Don't talk about AGI with "Yiwu people" | DeepKr Lite

Text | Zhou Xinyu

Editor | Yang Xuan, Su Jianxun

In 2023, two forces are simultaneously flowing towards the AI trend.

The first force is glamorous and conspicuous. You can easily find these people at high-end exhibitions like the "World Artificial Intelligence Conference": they are from prestigious universities, have published articles in top journals, and have overseas experience.

In their conversations, grand topics such as "AGI (Artificial General Intelligence)", "Surpassing OpenAI", and "Killer App" often come up. If asked "Does AI make money?", they will tell you: It's too early to tell.

But to find the second force, you need to stay away from Beijing, Shanghai, and Shenzhen and go to more down-to-earth places.

For example, in Yiwu, a county-level city famous for its "City of Small Commodities" and with an image that has nothing to do with high technology, the east wind of AI has not blown into the conference room of an investment committee, nor is it in a high-end office building.

But when you enter a high-end restaurant, you will probably see several local young people holding tablets and wine glasses, demonstrating AI-generated product image technology to merchants in a private room.

According to Tianyancha statistics, since 2023, the attractiveness of Beijing and Shanghai to AI startups has been declining, while the number of AI new companies in non-first-tier cities has been increasing.

Graphics: Intelligent Emergence

In Yiwu, "AGI" and "Killer App" do not impress customers. The only measure is whether it can reduce costs and make money.

A MCN manager in Yiwu recalled to "Intelligent Emergence" that she once interrupted an engineer selling the so-called "AI Agent" in person: "Young man, just tell me directly, how much money can be saved by using a virtual anchor for a month."

When Feng Yawei from Wuhan rushed to Yiwu with an office assistant that "covers the shell" of GPT, he clearly knew that his resume was not outstanding in Beijing, Shanghai, and Shenzhen: he graduated from a non-computer science major in nuclear energy; his previous Internet entrepreneurship project ended in failure due to commercialization.

But Yiwu merchants do not care about other things, but whether Feng Yawei's AI product can manage the complex process of the live-streaming room more efficiently.

The 200 business introduction brochures brought from Wuhan were snatched up in less than an hour. Feng Yawei hurriedly ran to the print shop and printed several hundred more.

Feng Yawei explaining the AI product to Yiwu merchants. Source: Provided by the interviewee

If "AI" is really useful, the benefits in the lower market are real.

Two men entered Xia Nan's office carrying several large bags, which were filled with stacks of 100 pink banknotes. When placed on the table, they made a dull thud.

Xia Nan opened the bags and quickly counted them. There were approximately forty to fifty stacks.

Those who brought cash to the door were local "bosses" of MCN in Fujian. They used cash to show their sincerity in order to purchase Xia Nan's AI face-swapping technology.

In mid-2023, short dramas going overseas were very popular, and these bosses urgently needed cheap and efficient AI face-swapping technology to replace the faces of Chinese actors with more popular foreign faces that overseas audiences prefer.

The cool plot of grassroots counterattack has come into Xia Nan's reality. "I've never seen so much cash in my life," he described. Two years ago, he was still a frustrated person who returned to his hometown in Fujian from Hong Kong after struggling for 6 years but failed repeatedly.

The lovable aspect of the lower market is precisely that it is extremely inclusive of the resumes and business models of AI entrepreneurs.

Before being labeled as "AI entrepreneurs", the identities of "Xia Nan" and others are diverse: graduates who can't find jobs, employees who have been laid off from large companies, and even Douyin anchors and Taobao merchants...

AI terms in the mouths of elites such as Langchain and Agent are mostly not understood by these entrepreneurs. But they have a stronger ability to implement and are willing to do the "Dirty Work".

Qu Kai, the founder of the FA institution 42 Capital, summarized to "Intelligent Emergence" that the characteristics of these people are: While large companies and star teams are still struggling to think about whether it can be successful, he has already made money.

Catching the Opportunities Beyond "Wang Xiaochuan" and Others

In the corners that the elites cannot see, everything is requiring AI entrepreneurs to break the inertia of traditional entrepreneurship.

"In doing AI business here, you can't be too 'academic' and 'theoretical'. Just think, can Wang Xiaochuan do business with e-commerce anchors using the 'Life Model'? " An FA told "Intelligent Emergence", "The BP (Business Plan) cannot help you make money."

Leaving aside the BP, how to quickly "do business" with people is instead a skill that many grassroots entrepreneurs are good at.

What they are selling is not only technology, but also emotions and a "human touch". For example, a Wuhan couple who started an AI self-study room in Meishan, Sichuan, told "Intelligent Emergence" that they spent half a year learning the Meishan dialect and playing mahjong to a certain extent.

"Not every entrepreneurship has to be a 'ByteDance'. " Li Kaifu once said, "You may have made a profit in the first three months. Of course, this may not be a scalable profit and cannot become a listed company, but this is also entrepreneurship."

In Yiwu, compared to competing with large companies and star entrepreneurs, Feng Yawei is more afraid of the locals who do AI business.

The locals are much more sensitive to changes in the water temperature than Feng Yawei and others. A local merchant mentioned to "Intelligent Emergence" that as early as mid-2022, when AIGC became popular, the second-generation store and factory owners in Yiwu who were supposed to inherit the "family business" began to "climb over the wall" to engage in model API. "Locals understand the needs of locals better."

In February 2024, a merchant in Yiwu International Trade City plays a multilingual product introduction video made by AI. Source: Xinhua News Agency

The path for ordinary AI entrepreneurs to discover business opportunities is often not through precise market research, but by relying on sensitivity to quickly grasp the vivid needs of neighbors.

Having come from Hong Kong to Fujian for 8 years, Xia Nan has done mobile games and virtual human live-streaming. Each entrepreneurial venture ended without a hitch due to either the epidemic or policies.

In early 2023, the opportunity came. At a dinner party, an old MCN customer mentioned that they wanted to use AI face-swapping to produce short dramas going overseas. Xia Nan immediately got to work. It only took him a few weeks to pull a few open-source models and transform the digital human generation platform left by his previous business into an AI face-swapping platform.

Then, he used the AI face-swapped video to post on Moments to promote himself and also gave the product to several old customers for a free "trial".

A few days later, the news that Xia Nan started doing AI face-swapping quickly spread in the Fujian MCN circle. Xia Nan received more than a dozen friend requests every day. "The short drama business has increased several times, and the unit price is not high, but the demand has increased."

What's more, what allows entrepreneurs to earn their first pot of gold in the AI trend is their own needs.

Zhang Luyu, who dropped out of school in the second year of junior high, registered his AI company Dify in Suzhou, a city that seems a bit marginal in the AI trend, deliberately avoiding the elite-packed Beijing, Shanghai, and Shenzhen.

"The consensus among these investors and star entrepreneurs in Beijing is too strong. They get together every day to talk about the same things. But isn't entrepreneurship about doing different things?"

The best methods he believes in finding an entrepreneurial direction are two: First, see if his past abilities can match an unmet need. Second, start from his own real needs.

"I can't empathize with the needs of others very well. Just like when you open a restaurant, some people tell you that they want to eat hot pot, but if you don't like it, you really can't open a good hot pot restaurant." He told "Intelligent Emergence".

But does his own need represent a huge market scale? Zhang Luyu does not have a definite answer.

It's just that he thinks that if he follows the logic of traditional entrepreneurship and calculates the market size clearly, he will have to set up various hypothetical enemies in the market competition - a big shot enters the field, and a large company comes to seize the market - "From this perspective, everyone should simply not do it."

On November 31, 2022, the night ChatGPT was released, Zhang Luyu chatted with this "silicon-based new species" all night long.

However, he soon realized that it is quite difficult to develop new applications using ChatGPT's API. "Not including the engineering aspect, at least you need to continuously debug the Prompt and it takes three to four months." But at that time, there was actually no tool to assist in the development of large model application.

Just like the apple hitting Newton, he slapped his head: "Isn't this an entrepreneurial opportunity?"

Just Go Where There Is More Money

In April 2023, one month before Dify was launched, Zhang Luyu tentatively met with two or three investors. From their lackluster responses, he knew that financing was definitely out of the question.

"This round of financing is completely different from the previous round of mobile Internet financing, unless your background is very outstanding." He told "Intelligent Emergence". Product data and commercialization achievements, which were once the requirements for the second and third rounds of financing, have now become the stepping stones for the first round.

In an interview, Li Kaifu once said outright that the current financing and other environments are the worst in the past decade. Qu Kai told "Intelligent Emergence" that in the capital winter, investors are more stable and more trusting of verified entrepreneurs - that is, the so-called "elites".

The realistic pressure of financing makes commercialization an inevitable proposition that grassroots entrepreneurs have to consider from the first day. This also makes the current AI businesses of these ordinary people often have beautiful revenue data.

The "Enterprise Alliance AI" that Feng Yawei and his team pieced together from the open-source community has a stable monthly income of several hundred thousand yuan; Xia Nan's company's current revenue has allowed Xia Nan to move to a better location with a more expensive office building. Opening Jike or Xiaohongshu, stories of "Monthly income of 100,000 yuan" and "Annual net profit of one million yuan" happen to many ordinary people who do AI design and AI writing.

Relatively, "At present, the ARR (Annual Recurring Revenue, annual recurring income) of most AI companies in the market is not even as high as the daily wage of a 400-yuan domestic helper." Qu Kai summarized.

For many star entrepreneurs, whether the market is overseas or domestic is a difficult problem that needs to be carefully considered at present: Which side's financing to take? Which side's users to target? Where to register the company entity?

But for grassroots entrepreneurs, there is only one correct answer in their hearts: Just go where there is more money.

In 2018, when Zhang Luyu started an entrepreneurship to develop a developer collaboration product called "Moth", he realized that SaaS cannot survive in China. Although the product has a good reputation, the income poured cold water on him, "The gap between the income and my ideal is very large."

After a painful reflection, Dify chose to go overseas. When the product was just developed, Dify made three language versions at one go: Chinese (simplified and traditional), English, and Japanese. What if no one in the team is proficient in Japanese and English? Then use Google to translate word by word.

In Zhang Luyu's view, the United States and Japan are the two markets that SaaS products must not miss: "The United States goes without saying, and in the Japanese market, Chinese people are more accessible, customers have strong payment ability, high loyalty, and a high concentration of global companies."

At the beginning, Dify did not have staff overseas or funds for promotion. For overseas business, Zhang Luyu set his expectations very low, "I didn't expect the data to be very good."

To his surprise, an article with the theme of "Uncovering the Team Behind Dify" became popular on the developer forum Note, known as the "Japanese GitHub".

With the boost of public opinion, less than two months after Dify was launched, overseas businesses came in one after another. During that period, the company always burst out with cheers: "Another cooperation email has been received" - Looking at the sender, it is another Fortune 500 company.

However, even though Dify has gained some reputation, how to charge fees still needs to be carefully considered. "If you charge too early, growth will stagnate, if you charge too late, the business model will not work, and users will assume that you are free first."

In terms of making a profit, the consensus that can align elites and grassroots entrepreneurs is: The premise of charging is the user base.

It wasn't until 7 months after its establishment that Dify's global installations exceeded 100,000, and Zhang Luyu felt that it was time to charge, "The market has an understanding of your product, and you can grasp the psychological price of customers. At this time, it is almost time to set the price."

The pricing of Enterprise Alliance AI was also urged by customers. Their office intelligent assistant was launched in May 2023, and in that month, they received nearly 100 enterprise trial application requests. Less than a week after the trial, the enterprise's repurchase rate was close to 100%.

The money came several months earlier than the several co-founders expected, and there was not even a payment plan on the official website of Enterprise Alliance AI. They simply and roughly referred to the pricing of a similar tool and announced the charging model on the official website. Later, by sending several months of benefits to old customers and collecting a wave of feedback, Enterprise Alliance AI gradually figured out its own market pricing.

In the Spring Festival of 2024, Zhang Luyu was pleasantly surprised to find that: 8 months after its launch, Dify has made a profit, with a gross profit margin of 80%, and overseas income accounting for nearly half.

He remembers that in the month when Dify was just launched, in order to raise funds for technology iteration, he flew from Suzhou to Beijing and met with more than 30 investors in one week.

At that time, the business plan was also very simple: a screenshot of an iPhone memo with a few lines briefly written: Who am I? What product am I going to make? How is the product data?

At that time, he set his financing expectations very low. He thought that if he talked to 20 companies, as long as he had a 5% success rate, it would be enough.

Now, facing