ChatGPT is an entrepreneurial opportunity: rich model, no money application layer

Author: Guo Hongyan

Source: Titanium Media

Image source: Generated by Unbounded AI‌

After ChatGPT was born, the dialogue between human beings has become "technical".

As far as the 18th-line baby boy, asked, "I heard that ChatGPT can write manuscripts by itself, will you be replaced in the future?"; near the primary market, investors who have traded tens of millions of transactions, asked, "Your Can the product be combined with ChatGPT?"; and when I was 40 years old, I saw the advertisement in Moments, consulted the course "ChatGPT Improves Future Workplace Competitiveness", and asked, "Don't talk about salary increase, can learning keep me from being unemployed?"

ChatGPT for all people, some people embrace it, some people fear it, some people still don’t know why it continues to be hot, some people use the ChatGPT model to get 125 million US dollars in financing, and some people use ChatGPT to make the company’s ARR (Annual Recurring Revenue) reach 10 million Dollar.

So is ChatGPT's "rich-making" myth true? With ChatGPT not open to mainland China and Hong Kong, how can domestic entrepreneurs take advantage of this wave of new entrepreneurial opportunities? Will China have a latecomer advantage in the competition for large models?

ChatGPT detonated the AI boom

**ChatGPT is not the end, but the beginning of detonating AI. **

Once ChatGPT was launched, users around the world flocked to it. The number of users exceeded 100 million within two months, setting a record for the fastest Internet application to exceed 100 million. To know this data, TikTok used it for 9 months, Facebook used it for two years, and Instagram used it for 2 years. 30 months.

A devastating third technological revolution kicked off. Kuhn, a philosopher of science, once proposed that the development of science is not a smooth linear development model, but will be like walking up steps, step by step, just like the advent of ChatGPT, bringing artificial intelligence to a new level.

OpenAI CEO **Sam Altman predicts: AI large-scale model technology will become the largest technology platform in the future after the mobile Internet, and chatbots are used as the interface, plus images, music, text and other multi-mode With the development of the ecological model, a world-class large-scale enterprise will be born. **

The booster behind it, **Microsoft CEO Satya Nadella also stood up and said that the development of the field of artificial intelligence at this stage can be described as "exponential". **

Sun Zhengyi, who has retired behind the scenes, also publicly stated that he will end the relatively dormant state of all in artificial intelligence. "I am most interested in the artificial intelligence revolution. I believe that human beings will be surpassed by computers or artificial intelligence. We hope to be the leader of the artificial intelligence revolution!" At the same time, he said that he is a heavy user of ChatGPT, and also talked with OpenAI CEO Sam Altman talks a lot.

**ChatGPT is not the whole picture of artificial intelligence, but a form of generative AI. Sequoia Capital predicts that the wave of innovation in generative AI applications inspired by large models will be trillions of dollars in economic value. **

The primary market is also full of generative AI entrepreneurial atmosphere. In foreign countries, according to PitchBook data, since the release of GPT-3 more than two years ago, venture capital investment in AIGC has increased by more than 400%, reaching an astonishing $2.1 billion this year.

Generative AI venture capital situation, source: PitchBook

The Y Combinator W23 Demo Day Batch has a total of 268 projects,** of which nearly 50% are AI-related projects, mainly focusing on the middle layer and application layer, and the fields include generating text, pictures, videos and other fields. ** "This also reflects YC's investment orientation, investing in small but promising and innovative projects." In the opinion of Xu Siqing, the founding partner of Alpha Commune.

Domestically, between November 2022 and April 2023, AIGC’s financing scale will reach RMB 2.24 billion, equivalent to 41.4% of the total financing amount in the past two years.

After Sequoia China’s startup accelerator YUE opened registration on February 28, there were 300-400 AIGC themes among the 2,000 registered projects; 41 are artificial intelligence themes, double compared to last year, and 39 of them are related to large models. It can be seen that the attention of the primary capital market to the AIGC track has increased sharply.

Overall, the number and scenarios of generative AI entrepreneurial projects in China are far less abundant than in foreign countries.

An overseas AI entrepreneur told Titanium Media Venture Capitalists: "First, it is more convenient for overseas entrepreneurs to call the ChatGPT interface, which is convenient for fine-tuning and developing new products; Thirdly, open source communication communities like Reddit have a strong atmosphere, and everyone will put their understanding of ChatGPT and new applications on it for public discussion and innovation.”

In contrast to China, when ChatGPT is not open to registration in mainland China and Hong Kong, how can entrepreneurs and major manufacturers get on the train of the "third technological revolution"?

"Money big model, no money application layer"

Entrepreneurship is not only about dreams, but also about how much money is in your pocket. Computing power, data, algorithms, talents, and capital reserves are indispensable for building a large model, but this is not a cost that ordinary small and medium-sized startups can afford.

For example, computing power resources, the data shows that the total computing power consumption of ChatGPT is about 3640PF-days (it takes 3640 days to calculate 1 quadrillion times per second). It is said that 10,000 Nvidia A100 chips are the computing power threshold for a good AI model. At present, there are no more than 5 Chinese manufacturers with more than 10,000 GPUs, and among them, there is at most one company with 10,000 Nvidia A100 GPUs.

For example, funds, according to the Guosheng Securities report "How much computing power does ChatGPT need", the cost of GPT-3 training is about 1.4 million U.S. dollars, and for some larger LLM (Large Language Model), the training cost is between 2 million U.S. dollars and 12 million U.S. dollars. between million dollars. Based on the average number of unique visitors of ChatGPT in January of 13 million, the initial investment cost is about 800 million U.S. dollars, and the daily electricity cost is about 50,000 U.S. dollars.

No matter how expensive the cost is, the mission of the Chinese version of "ChatGPT" must be achieved. **"China will definitely have its own big language model, which is determined by the current status." **Xu Siqing told TMTPost Venture Capitalists.

China's major manufacturers are already gearing up and entering the state of a hundred models. Baidu, Ali, Huawei, Tencent, 360, SenseTime, JD.com, iFLYTEK, ByteDance and other giant manufacturers have all set foot in the research and development of large models in combination with their own business and strategic layout, striving to be the first "China ChatGPT", loaded History of artificial intelligence in China.

According to Leonis Capital, **China seems to be the only country other than the US and UK monopoly that develops its own model infrastructure. **First, in terms of the number of models, four of the top ten developers in the world are Chinese companies, including technology giants such as Baidu and Alibaba, top universities such as Tsinghua University, and the Beijing Institute of Artificial Intelligence (BAAI) and other government-sponsored laboratories; second, in addition to model training infrastructure, Chinese developers have also become independent at the hardware level. In response to the threat of US sanctions, Chinese companies are increasingly turning to domestic GPUs, such as Huawei's Ascend 910; Third, although the Chinese model is about a year behind the top Western models, it is only a matter of time before they catch up.

Xu Siqing also agrees with this: "Open AI is on top, and its technological breakthroughs have given them an absolute leading position in the field of AIGC. China's large models are quickly following, following the path explored by Open AI, and quickly Iteration and catching up; American companies are more at the forefront of exploration, which is determined by the characteristics of the companies of the two countries and the resources at hand.”

"We also have a point of view, that those who get the talent get the model, and those who get the model get the world. The models here include vertical models trained for specific scenarios, including ToC and ToB models trained and deployed for professional scenarios Compared with large models, we are more optimistic about this field. The reason is very simple. More than 98% of the computing power of training a model is spent on pre-training, but the next three steps of supervised finetuning, reward modeling and reinforcement learning only use a small amount of time. Part of the calculation time is long." Xu Siqing said.

"For small and medium-sized enterprises that do not have sufficient funds and do not have a reserve of cutting-edge technical talents, it is a good choice to explore opportunities at the application layer early and create some first-mover advantages." A Silicon Valley investor told TMTPost Venture capitalists said.

"**I think that the computing power of China's large-scale models is not suitable for ordinary entrepreneurs. Entrepreneurs who understand artificial intelligence have a great opportunity, because they don't need to start from the most basic model to compete for computing power. The comparison is knowledge accumulation, team engineering ability and Iterative speed, for superficial product innovation, this type of entrepreneur is a blow to dimensionality reduction. **Of course, technology is important, but industry experience and selected application scenarios are equally important. Our seed round investment in HiDream.ai (Zhixiang Future) project Mei Tao, the founder, former vice president of JD.com, and foreign academician of the Canadian Academy of Engineering, is based on this consideration.” Xu Siqing said.

In addition, Xu Siqing also pointed out that the breakthrough of AIGC technology has brought three levels of opportunities to Chinese entrepreneurs. First, is the opportunity of LLM, a big factory or the national team; second, it is to use a large model to do shallow application or product-driven application,**no matter To B or TO C, there are many opportunities , the characteristic is that it is difficult to judge who is easy to win, and the difficulty is like predicting which one will be successful at the beginning of the emergence of APP; The third is to establish a vertical model to solve the problem at the application level, which is widely used in ToB and ToC Prospects, because the threshold is high, the certainty is relatively strong. They are characterized by strong modeling ability, can work hard on the model, and solve problems that cannot be solved by product innovation.

**Finally, he emphasized that there will be many opportunities for infrastructure services associated with various language models and applications, such as privatized deployment, cross-cloud, and cross-model service tools. **

Foreign large models, AI applications, full of carnival

In Silicon Valley, Open AI wants to stand out, but Google, Meta, and Amazon cannot agree.

Recently, Amazon, which has been incognito, released the Amazon version of ChatGPT. The new model is called Amazon Titan. The Titan series models are divided into two types, one is a text model for content generation, and the other is an embedding model that can create vector embeddings. , for creating efficient search functions, etc.

Meta CEO Mark Zuckerberg did not play the cards according to the routine, directly penetrated the "low price", and declared the "hegemony" of Open AI with open source. Meta licenses the open-source large-scale language model LLaMA to researchers in academia. On June 16, according to foreign media The Information, Zuckerberg is considering commercializing a new version of the open source large-scale language model LLaMA, allowing other companies to use LLaMA for free to develop related AI software, and allowing developers to use these products for business use.

The war escalated. OpenAI announced that it plans to build an epic LLM "app store" to open up all ChatGPT applications. The platform can realize two-way docking. Developers can put their products based on OpenAI technology on the shelves, and enterprises can also use LLM on the market as needed to completely seize AI. model ecology.

"The big model competition in the United States must be full of flowers, and the degree of openness will become higher and higher. The history of IT development continues to prove this. From Unix to Linux, from iPhone to Android are examples. The leading openness is insufficient , Manufacturers in a slightly passive position will use openness to build their own ecology, which is a very positive phenomenon for the industry." Xu Siqing told Titanium Media Venture Capitalists.

Looking at the application layer, there are not a few cases of commercial verification. The free and open-source Stable Diffusion, a start-up company raised $100 million in financing in just one month, and its valuation soared to $1 billion; Jasper.ai, a SaaS company that builds applications based on GPT-3 API, was founded 18 months ago and won US$125 million in A-round financing, with a valuation of US$1.5 billion; Copy.ai, which helps companies write promotional copy, has an ARR of US$10 million in just 2 years.

"Overseas projects are different from domestic projects. One is that the capabilities of overseas large-scale underlying models are more mature (OpenAI's API); the other is that the commercialization environment is better, and users are willing to pay for high-quality single-point tools. This is also experienced The result of the development and education of the SaaS industry; the third is that the ecology of overseas projects is better, there is no squeeze similar to domestic Internet companies, and there are explorations for new products and new business models at the C-end and B-end.” Yunxiu Capital AI The person in charge of the research, Shen Huaxin, told TMTPost.

According to PitchBook statistics, the investment circle will invest a total of 1.37 billion U.S. dollars (equivalent to about 9.369 billion yuan) in generative AI companies in 2022, almost reaching the sum of the past five years.

Image credit: PitchBook

Whether it is text generation, image generation, or video generation, they are not new scenes. The difference is that the original efficient scene is recreated in the form of ChatGPT.

Li Yanhong, the founder and CEO of Baidu, also publicly stated that generative AI will subvert all existing content production models, and it is possible to reduce the cost of original content production to one-tenth of the current one, and make production Processes are sped up hundreds or even thousands of times**.

It can be seen that the **generative AI is not a trick. Its ultimate proposition is to solve the problem of cost reduction and efficiency increase, release repetitive labor, and allow humans to have more time to do more creative things. **

At the level closest to C-end users, generative AI may become a personal assistant for humans. For example, if you want to book a plane ticket, tell ChatGPT the time and place, and it will call Expedia (book plane and hotel) to help you check the plane ticket and make a reservation. If you want to order a meal, it will first call OpenTable to help you select restaurants and menus. There are already 11 apps in ChatGPT’s plug-in store, including air ticket, hotel, shopping, meal ordering and other apps.

The momentum is overwhelming, as if returning to the "iPhone era" of the mobile Internet. Startups, large and small, develop apps based on the iOS system, forming a new software application pattern.

Today, the new wave of generative AI applications inspired by large models will surely have killer applications.

at last

China and the United States will inevitably take a differentiated route.

"The behavioral differences between Chinese and American companies have resulted in the United States usually leading the way in exploration, while China is catching up quickly, and relying more on iterative development at the product level." Xu Siqing said.

Shen Huaxin also agreed, "There will be certain commonality, but the differences will be greater than the commonality. The commonality is mainly in the segmented track, such as chat robots, text/picture/video content generation, etc. But Due to the differences in market demand, cultural background and regulatory environment between China and the United States, entrepreneurial projects may be adjusted and optimized for their respective markets. For example, products in the United States only need to provide a single point of function. It can generate payment for B-end and C-end. Chinese companies are less willing to pay for pure software, and often need to provide a complete set of solutions or integrate software into hardware to generate payment. Therefore, it is possible to start a business in China and the United States The underlying capabilities of the projects are similar to a certain extent, but the business models based on capabilities will be quite different**."

"Most of the domestic VC's actual sales volume is far lower than their enthusiasm for researching ChatGPT. For the domestic market, they mostly research + wait and see. For foreign markets, there may be an atmosphere of investing and trying. You have been to Silicon Valley. I know, it was the same AI entrepreneurial boom as in Zhongguancun back then." A fund partner said.

"**This is a big opportunity no less than the Internet, and it is also a cognitive competition. There is a great opportunity for foresight and foresight. At present, the investment boom of China's AIGC is not as violent as that of Silicon Valley. , but we believe that as the application scenarios continue to be proven, China’s investment will also heat up, and will definitely reach a new high in the future.” Xu Siqing told Titanium Media Venture Capitalists. (This article is the first release of Titanium Media App, author | Guo Hongyan