With 100 million monthly lives in 60 days, OpenAI's ChatGPT was born, making the current innovation of AI more well-known to the public, and then more companies such as Google, Microsoft and domestic Baidu joined the competition sequence.
So, under the well-known AI established opportunity, who is thinking and who is acting? There are investors who make a move, and there are investors who wait and see; There are entrepreneurs who are positive and optimistic, and there are entrepreneurs who are cautious.
Of course, the continuous entry of Wang Huiwen, Zhou Bowen, Tang Jie and other industry insiders has added a handful to the AGI represented by ChatGPT, which has been warmly welcomed by many investors, not to mention the voice recognition, image recognition and other AI subdivision unicorns have long been in front of the front, so that regardless of their own attitude, it has become a very correct thing to shout for the advancement of AI innovation.
But the problem is that not every wave of technology dividends can be really seized by the previous generation of large companies, nor necessarily startups can necessarily seize the opportunity of innovation, so the question is, artificial intelligence is rapidly becoming a general technology to empower all industries, into the era of general large models, who can really seize this opportunity, take the opportunity to independently lead?
In this context, on February 28, Tozhong Information made a closed-door salon on the "AI New Singularity", in this AI closed-door salon of the Touzhong 21/2 series of salons supported by Lightspeed China, more than a dozen guests provided a lot of industry worthy of thinking and investment directions for the current situation of AI to promote development.
To vote or not to vote is to support it unequivocally
At the salon, both RMB funds and US dollar funds were observing AI-related opportunities and choosing the opportunity to shoot.
The only difference is that, in the words of Wang Bing, a partner at Oriental Fuhai, RMB funds are more inclined to invest in the underlying infrastructure, have a clear business model and core barriers, and the rate of return is very clear, which is worth betting on.
But dollar funds are more daring to take risks and heavy positions, Red Dot China partner Liu Lan said that for dollar investors, the bet of large models is also inevitable, hoping to catch up or even catch up with more investment, especially entrepreneurs like Wang Huiwen and Zhou Bowen, with very high financing capabilities, and very strong team capabilities, it is worth a try.
However, everyone also sighed that at present, many companies are rushing in the direction of GPT, Baidu is already the most determined company in China, and it is necessary to wait for Wen Xin's products to come out to see how the development is, and many startups need to find opportunities in the "vertical" direction, so as to find their own advantages in the differences and form a real ecology in China.
In the words of Zhang Qing, managing director of CICC Capital, whether AIGC or AGI at this stage is more similar to the meta-universe that was particularly lively but rarely landed in the previous two or three years, or similar to the frequent emergence of Internet and mobile Internet giants, opening up a broad track, and it has just begun.
Even if it is advanced to the current AIGC level, as for practitioners or investors, it is not a new story, how can the valuation be reasonable? How regulated is it? How does the business land? All are subject to further verification.
After all, in the process of AI development for so many years, the imagination space such as voice and vision is relatively limited, and the current chatGPT has made many people feel beyond expectations, and then if the space of this "miracle" is no less than the space depicted by the meta-universe, and even the peak industry and industry, it is necessary to truly make a miracle.
In the end, AI will return to the core elements, computing power, algorithms, data and sensing, these four are coordinated and spiraling, only a specific time to build the highest barrier to the four elements, why do you think OpenAI is unmatched? It is precisely because the above equilibrium has developed to a certain stage. A bright future is unfolding for investors and entrepreneurs, but it is a future where the bar is significantly raised.
Zheng Can, Managing Director of Linear Capital, also elaborated that Linear has paid attention to and has a number of investment cases in AIGC scenarios, and many AIGC application scenarios mentioned today have existed before, which is nothing more than today's large model to let everyone see the ability to accelerate generalization, followed by huge infrastructure workload and new imagination space and opportunities.
In this way, "money, people, cards, data" several key factors, "money" and "people" are relatively easy to do, money always has someone out, everyone knows where to dig, nothing more than the problem of digging or not digging and how much money, "card" is not so worried, there are always two or three years to do things, but more worried about the data problem, firstly, the data cleaning work is very demanding, and secondly, the commercial return is relatively small, after all, only good Chinese data sets, and the future connection with other models, can continue other innovation and development.
Liu Lan, partner of Red Dot China, took his investment in two computing power chip companies as an example and found that China has a huge amount of data, so start-ups including AI security and other subdivisions can grow, but the core of the current wave has become an algorithm model, but because this matter is very meaningful in China, even if the odds of the large model are the highest, you must participate.
In particular, only China has its own algorithm model, can better pull up talents in large factories or entrepreneurship and even in universities, so as to form a mature business model, at present, many practitioners have poured in, has begun to form a hundred flowers blooming situation, so it is completely possible to continue to bet according to the logic of investing in the model first and then investing in the application.
Cai Wei, partner of Lightspeed China, also said that in the last wave of AI entrepreneurship, many investors have bet and gained, but they also encountered the innovation wall of AI in the middle, until the application of large models in 2021 is gradually widespread, and in 2022, excellent applications such as Jasper appeared in generative models, coupled with Lightspeed invested in the United States Stability.AI, and now see the ChatGPT model continue to rise, and the innovation wall of AI has been broken. Make yourself more determined to continue investing in this area.
After all, when NLP has a surge effect, will the image also have a surge effect? Will the future big model directly become a vertical big model, or will it become a multimodal big model that will unify the world? Once it changes, it will change many AI development trends, resulting in more new landing business models, and can also empower more different industries, which is where many opportunities lie.
Liang Junzhang, founding partner of Kunzhong Capital, has been discussing with many AI companies since last year how to take advantage of the new wave of opportunities to upgrade and expand their businesses, so he pays more attention to whether more companies can make targeted industry solutions for industry applications, in addition to screening targets according to their own investment habits on many dazzling applications.
In addition, according to the current different situations in China and the United States, in the Chinese environment, To B companies usually cannot rely on products alone to solve the pain points of the industry, but still have to provide a complete set of solutions and services, so that the company that is already doing can promote the development of the company in the original industry knowhow, customer accumulation, channels and other advantages.
On the contrary, there may be new companies running out in the To C field, whether it is the game or entertainment field, the metaverse level deserves more attention, just as China can always run out some game companies, it is worth looking forward to.
Shen Libin, CEO of Leyan Technology, also bluntly said that there are many startups based on OpenAI in the United States, but it is still unknown whether China will have this wave of entrepreneurial opportunities. What can be seen more is that large model companies currently have two opportunities, one is to do open source, the other is to do vertical, after all, the general model can not be directly used in the vertical field, there is still a certain time window.
SaaS software companies that make good use of large model technology, especially in the field of heavy interaction, will form a capability advantage in homogeneous competition. Technology has a window period, and finding usage in vertical fields in a way similar to a large model may lead by half a year to a year, and the possibility of no one catching up in two years is too low, so more opportunities seem to be left to companies that have reached a certain size in the industry, rather than companies that have just entered the industry.
If you dig deeper, can you look at the so-called "devil in the details", taking the customer service field as an example, whether each round of different words can be formed according to the different operation strategies of each store in the process of interaction, so that technological innovation and industry knowhow are closely combined, but it is easier to form their own barriers.
To do or not to do, it is necessary to be more discerning between real and false opportunities
At the salon, many guests hotly discussed why OpenAI was not invested by market-oriented VCs. OpenAI was originally a non-profit organization, the goal is to study the subsequent impact of AI on human beings, there is no commercial KPI, any commercial VC must be invested in the target with a clear commercial landing goal.
Therefore, when OpenAI made GPT-3, Microsoft saw the potential of the former, and Microsoft wanted to take the opportunity to do its own cloud services well, so it finally invested in OpenAI. Even so, Microsoft's investment is business-oriented, that is, 80% of the money goes back to Microsoft, becoming Microsoft's performance, and the stock price takes the opportunity to rise a lot.
So, how can more startups take advantage of this wave of opportunities to develop and grow themselves?
Wang Bing, partner of Oriental Fuhai, said that throughout the history of OpenAI, it has probably begun to train models with unlimited fees as early as 4 years ago, so under the premise that the United States is not open source, it will take a long time for many domestic companies to explore it.
Among them, there may be three barriers: on the one hand, there are barriers at the hardware level, the head company may be fine, but most startups are difficult to solve this problem; On the one hand, there are human barriers, in addition to pure academic barriers, even experienced talents need to be accumulated; In addition, it is the accumulation of data accumulation, 95% of the academic literature is in English, and the accuracy of these data is very important.
Why didn't the "AI Four Tigers" make money until the back? Wang Bing analyzed that in the final analysis, it is not because of dedicated capabilities, but because of low barriers, so whether it is general artificial intelligence or special artificial intelligence, high technical and data barriers are the key to making money.
Li Kejia, partner of Chuxin Capital, as a technology entrepreneur and investor, shared his views, from the perspective of users and data, OpenAI is still unique, strengthening the closed-source large model, with the significant reduction of costs for developers, the "ecological (parasitic) entrepreneurial opportunity" based on OpenAI is still the most certain; OpenAI is also reinforcing this assumption through capital, launching the accelerator Converge, which, in addition to funding, offers special incentives, including licensing discounts and early access to new technologies such as GPT-4.
Open source big models, such as Meta's LLaMA are still worth paying attention to and investing in, as a latecomer, superimposed on some security and compliance issues, open source and closed source will promote the development of the entire industry between you catch up, we can find a familiar path reference in the development of chip architecture and operating system in the past three decades.
For entrepreneurial opportunities outside of self-developed large models, some assumptions can even be made, if the main differentiation of the final product is AI itself, then verticalization + middle layer (training and hosting of large models to Developer's large models and domain models) is likely to win. But at the application layer, the demand for AI will have long-tail features, so it is more likely to be leveled. That is, over time, we should also see more traditional moats established, returning to the essence of business and efficiency, including the original intention capital may also see new moats gain a foothold.
Xie Yujuan, partner of Chuangfang Investment, highlighted that in the future, ChatGPT applications and product ecology need to focus on industry supervision. Especially in some specific industries, which pay more attention to independent property rights, then how the hardware ecology is combined with software will determine the choice of product path and technical path at the practical level. In which industry can ChatGPT applications run out of unicorn companies, it is necessary to analyze which links the core resources in the industry master, and whether ChatGPT can help achieve breakthroughs in more industries.
Specific to the application layer, how to form a cost-effective solution based on the application scenario, in the end, it is necessary to start from the customer's perspective to see how the product or solution can form a closed loop at the specific business level while providing customers with a higher cost-effective advantage. In general, from the perspective of the closed loop of the business model, the people who ultimately pay in the industry chain, as well as the regulatory policies at home and abroad, will have a significant impact on the choice of the underlying algorithm model and hardware of the startup.
Tao Fangbo, CEO of Mind Universe, also said that he did not expect ChatGPT to come so soon, even if he paid attention to AGI very early on because of his experience, he only estimated as an optimist that it could come out in about 5 to 19 years, but now, a year has appeared, so there may be good and bad for the industry.
After all, from the original perspective, a lot of AI work may have no meaning before, but need to be reconstructed, for example, one day, the software will be refactored all again, so GPT should be regarded as a new "brain resource pool", specifically ChatGPT is a brain resource, is a new era of CPU.
Today's AIGC and AGI are not the same, AGI may subvert the entire industry, Tao Fangbo even said that it is a greater opportunity than the mobile Internet, even if China does not have it now, and it hopes to have, even if the success does not have to be in me. How to do it specifically, it is necessary to do a good job of the large model layer first, so that Chinese entrepreneurs can look up, and better lay the middle layer on it, and put the large model into the application in a targeted manner, so as to produce better applications through scheduling.
Zhang Lei, founder and CEO of Stardust Data, pointed out an overlooked question, that is, is there an AI infra company in China? The answer is that only foreign companies have, domestic practitioners are too focused on the methodology, and the methodology is public, but in fact, the content that is not disclosed has more knowhow and barriers. In addition, it is the accumulation of data accumulation, 95% of the data set is generated by the English-speaking world, and there is currently a relative lack of data sets in the Chinese environment.
Therefore, Zhang Lei thinks from his own point of view, what can he do from the perspective of AI? Combined with itself, but feel that the direction of data is ignored by everyone, and the importance is high enough, a certain algorithm background is also required: 90% of the data can be automated, and the people who want to talk to the algorithm also need a good data strategy, plus the model is constantly iterating, and the data also needs to be constantly iterated, which is worth paying attention to and doing.
Deng Yulong, founder of Xiguang Technology, also speculated that the social aspects represented by chat tools, as well as the video CV aspects, may be greatly subverted. Why? In addition to a large number of QQ user system escort, there are also important functional points such as mobile chat and voice, so it is an inevitable trend for social networking to be subverted, only regardless of the length of time.
For startups, whether it is OpenAI or Baidu, it does not matter who does the underlying framework, the platform is good enough to use, but it is more aimed at its own play space, do its own part, and when it achieves a certain scale, it can also borrow a large model to make its own industry model.
Wang Zhiwu, CEO of Yuanjing Technology, a subsidiary of Tianyu Digital, said that he has been insisting on the vertical track of virtual digital humans, and is also a relatively early company in the industry to access ChatGPT capabilities, at that time, he just wanted to use NLP small models to make virtual humans move and do applications in tourism and cultural tourism.
But in the application process, it was found that the delay of the virtual pseudo-digital person will reach 6-8 seconds, and the context will not be right during the live broadcast, a serious nonsense", so the final is to use the ChatGPT model to train NLP, and use the large model to train the small model, so that you can continue to cultivate in the vertical track such as virtual customer service, live broadcast virtual anchor field.
Yu Wei, founder of Ruiqi Technology, also said frankly that it was very difficult to see AI empowering enterprises when he was at Microsoft, one of the important reasons for this is that the technology is not mature enough, such as voice recognition and NLP provided to customers are not very accurate, but on the contrary, the business is very complicated, but after doing it yourself, you can find it, go deep into the scene, polish the product, and iterate the technology to really land.
Therefore, Yu Wei focuses on, how good is ChatGPT today? To this end, Ruiqi Technology has recently done a lot of tests and found that ChatGPT as a general model can be used to do a lot of work, but there is still a big gap with professional industry models, only by going deep into the industry and truly integrating AI capabilities and business can we truly realize the landing of AI technology and solve key efficiency and productivity problems in the To B process.
Perhaps, to say a thousand ways and ten thousand, it still requires many practitioners in China to collect firewood and work together to truly seize the opportunity of AGI this time and take the opportunity to present more flowers of ecology.
The good news is that in the past two days, in addition to Wang Huiwen and Zhou Bowen jumping to choose to start a business again, Tang Jie, Wang Xiaochuan and many other bigwigs are also on the way, Baidu's Wen Xin was directly released in March, and China's AI is also just like the weather in early spring, and it has new vitality. (Text/Zhang Lijuan, source/Touzhong.com)
![They fooled the richest people[fig]](data:image/gif;base64,R0lGODlhAQABAIAAAP///wAAACwAAAAAAQABAAACAkQBADs=)