Depth丨Light years away from Meituan, but big models are not good brothers everywhere

Source: Beijing Business Daily

Reporter: Yang Yuehan and Wei Wei

On the other side of the seemingly lively wave of artificial intelligence entrepreneurship, is there also realistic pressure in terms of capital, technology, and competition, and these are likely to be the real "light-years away" under the filter of fanaticism.

"I applaud those who have the courage to embark on this road, and those who embark on the road are all warriors."

In April of this year, around the topic of entering the AI field, Wang Huiwen once said such a sentence. At that time, it was only two months since he posted the "AI Hero List" in a high-profile manner, announcing the establishment of light-years away.

This dialogue allowed the outside world to see the enthusiasm of the Internet veteran, and also saw the long-lost burst of epoch-making potential in the technology circle. But what is surprising is that two months after the end of this dialogue, Wang Huiwen fell ill, and Light Years Away also fell into speculation about where to go.

The answer was soon revealed, and Wang Xing, the founder of Meituan, took over the entire team of Good Brothers. On June 29, Meituan announced that it had completed the acquisition of 100% equity of domestic and foreign entities light years away, at a purchase price of approximately 2.065 billion yuan.

Not long ago, around the topic of "should Chinese entrepreneurs compete for large-scale models", the "fierce debate" between Fu Sheng, CEO of Cheetah Mobile, and Zhu Xiaohu, managing director of GSR Ventures, just came to an end.

In the six months since artificial intelligence became popular in the technology circle, these two events seem to be a key turning point, prompting people to think about whether there are also realistic pressures in terms of capital, technology, and competition on the other side of the seemingly lively wave of artificial intelligence entrepreneurship, and These are likely to be the real "light-years away" under the fanatic filter.

Light Years Away is Meituan

Although Wang Xing and Wang Huiwen are well-known brothers and lovers, and Wang Xing is an investor light years away, it is still surprising that Meituan completed the wholly-owned acquisition so quickly. Meituan explained that the acquisition of leading AGI (General Artificial Intelligence) technology and talents through the acquisition has the opportunity to strengthen its competitiveness in the fast-growing artificial intelligence industry.

Light Years Beyond is a leading AGI (General Artificial Intelligence) innovation company in China. It was founded and controlled by Wang Huiwen, co-founder of Meituan, former director of the company and related person, and is committed to promoting the realization of AGI inclusiveness. Net cash totaling about $285 million is currently light-years away. Sources said that Light Years Beyond completed the team building with products and technical talents within two months, attracting many top experts and entrepreneurs in the field of artificial intelligence to join. The current team size is about 70 people.

Compared with other large-scale entrepreneurial teams, Light Years Away has obvious advantages. Wang Huiwen entered the game with capital, and also has entrepreneurial experience in creating companies with a market value of 100 billion yuan. He has outstanding organization, strategy, rhythm control, and product capabilities. In the capital market, light-years away is also a "sweet pastry". In early June, it was reported that Light Years Beyond had completed a new round of financing of US$230 million, with a pre-money valuation of US$1 billion. In this regard, Wang Huiwen once told a reporter from Beijing Business Daily: "The data is not accurate and will be disclosed later." According to the information provided by Zhang Peng, the founder of Geek Park, "The actual amount of financing that has already arrived in the A round is light years away. Higher than the rumored $230 million."

Everything seems to be going smoothly, and the battle of the giant model gods is intensifying. Any problems in management and funding of the entrepreneurial team may affect the order of appearance in the future. Because of this, after Wang Huiwen fell ill, the destination of light-years away became the key to determining whether light-years away could maintain an advantage.

"This transaction was processed very quickly, half favored, half sold." Wang Chao, founder of Wenyuan Think Tank, told a reporter from Beijing Business Daily.

It is rumored that after Wang Huiwen realized that he had a clear health problem and needed to be forced to leave his job, he had already communicated with the team and investors, and contacted Tencent, ByteDance, Meituan, and Kuaishou before the news came out. Explore various potential options.

The above-mentioned big companies have their own origins with Light Years Beyond. Among them, Tencent has made a big bet on Light Years Beyond. Meituan and Wang Xing are very interested in AI, and Wang Xing said in March that he will participate in the A round of Light Years Beyond Invest and act as a director. Su Hua, the founder of Kuaishou, invested nearly 40 million US dollars in the A round of financing.

To say that it is most likely to take over the light years away, Zhang Peng once judged that "Meituan is a more likely choice", and he analyzed this from the perceptual and rational levels.

On the perceptual level, "Wang Xing and Wang Huiwen are classmates and roommates of Tsinghua University. They have fought side by side several times on the road to entrepreneurship. After Wang Huiwen announced his entry into the big model, Wang Xing immediately followed up, spoke out, and paid for Wang Huiwen The entrepreneurial project. The entrepreneurial partnership relationship between them may not be easily understood by the outside world, but it is also the most important element that cannot be ignored in many cases.”

On the rational level, “Since Meituan’s strategic upgrade to ‘retail + technology’, both Meituan and Wang Xing himself have shown interest in exploring cutting-edge technologies such as AI. In Meituan’s earnings conference call last quarter, Wang Xing once said that Meituan has established an internal team to carry out large-scale model and application-level research and development, and is open to external technical cooperation opportunities and investment opportunities.”

In terms of emotion and reason, it is not an accident that Meituan took over light years away. For the future, Meituan's plan is: After the merger is completed, it will support the Lightyear team to continue exploring and researching in the field of large models.

Model layer VS application layer, who is the king

"Half of the start-ups in Silicon Valley started around ChatGPT, our investors can still be so ignorant and fearless." On the evening of June 26, Fu Sheng reposted an article titled "Zhu Xiaohu: ChatGPT is very unfriendly to start-up companies" in Moments, and attached such a comment.

This is the beginning of this debate, and it is also enough to become the beginning of the phased thinking of China's large-scale entrepreneurship. Ma Qianli, the co-founder of Unbounded AI, analyzed to the reporter of Beijing Business Daily that the core of this debate is actually the question of which has more entrepreneurial opportunities, the model layer or the application layer.

Zhu Xiaohu believes that the model layer creates 99% of the value, and the application layer only integrates an "interface" on top of the model layer, which does not create much value, so there are few opportunities for entrepreneurship. Fu Sheng believes that although the underlying model is valuable, problems or needs cannot be solved simply by relying on the underlying model. The application layer is also needed, so there are still opportunities for application layer entrepreneurship.

In the past period of time, the entrepreneurial wave of large models was once dubbed the title of "Hundred Models War". The entrants include Internet veterans such as Wang Huiwen, the co-founder of Meituan, and Wang Xiaochuan, the former founder of Sogou, as well as Baidu, Ali , Tencent, Huawei and other major manufacturers, as well as star enterprises such as SenseTime and Kunlun Wanwei.

The investment requirement of "sky-high price" has indeed raised the threshold for entrepreneurship of the large model itself, and the identity of the entrant speaks for itself. Ma Qianli predicts that the opportunity of the model layer and the application layer will be decided soon.

Judging from the intensity of competition, Ma Qianli believes that the model layer is much more intense, and the model layer may have a "winner takes all" situation in the future. Every field, whether it is a language model or a Vincent graph model, may form The pattern of oligopoly competition. Corresponding to the application layer, there may be a situation of "a hundred flowers blooming", that is, as long as a specific problem in a certain field can be solved, continuous value will be generated, and even the competition of the underlying models will benefit the upper-level applications, because each underlying model is in Do everything possible to build a prosperous application ecosystem, there may be some incentives to stimulate the application layer.

Unbounded AI bet its "treasure" in the field of Vincentian graphs. Ma Qianli said that compared with the big language model, the Vincent graph model has a better open source ecology in the development, which brings many opportunities to the application layer - that is, to use the open source model for secondary development, and to achieve from 0 in a short time To 1, the upper layer uses small steps to verify business ideas.

According to reports, the Vincent graph model layer can combine market demand to train various models in various subdivided fields, from 1 to 100. From the experience of unbounded AI, it is important for entrepreneurs to embrace open source and strengthen self-research It is a good choice of direction, and it is also microscopic enough.

Lin Laini, the vice president of the commercialization of Huayuan Digital Homo sapiens, also analyzed to the reporter of Beijing Business Daily that at present, entrepreneurship in the field of AI is more concentrated on the development of the application layer based on the large model, because the training of the underlying large model requires a large amount of data. Researchers first need to solve the problem of where the data comes from and whether it is compliant. At the same time, the training cost and time cost of large models are quite high, so Maas, or Model As Service, will become a common phenomenon.

"What's interesting is that at the end of the debate, the two seem to have reached a consensus that entrepreneurial opportunities at the model layer are 'BAT'-level opportunities, while entrepreneurial opportunities at the application layer are scattered and smaller opportunities." Ma Qianli concluded. Just as Zhu Xiaohu added an explanation in his circle of friends: Don’t be superstitious about the general model. For most entrepreneurs, the scene is the priority and the data is the king.

Large-scale model entrepreneurship is easier said than done

Kai-fu Lee, chairman and CEO of Innovation Works, once described the AI model as "a historical opportunity that China cannot miss." In his view, the popularity of generative AI represented by large models such as GPT-4 is spreading globally, which means that the era of AI 2.0 has arrived, and it will bring opportunities ten times larger than that of the mobile Internet era, penetrating all walks of life Greatly boost productivity.

The clash of views between Fu Sheng and Zhu Xiaohu reflects to a certain extent the state and reaction of the two different roles of entrepreneurs and investors under this unprecedented historical opportunity.

If you want to choose an iconic figure among entrepreneurs, it may be Wang Huiwen. Four months ago, Wang Huiwen's high-profile posting of "AI Hero Posts" is still vivid, but all this was disrupted by a message four months later.

On the evening of June 25, Meituan announced that Wang Huiwen resigned from the company's non-executive director and member of the nomination committee of the board of directors due to personal health reasons. On the evening of June 29, Meituan issued an announcement to acquire all the rights and interests beyond Light Years.

Fu Sheng's passion and Wang Huiwen's departure due to illness are like two sides of a large-scale entrepreneurship: one is the infinite enthusiasm in front of opportunities, and the other is the realistic pressure that is not humane outside the spotlight.

Money is the main source of stress, especially for large-scale entrepreneurship known as the "gold swallowing beast". Some people in the industry once concluded that without a start-up capital of US$1 billion in China, it is impossible to participate in this large-scale entrepreneurial competition. Ma Qianli also mentioned that whether it is the model layer or the application layer, AIGC startups have particularly high requirements for core elements such as capital, computing power, and talents.

The second biggest pressure may lie in "volume", especially at the application level. The founder of a company that focuses on the implementation of artificial intelligence technology education scenarios in overseas markets told a reporter from Beijing Business Daily that "there are too many AI companies in China, and the volume is not enough." In addition, he also emphasized that capital and technology are problems, and there is no business model.

Anxiety about rapid technological iteration is also likely to resonate. Ma Qianli said that the development speed of AI is too fast, and the endless application layer and model layer are constantly improving, and it is a global competition. "Maybe the research results of a university in Washington that morning can affect the whole world that night."

Wang Peng, an associate researcher at the Beijing Academy of Social Sciences, analyzed to a Beijing Business Daily reporter that the pressure on AI entrepreneurs mainly comes from three levels. The first is the technology itself, including whether the chosen technical route and track are sustainable, more competitive, and more effective. The second is resources. Large-scale model training requires data, computing power, and talents, all of which require long-term financial support. The third is industrial application. Even with funds and a well-established team, whether industrial application can be formed in the future, whether the market can pay for it, and whether it can be responsible for investors, investment institutions and shareholders are completely different things. .

"The core problem of AI entrepreneurship is that it is easier said than done." Shen Meng, executive director of Chanson Capital, concluded. Regarding entrepreneurship in the field of AI, Shen Meng also cited an example of Nvidia.

Although Nvidia seems to "dominate the world" now, if you deduce its growth process, you will also find that it was on the verge of bankruptcy. Behind its success, countless similar companies have closed down. This cycle cannot be sustained by ordinary investors or investment institutions. After all, the market competition in the big waves is very cruel. It is difficult to determine whether the current choice is the future Nvidia or the one that will be drained.

“It’s like survivor bias, where all we end up seeing is survivors and think we should emulate survivors, but the problem is that survivors may have experienced a lot of intense competition that we haven’t seen. If we only target survivors , is likely to fail." Shen Meng said.

Entrepreneurship and investment, boom and bubble

While large-scale entrepreneurship enters the second half, the attitude of investors in the primary and secondary markets may also usher in a turning point.

On June 28, the artificial intelligence sector once fell by more than 5% in intraday trading. According to media reports, although this sector has become one of the hottest investment lines this year, there are signs that the sector is cooling down. The recent adjustment of the AI sector is mainly related to the "get off" of some investment institutions after making profits. However, the analysis also pointed out that in the medium and long term, the industry is still in a stage of rapid development.

A crazy research story has also been outlined in the primary market, and the number of AI projects exposed to it every week or even every day has suddenly increased. So much so that in the cognition of the outside world, investors seem to generally believe that this wave of AI is different from the previous hot spots, and this brewing may be an epoch-making opportunity.

Shen Meng believes that in the past period of time, the investment boom in the AI field was more due to the changes that ChatGPT brought to AI, coupled with the rising stock price of Nvidia, the entire international level of funds may be paying attention to AI, not Chinese investors have this preference, on the contrary, this is a global trend, "under such circumstances, large models, GPUs, etc. have become the most eye-catching investment targets."

In an interview with a reporter from Beijing Business Daily, a partner of an investment institution also mentioned that the investment market is indeed relatively hot. Some funds sell more, but some funds are in a state of seeing more and investing less, which may have something to do with the positioning of the fund or The background of the practitioners.

But just like whether AI entrepreneurship has sparked a bubble, the investment field has always been unable to avoid this topic. So in the near future, news that investors' attitudes have turned cautious has gradually appeared in the newspapers.

An Guangyong, an expert from the Credit Management Committee of the Quanlian Mergers and Acquisitions Association, told the reporter of Beijing Business Daily that some changes can indeed be observed in the attitude of investors. The main reason is that there are certain restrictions and regulations on AI research and technology development in China, which makes Investors are more cautious about investing in the AI field. The requirements of policies and regulations on data privacy and algorithm security have also increased the compliance pressure of enterprises.

In addition, the source technology is not in China, which causes investors to have certain concerns and uncertainties about the future development of the AI field. And despite the huge commercial potential in the field of AI, it also faces many technical, ethical, legal and other challenges and risks. Investors pay more attention to the commercial viability, core competitiveness and long-term development prospects of start-up companies, rather than just chasing hot spots.

Angel investor and senior artificial intelligence expert Guo Tao also analyzed that as the craze gradually fades, investment institutions have realized that the AI track faces large investment, long return cycle, low success rate, fierce competition in the industry, and increasingly stringent supervision. It is difficult for start-ups to compete with giants with advantages in technology, data and ecology, and investment attitudes are gradually becoming more cautious.

"Now is a very important watershed for AI investment." Shen Meng judged that AI investment can be divided into three levels. The first level is the basic key, similar to GPU, large models, etc.; the second level is to apply AI technology to transform oneself Core business and core products; the third layer is to judge whether AI will form incremental performance, which will not affect its basic business and generate additional revenue.

Due to the high investment requirements, the first layer is usually not set up by general investment institutions. The second layer may also face the risk of changing user experience and user habits and failing to get good market feedback. In comparison, the third layer has less risk and at the same time retains the leverage of the AI business on its overall business. This product may be more feasible and reliable for investment.

In his opinion, the project with the most investment value is a model trained with private domain data based on the technology of the existing model. These private domain data are the most important manifestation of differentiation, and unique parameters can be used to create unique models. Moreover, the accumulation of private domain data mainly depends not on a large amount of investment, but on time, without taking great risks.

From the perspective of investment, Lin Laini generally focuses on several key points: in terms of product strategy, whether new market opportunities have been created and the opportunity hypothesis is verified; if a large model is used, attention should be paid to the accuracy, stability and Iteration speed; resource endowment and successful cases of entrepreneurs, organizational structure of the entrepreneurial team and future organizational structure concept; GTM strategy of the project, who is the target customer, what value is provided for it, business model, product matrix, pricing strategy, Customer resources and sales forecast, etc.

"To sum up, what investors are concerned about is how much value the commercial application of the application of the large model may create, and whether entrepreneurs can build a moat." Lin Laini said.