At the end of last year, Sogou’s former CEO Wang Xiaochuan was convinced that “AGI is here” after just experiencing ChatGPT a few times.
Later inEntrepreneurshipAt an online exchange meeting of the AI community, when several entrepreneurs in the AI field defined ChatGPT's progress only in terms of function, Wang Xiaochuan made it clear that "everyone underestimated this matter."
Wang Xiaochuan is not the only one who believes that ChatGPT can revolutionize the world.Large ModelEntrepreneurs in this field all expressed this view in an interview with Tech Planet.
In business history, before ChatGPT ignited this wave of entrepreneurship, the last round of revolution triggered by artificial intelligence technology had gradually calmed down after nearly ten years of development.
Source Note: The image is generated by AI, and the image is authorized by Midjourney
The four AI unicorns that were once born with a silver spoon in their mouths: SenseTime, CloudWalk, Megvii, and Yitu, have fallen from the darlings of capital to questioned projects due to long-term losses and insufficient hematopoietic capacity, which led to the market's disenchantment with AI.
Therefore, when ChatGPT once again set off a wave of entrepreneurship, every entrepreneur involved will hear more doubts from the outside world: Can you implement this? What is the use? Don’t be a scammer... A large model application entrepreneur told Tech Planet: "There are too many such voices."
Compared with the previous entrepreneurial environment, today's market and investors are more realistic. While they see technological innovation, they also need to see the business space behind the project. "We don't deny the technological revolution, but in the business world, the ultimate mission of technology is to land and bring commercial value," said the CTO of a listed company.
The harsh entrepreneurial environment does not seem to have dampened the enthusiasm of entrepreneurs. There are still countless entrepreneurs who choose to catch up with this round of opportunities: some of them resigned from their jobs in the United States and returned to China, some started their research while in school, and some entrepreneurs who have been successful before...
Li Lingfei, founder of Photon City, said that the last wave of AI did not explode because it was a customized technology, while the big model is a general technology, and the marginal cost of such general technology in commercialization will be lower, which is actually what the new round of entrepreneurs firmly believe.
However, they also know that it is not easy to seize opportunities in this environment.
Making large model bases is not a game for entrepreneurs
At the beginning of this year, Li Yu (pseudonym) returned to China from the United States and started his second entrepreneurial journey.
FirstThe first startup was also in the Chinese market. In five years, he received financing from multiple institutions and had more than 6 million users and 100,000 corporate users. Later, considering the prospects of commercializing the tool, he chose to sell the company to a large manufacturer.
Although the outcome of going public did not conform to mainstream values, this entrepreneurial experience gave Li Yu a complete experience of domestic entrepreneurship and made him more calm in this entrepreneurial process.
At that time, domestic large-scale model entrepreneurship focused on the foundation, and funds and talents poured into the underlying large-scale model construction, but Li Yu disagreed.
In his opinion, compared with the PC era, the giants of the mobile Internet era emerged in a specific era, but today the dividends have disappeared. These giants that have already occupied the top spot in fixed fields have funds and resources, and are more difficult to be subverted by start-ups.
One piece of data confirms the crowdedness of the big model startup market. So far, about 200 big models have been released in the domestic market, and more than 20 big models have been approved. Baidu founder, chairman and CEO Robin Li mentioned in a forum that the combined call volume of more than 200 big models is not as much as that of Wenxin Big Model alone.
This is a warning signal. Perhaps in the future, large model suppliers in the market will soon face the problem of high costs and imbalanced income. Li Lingfei and Li Yu realized this very early: the base is not a game for entrepreneurs after all.
In their view, giants have more advantages in computing power, capital, and even data, and have highly mature commercial monetization systems. It is not wise to compete with giant companies for the big model base.
An investor also told Tech Planet that startups have limited resources, and vertical applications in the industry chain are their main battlefield. However, due to differences in experience, each entrepreneur sees different product routes in vertical applications.
"Redo what AI does on the large language model." With this belief, Li Yu saw the opportunity: "Machines have lowered the threshold for ordinary people to become programmers, so can I make a tool to allow more entrepreneurs, even those who can't program, to participate in this wave of entrepreneurship?"
Li Lingfei has worked for ByteDance's overseas and Huawei teams. He not only has many years of experience in artificial intelligence, but also has a deep understanding of the legal industry. Based on this, he created a personalized AI assistant "Hai Rui Zhifa" for the legal industry, and continued to extend it to areas beyond the law.
Although there are differences in the way big models reconstruct different products, entrepreneurs all have a common starting point, which is to improve the original human-computer interaction method and improve efficiency.
Technology, direction and team
Yingmo Technology has been out of the circle for a long time. In 2021, a face-pinching app called Wand topped the App Store's graphics and design rankings.Firstname.
Based on generative AI technology, this is theFirstThis work: The user's simple sketch on the canvas is used to generate a cute cartoon avatar through AI algorithm. This is an APP with playability, but that's all.
This put the founding team in a state of distress. "Most of the products we developed only met academic standards, and rarely considered real market needs." Zhang Qixuan told Tech Planet that the technical direction of the research and development at that time was difficult to bring commercial returns to Yingmou Technology.
"If you want to retain users, there are only two options. One is to turn this product into a tool, and the other is to turn it into a community." But after repeated evaluations, Zhang Qixuan and his team felt that neither was a feasible path.
"We could neither think of a good charging model nor a way to balance the users' computing power expenses." So, they finally chose to give up.
Before starting the new project, Zhang Qixuan held a long meeting with the team, the core direction of which was the focus of the new project.
Having learned from the lessons of the previous project, the team has become more pragmatic and has focused on R&D in the field of digital humans, developing system technologies that are compatible with existing film, television and gaming businesses.
At the end of 2022, Yingmou Technology completed the development of the Dreamface product, and this year launched a product called ChatAvatar based on the big model. Dreamface has promoted commercialization since its birth, and ChatAvatar has also entered the public beta stage.
The idea of developing ChatAvatar came from a project the team was trying to do. "When we were working on a project internally, we had to give out more than 100 character examples. We found a problem: it's difficult to describe the appearance of one person, let alone more than 100."
So Zhang Qixuan and his team used ChatGPT to successfully complete the construction of the role sample. As a result, Zhang Qixuan and his team added a Chatbot to the product as a tool to improve efficiency. ”
Many entrepreneurs believe that in this wave of entrepreneurship in ChatGPT, the free model isabsoluteIt’s not feasible. “This is not a path for startups to achieve a virtuous cycle.”
In addition to the technical path and direction, another challenge comes from team building. Zhang Qixuan chose to start a business since he was a student, while Li Lingfei has been working in a large company.Firsttimes of entrepreneurial experience.
"Building a team is a long process." Zhang Qixuan told Tech Planet that outside of functional departments, they are more rigorous in selecting technical talents. They all hope to find like-minded talents with relevant work experience.
In the two years since its establishment, Yingmo Technology has expanded from eight people to about 30 people. Li Lingfei's team has expanded from four people to about 15 people in half a year.
From the moment we opened, we were doing business.
In the past AI startup boom, investors and entrepreneurs loved to talk about "long-termism." They talked about R&D time, R&D investment, and R&D talent, but rarely mentioned revenue and profit in the early stages.
This is the traditional trilogy that entrepreneurs go through, from technology, products to commercialization. But new changes have emerged. In the entrepreneurship of large-scale model applications, more and more entrepreneurs have chosen a more "pragmatic" path: they start to conceive the commercialization model from the beginning of their business. They want to innovate and make money.
In the era of industrial Internet, the profit model of some artificial intelligence and SaaS products is often to earn revenue through the B-side: providing a relatively standardized product and opening an API interface for government and enterprise customers.
However, this model can easily evolve into a customized project at the back end. An investor told Tech Planet that paying for products is the logic of expanding the business scale. If you pay for projects, it will undoubtedly be a huge challenge for startups in terms of cost and scale expansion.
According to Sensor Tower data, in the first half of 2023, the US market contributed 55% of total AI application revenue, the European market accounted for 20%, and other markets including China accounted for only 25% in total.
The core founding team formed by Li Lingfei includes algorithm engineers from Facebook and hardware companies, as well as professional legal professionals from Red Circle Law Firm.
Before deciding on the product direction, Li Lingfei and his team chose to work with legal professionals to polish the product based on the core and practical needs of the target users. The same is true for Yingmo Technology. When the product was launched, Zhang Qixuan and his team gainedFirstWe have selected Unity for each of our customers and are continuously improving our products through co-creation with them.
The direct business model allows them more freedom in customer choice.
For example, Li Lingfei does not want to be a pure To B business, so he does not dwell on the B-end and C-end markets, but directly targets professional consumers. Yingmo Technology's SaaS service platform will also be open to ordinary users. "Ordinary users here have a common feature, in fact, they are all professional users with a certain professional background."
In Li Lingfei's opinion, a large-scale model application with the potential to become popular must first be interesting. "Just like WeChat, which broke the circle because of the shake function." And for users, it must be a useful productivity tool, based on which the logic of paying will arise.
At present, the products of Li Lingfei's team are divided into regular version and professional version, and fees are charged according to the account. Among the nearly 30,000 customers, there are ordinary users, professionals, and some law firms and legal groups.
As Zhu Xiaohu, managing partner of GSR Ventures, said, when starting an AI business in China, one must consider the scenarios in which the business can be implemented andFirstEvery day you can make money.
Whether it was Zhang Qixuan or Li Lingfei, they all realized the same truth.
When Fundraising Goes Quiet
AI unicorn companies have received a lot of hype.
The most typical case is that Xu Xiaoping of Zhen Fund, Shen Nanpeng of Sequoia Capital, and Feng Bo of United Capital Investment talked about the future valuation of Deepin Technology. Xu Xiaoping said it would be at least US$500 billion, Shen Nanpeng said US$100 billion, and Feng Bo gave a compromise price of US$300 billion.
But six years later, Green Vision's market value is less than 6 billion yuan today.
For a long time, the commercialization problems of leading AI companies have not been solved. The influx of hot money has become lines of code piled up on technology platforms, which seem to be unable to be realized forever. Therefore, no one mentions the former star unicorns in either the primary or secondary markets.
The more profound impact comes from the rift in "trust", which is a two-way rift.
In the past, investors, especially US dollar investors, dressed in fancy clothes and ran around the first-tier cities of Beijing, Shanghai, Guangzhou and Shenzhen. They were the guests of honor of entrepreneurs. With a stroke of the pen, money flowed from one account to a project. But this scene is no longer the case. A set of data from the secondary market shows that from January to August 2023, a total of 247 A-share IPO companies were listed, of which 225 were approved, 12 were rejected, 8 were suspended, and 2 were suspended and then withdrawn.
An investment manager said, "It's like returning to real life. Every investment must be carefully calculated."
As investors become more stringent in screening projects, entrepreneurs in the big model era also have new views on financing.
At the beginning of his business, friends in the investment circle told Li Lingfei that if he could run the business with his own money and continuous income, he should not consider financing. "If I have to give up my voice just to get funding, that's not what I want to do."
Since the middle of this year, more than 20 investment institutions have found Li Lingfei, including sometop notchInvestment company. Compared with getting money, what makes Li Lingfei more excited is that his entrepreneurial direction seems to have been recognized by the capital market.
Incubated at ShanghaiTech University, Yingmo Technology has completed two rounds of financing, with investors including Qiji Ventures and Sequoia China. In terms of the selection of future investors, Zhang Qixuan's requirements are becoming more and more clear. He hopes that investors will not only provide financial support, but also give him a broader vision in terms of strategic direction.
Whether it is Li Yu who returned to China from the United States at the beginning of the year, or Li Lingfei who chose to resign from Huawei to start a business in the middle of this year, they have made great determination. On one hand, there is a comfortable life, and on the other hand, there is the dream of entrepreneurship.
But when the competition among big model entrepreneurs begins, what makes them more certain is not to be a bystander in this wave.