, can they produce any decent products?

"Tide": have any investors suggested Xiao Bing to do a similar ChatGPT recently? What are your considerations?

Li Di: some investors will say that you should also make this kind of super-large-scale model with super-large parameters. However, I will not launch a thing that allows students to copy homework. No matter how much money it makes, we will not do it. So we hide, not particularly like to steal the limelight.

The focus of our attention has not changed, as it was before and will continue to be. When the meta-universe came out, we also made it clear that we did not understand the meta-universe. At that time, some investors came up to us and said, "even if you are the meta-universe, we do not want to talk about it, because he does not agree with our ideas."

I can launch a similar application in China, and then a group of students use it to copy their homework. What am I looking for? It is not our pursuit. I prefer to use the investor's money in other places, such as thinking about what happens after ChatGPT and how to optimize the product.

Second, the strange circle of AI's business model, ChatGPT alone can not break the "trend": what do you think of the business model change brought about by AIGC?

Li Di: the biggest problem with the AIGC business model has nothing to do with technology. What does it have to do with? It has something to do with what role you, as an artificial intelligence company, play in the whole AIGC ecology.

For example, bilibili, Douyin, and Kuaidi have a lot of film and television commentary videos. The voiceover in this one is actually not human at all. The voice is synthesized by an API interface of Microsoft Speech Cognitive Service. You can get that sound by calling that API interface and typing the text.

An and B, two UP owners, A made a video with this service and made a lot of money from proper operation, while B made no money at all for a variety of reasons, and the two videos have very different value in the content market. But from the point of view of AI, it has not been able to get different benefits from these two very different videos, which can only be charged for three minutes.

At the same time, because he provides API, specific to the sound synthesis API, scattered to this extent, in the future he must have many competitors, profit margins will be further compressed. So, artificial intelligence companies, if they only offer this ability to invoke API, have not actually achieved a real return on value in any of the ecosystems it creates great value.

"trend": is the emergence of ChatGPT possible to break the business model of AI in the past?

Li Di: no, technological innovation and breakthroughs can not solve the problem of business model. A pupil wrote a composition in ChatGPT and handed it in, which was equal to the money earned by ChatGPT in writing an academic paper for a college student, without any difference. This model, in the end, also determines that its income will gradually decline, because the profit margin is obtained from reducing costs, rather than constantly gaining new value-added value.

"Tide": if ChatGPT becomes only a software tool, will there be a very low return on commercial cash?

Li Di: yes. Everyone knows that game manufacturers spend a lot of money on art resources. Some AIGC companies are already customizing models for game manufacturers to help produce a variety of game art resources according to the style of game manufacturers' art resources, which can save game manufacturers millions of dollars, but do you know how much they are willing to pay for this model? Less than 100000 RMB.

So AIGC's business model has to change, and that's what we're doing. Artificial intelligence actually provides intelligence or the value of wisdom, rather than just the value of tools or information retrieval like software. If you still follow the thinking of the software and only provide API calls, you will think that the whole world will transfer my API and naturally make a lot of money. This idea is wrong, and one day we will form a new consensus that we have to go deep into the revenue of the terminal product.

Tide: if you choose to do a vertical scene application, do you still need to develop your own large model?

Li Di: with the iteration of Open AI technology, companies like Jasper, which rely on Open AI technology, are bound to have problems. Just like the AI's own business model dilemma just mentioned, if AI only provides API, you will soon find that you can't make money. I have created a lot of value upstream, but these values have been taken away by downstream companies like Jasper.

So he will definitely end up, there is no other possibility, it is possible to use capital operation to unite with Jasper, but if he uses a newer version of technology to do the same thing as Jasper, it is easy to bring over customer resources, which is almost certain to happen.

Trend: what is your advice for these startups?

Li Di: I advise you not to use anyone on the market who has launched such a thing that everyone can use. If you do it, he will eat it for you. There are no bones left. API invokes the business model that he can't make money, so even for his own development, he will jump in and do the things of other enterprises on his ecological chain, the first to bear the brunt is the people who directly apply his technology to provide services.

You might as well make your own big model, which doesn't have to be as big as Open AI. For a company, it is make sense to make a vertical, large-scale model. You don't have to get hundreds of billions of parameters on such a large scale and burn so much GPU to do it.

Third, "the biggest commercial return on the AIGC track is digital people" and "trend": will Xiao Bing try to launch C-end products in the future?

Li Di: Xiao Bing actually started with to C, but the main problems of to C are safety and regulation, as well as self-regulation. To C (product) we all know that there is a huge gap between what it should be and what it will eventually be accepted. The same is true of ChatGPT. I see that people in charge like Open AI are now jumping out and taking the initiative to say that they want to be regulated.

"Tide": in your opinion, where is the level at which the AIGC vertical scenario can achieve the greatest commercial return in the short term?

Li Di: the biggest commercial return track of AIGC is actually digital people, this is our judgment, because its customer unit price will be obviously higher. At that time, we also went to generate terminal content, which could only be sold according to the dose, resulting in a content market similar to the vegetable market, with a very low unit price. Later, we tied the generating capacity to the digital person, and the average unit price increased rapidly from 200000 to 3 million.

"Tide": how to create a lasting differentiated business? From ordinary to professional, where is the threshold?

Li Di: in China, many start-ups often use a large foreign model, or the technology disclosed in the paper, to quickly appear and form a demo, so it leads to the actual lack of innovation, more often the realization of something in engineering, but there is not much difference between them.

Differentiation itself needs innovation. I've seen a lot of AIGC startups, maybe they were working on meta-universe in the first half of last year, or they were still working on blockchain in the first half of last year, and they can do whatever they want. What do you think he might have differentiated? It's unlikely.

If you want to get ahead on the AIGC track, I think the basic quality you need to have is to focus. You have been doing the same thing for many years. Even if the outside world denies it, if you really believe in this direction, you need to stick to it and take time to precipitate and polish yourself and your products.

"Tide": what is Xiaobing's play and plan in the direction of AIGC in 2023?

Li Di: our investment in AIGC is very clear. We will do a series of ecological environment building in 2023, and the AIGC will be bundled with IP. We think that the added value of the content industry is in the virtual human, and then we will work with the industry to do this kind of ecological strategy integration, in this way to improve or introduce a new AI ecological environment.

Fourth, the giant reshuffle, some entrepreneurs will suddenly rise "tide": Baidu, Ali, Tencent and other big companies have expressed their concern for the AIGC track, in your opinion, will big companies have more innovative advantages than startups?

Li Di: I think big companies may be slower on this, because for startups, the mechanism is more flexible, to put it bluntly, Nothing to lose, there is nothing to lose, the reaction may be faster.

There is a reason why the big factories are slow, because they need to be steady, they are fighting the group army, and the mobilization of the group army requires the organization and coordination of all parties, and it is certainly not as fast as calling out the small friends at the entrance of the village and go out as soon as they say so. But it is also for this reason that only big factories can fight the war of the group army.

Small companies pay more attention to opportunity, because of its bright side, small companies have the opportunity to get the capital to do this, which is very reasonable. However, the burden and burden of big companies will be heavier, and the current technology is really not accurate enough, so it is difficult to say whether they are really Ready.

Trendsetter: products like ChatGPT require a lot of top talent, cost a lot of money, and take time to precipitate. Is this track not very friendly for startups?

Li Di: we can't rule out the fact that technological innovation is always like this. When everyone went to do App, everyone thought that big factories would do it, and there was little chance in this market, but now we will find that some of them will really emerge as a new force, such as Little Red Book as a community. Just put into the artificial intelligence market, this thing will become more difficult, the technical threshold will be much higher, do not rule out this possibility, but this possibility will be greatly reduced.

What does it look like? My analogy may not be appropriate, the era of artificial intelligence is very much like a class-solidified society, it will become more and more difficult to get out of a poor family. Starting from scratch and starting from scratch, it will be harder to build a great company in the field of artificial intelligence than it used to be in the mobile Internet era.

"trend": will ChatGPT replace the traditional search engine market?

Li Di: we have experienced the era of portals in the past, and then we have search, and then to information recommendation. In these different times, we have produced platform-level products. In fact, the result of each round of technological evolution is to nibble away part of the original market. Until today, portals still exist, and they are not an alternative relationship.

As we mentioned two years ago, from portal to search to recommendation to dialogue, this is a technological evolution, so there will be a new platform with dialogue as the basic form of interaction, but it does not replace the previous search. Microsoft's addition of ChatGPT to search is just an addition, with an additional experience. The main reason why it can not replace search is that its accuracy is not enough, even if its accuracy is achieved in the future, it should appear in a new form, and search will still exist.

But it will greatly erode the market share of search, which is certain, but in fact, search has not just been eroded today, since the era of mobile Internet, search has been eroded, because search belongs to the era of PC.

"Tide": will Chat GPT be as fleeting as we have experienced many concept waves before, or will it become a real technological revolution?

Li Di: we all think that the big model corresponding to Chat GPT is a technological revolution in the future, but there is no need for excessive mythology. It is very clear that it opens a ceiling for innovation in the next few years.

But when Chat GPT really hits the ground, it faces the problem of business model, the phenomenon of letting students copy homework, including the safety of its supervision, and so on. These problems that have to be faced can not be solved by the existing technical means. Chat GPT itself, I think this may be the case in the next few months. But the discussion of the big model will not stop for at least a few years.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.