The survival experience of a Chinese start-up model company Doctor AI: Sink down, live, don't stand in the cracks

Every reporter: Ke Yang Every editor: Zhang Haini

Just imagine, as an entrepreneur of a large model company with an application layer, how to survive the year of the "100 model war"?

Flying up and down in the boiling wave of artificial intelligence, this feeling is mixed with luck mixed with confusion. Luckily, almost everyone says it's a time of unavoidable opportunity; Confusingly, no one knows exactly where the super scene that can be "monetized" is.

Visual China Diagram

A year ago (June 27, 2023), he and his partners founded Beijing Yizhi Information Technology Co., Ltd. (hereinafter referred to as Yizhi AI) to start the entrepreneurial journey of AI+ NLP's large language model. This entrepreneurial team of medical and health models from the Department of Computer Science of Tsinghua University uses MoE architecture large models + AI Agents to target the daily health management service market outside the hospital, trying to explore the application of large models in the field of medical and health.

Zhu Xiaohu, a representative figure in the field of venture capital and managing partner of GSR Ventures, has repeatedly explained his preference for the application layer of large models and his confidence in China's innovation in the application layer. But the fact is that when the story of large-scale entrepreneurship becomes crowded and saturated, few people turn their attention to non-well-known entrepreneurs. In the middle of the "100 model war", start-ups continue to play with investors, large manufacturers and themselves.

Competition is only going to get tougher

Earlier this year, Liu Chenghui was approached by a large company to combine the technology of Doctor AI with the company's products for in-depth customization. Prior to this, according to the initial assumption of the company, the company's path should be to focus on "solutions" and provide standardized products and services for offline chain organizations in the target industry.

But he finally decided to take the order, after all, "survival is the first priority".

The end of the "100 Model War" seems to be more distant than expected. Turning the clock back to 2023, in the face of the upcoming "100 model war", the first misjudgment the market may make: this race will be won and lost in 2024.

According to the incomplete statistics of the reporter of the "Daily Economic News", as of the end of April this year, a total of 305 large models have been launched in China. Among the more than 300 large models, excluding the large models from the leading manufacturers, among the startups, there are five "unicorns" such as the dark side of the moon, Baichuan Intelligence, Zero One Everything, MiniMax, and Zhipu AI.

Today, the only sure answer is that this pattern will not be the end. "Everyone has always thought that the 'war' will end quickly, and there may be only a few companies left in 2024, but these head large model companies are still continuing to raise funds, and who is the king of China's pedestal, the competition process may be longer than imagined." Liu Chenghui said.

The "war" has continued to this day, and Liu Chenghui feels that on the one hand, because the large model is one of the few investment targets that is still good, hot money has poured in again; Another reason is that the gap between the current domestic large model and the foreign head company does exist.

At present, the large-scale model "war" is still in full swing, and even if the base layer dies down, the application layer will not stop. There will only be more and more large model companies in the application layer, and the competition will only become more and more fierce. Liu Chenghui said: "This can never be stopped, just like making an app, there will be people in each track of the application layer large model, until the giant is produced, unless this era is over." ”

Sink to a more subdivided track

"First of all, medical care is a must for large factories, there are definitely opportunities, but it also depends on the direction." In Liu Chenghui's view, serious medical care and consumer medical care are two distinct paths.

In the field of health, serious medical care is a relatively difficult end, because it requires a large amount of hard-to-reach data support, has a long business model chain, and is mainly oriented to enterprises and government customers, which is not suitable for start-ups. The path chosen by Doctor AI is consumer medical care, which serves end consumers together with B-end (commercial) customers, that is, offline and out-of-hospital private consumer medical service centers such as Meinian Health, Happy Mouth, and Little Apple Pediatrics.

There is still a lot of room for development in the role of family doctors in China. Liu Chenghui observed that the daily needs of the health field are very rich and not limited to hospital diagnosis and treatment. In the past, Internet platforms could often only cover part of the market demand, and the significance of the existence of Agents is to make up for the shortcomings of traditional Internet platforms and effectively integrate various health needs.

Liu Chenghui believes that the best thing to do in the application of AI is health. He believes that good landing applications are divided into two dimensions: one is to generate revenue quickly, and the other is to have long-term value, and health applications have both advantages.

The startup's response is to sink into a more "specialized" track. "The logic of doing the application is that it must be very specialized, otherwise where is the barrier? The model layer will definitely crush your living space through model upgrades, so you have to be more specialized, more sinking, and more integrated with the business. Liu Chenghui said.

A more desirable investment comes from the B-side

Although a new round of funding is being brewed by Doctor AI, Liu Chenghui has decided to reduce the frequency of receiving investors: "It doesn't make sense. Talk about it every day, but it just doesn't invest in the application layer. ”

In 2018, Liu Chenghui had a successful entrepreneurial experience in the field of AI, and he had done a national-level project and reached a user base of more than one million. For this venture, Liu Chenghui is more handy. However, compared with the previous period of entrepreneurship, the current investment environment is more severe, the "water temperature" has changed significantly, investors are more cautious when looking at the project, not only looking at the growth space and future value when inspecting the project, but also have higher requirements for short-term income.

Based on the large-scale model entrepreneurship project, Liu Chenghui's observation is that the core of investors' current attention is actually very clear. First of all, it must be a "company on the table", and the core of the "company on the table" is Qingbei and other university departments, because this represents a deep technical background. Otherwise, it will be difficult to win the favor of top investors regardless of performance.

Second, companies in the application layer must achieve stable revenue, which means that companies must truly create value for customers. The size of revenue is not the only criterion, but companies must ensure that revenue is standardized and replicable at scale, which is a challenge for most companies. "There are still very few scenarios for real application now." Liu Chenghui said.

In the process of this round of financing, Liu Chenghui feels that the company does not necessarily need to introduce financial investment, and the more ideal investors come from the B-end customers served by the company - they have a huge amount of capital and a large number of business scenarios, and at the same time, the requirements for investment return and exit mechanism are more relaxed than those of financial investment institutions, which is a benign state.

Do what generic models won't do

"Entrepreneurship can't survive with technology alone, we must first maximize the use of technological advantages to create business value." The doctor AI led by Liu Chenghui chose to face the B-side and solve the problem of survival first.

Liu Chenghui said frankly, "We are also crazy about quantitative models and crazy looking for cheap computing power." The question of cost is not very "sexy", but it can make the difference between life and death for a company.

In addition, AI seeks to find "precise positioning" in the market and strives to take root beyond the general model. In actual business, there are many things that general models do not do, such as co-developing customized models with enterprises and AI Agents for landing applications.

"There are limitations to the ability of pure application models, for example, there will be (with) a lot of rule systems outside our model, and various patches will be added, which will be integrated into the business scenarios of enterprise customers like capillaries, which must be there when we really want to industrialize and implement them in batches." Liu Chenghui said.

He used the appointment booking agents recently launched by Doctor AI as an example. Completing the appointment through the guidance of AI Agents is "a function that sounds like an idiot, it doesn't seem to have any technical content, but it is very important for customers, not only to connect data, but also to have professional knowledge, so that they can serve consumers like real people, which requires very deep cooperation." Liu Chenghui believes that this is the advantage.

In the race of large models, start-ups seem to be standing between large factories and "unicorns", but they are not necessarily passive.

What Liu Chenghui believes is that there must be some things that big factories won't do. In his opinion, the key element in the health field lies in trust, and through cooperation with various medical institutions to provide services to customers, trust can be established, and the medical and health field must be inseparable from real people and offline, which is something that cannot be pushed down by large manufacturers.

Secondly, he believes that as far as model application companies are concerned, they still need to actively build their own barriers. "This barrier is not the technical architecture, but the proprietary model of a certain subdivision scenario, and the B-side privatized data is the core; Another core of our is the user's private data, and in the future, we will launch each user's own small model. Liu Chenghui pointed out that for medical AI, these private data mainly refer to the user's long-term health management data. This kind of data does not exist in the first place, and can only be gradually accumulated with the help of AI, which is also the unique advantage of AI.

Liu Chenghui revealed that Doctor AI is working with Tsinghua Laboratory to develop AI glasses equipped with a large model of Doctor AI, with the goal of combining health data to provide users with exclusive services and become portable AI Agents, just like Iron Man's Jarvis.

"There is no doubt that it is inappropriate for entrepreneurial teams to deploy intelligent hardware now, but we also have to look up at the sky, and we can have a more competitive advantage by laying out early." Liu Chenghui said.

National Business Daily