Both Zhu Xiaohu and Fu Sheng are right

Source: Chapter 42

Author: Qu Kai

Original Title: "Zhu Xiaohu and Fu Sheng Are Both Right"

Note: This picture is from Zhu Xiaohu's circle of friends

Recently, there have been a lot of discussions about whether AI will work, and even an article cue came to me.

In fact, I had a meal with the boss Zhu Xiaohu more than two months ago. He said at that time that half of their money will be invested in AI this year. In fact, they have already invested in several very good projects in this field, so this wave of discussion And questioning is actually a bit nonsensical in my opinion.

And after that screenshot was circulated, many people looked at this issue from the outside. From a front-line perspective, in fact, what the two people said made sense, so I will provide some facts and opinions of our own:

1. The FA business of Chapter 42 has received more than ten AI projects so far this year, five or six of which have been closed or are in the process of delivery, and three are currently running. As far as I know, this number should be the largest (one of?) in the industry. But I do feel that the market has cooled down a bit in the past one or two months. I understand that the root cause is that the qualitative changes in large models have decreased in the past one or two months, so there are fewer new entrepreneurs and new stories to tell than at the beginning of the year. .

2. At present, almost all US dollar funds on the market are looking at AI, and some RMB funds are interested in AI, but we have contacted about 50 or 60 funds in total, among which pure RMB institutions are estimated to be About ten. There are still many of them who look very positive, but in fact, there are not many shots in the end.

3. Based on my body sense (we maintain high-frequency communication with the funds that are actually selling, so the body sense should be used as a reference), there are probably more than 100 AI projects that have received money this year, and the mainstream funds are estimated to have invested more than a few Ten, as well as various underwater ones, or take the little angel, and the concept of rubbing together, I will take a subjective number of 100-200.

4. In terms of the stage of these projects, most of them are very early projects, and most of the products have not been launched yet, and basically there is no public financing report, so many people may feel that the market is very poor and no one is investing. feeling, but it still exists. In terms of direction, my feeling is that among the projects that get money, 10% - 20% do the bottom model, 20% - 30% do the infra/intermediate layer, and 60% - 70% do the application layer. (If you include those who haven’t received the money, it’s estimated that the application is at least 95%+)

5. From a more specific track, in fact, most companies talk about things that can be exhausted, nothing more than the bottom-level large model, multi-modal large model, AI + various 2B SaaS (legal, marketing, customer service, CRM, BI, etc.), AI + personal assistant, AI + game, AI + social, AI + manga, AI + education, AI + tourism, sound generation, 3D generation, video generation, Chinese version of Civitai, various intermediate Layers, privatization models, vector databases, computing power acceleration, distributed computing, etc. How to differentiate at this time is a difficult problem for all teams to overcome.

6. At present, projects that have demo or launched accounts for about 10% of the total projects that have received money. This matter is indeed a bit frustrating, but we have also contacted several hundreds of products that have already landed through AI. 10,000 or even higher income companies, and we have also seen and heard some very innovative and exciting products and ideas. So I personally still have the enthusiasm and confidence in this track at the beginning of the year. If someone in the market doubts it because they haven’t seen enough good products, I think it’s a good thing for those who believe in it.

7. Every track and hotspot has ups and downs, which is normal. I judge that the next wave of AI hotspots in the market will be about two or three months later, because a large number of projects that received money in the first half of the year will take a few months to actually go online. By then, we can Let's see if there are more and better killer apps, and we can also see who will be the leader of the application layer. In short, the next stage is to fight for the actual landing data.

8. At the same time, there are some things in the AI market that disappoint me. For example, the more I understand the big model, the more I find that the ability of the big model is limited. This has brought great challenges to many practitioners, so I always think that everyone still overestimates the ability of large models and underestimates the difficulty of engineering implementation. (We have talked about this non-consensus for several months, and it seems that the public's views are also changing)

9. Therefore, the current real differentiation of the AI track as I understand it is: execution and product landing capabilities. On this basis is the data closed loop, industry awareness, underlying algorithms, etc. that many people are talking about.

10. From the perspective of model application, the most typical practice we have seen so far is the combination of large model + open source model. I suspect that most companies will become more or less so-called "end end-to-end" company. That is to say, everyone first connects to GPT and other models, and then continuously accumulates data in operation and trains their own models through open source models, so as to constantly adjust the proportion of model usage, which may gradually change from 100% to GPT, such as 50%. GPT is used for professional scenes, and our own model is used for the other 50% professional scenes.

11. Although the ability of the model is not satisfactory in the short term, most people think it is either too simple or too complicated. The actual things that a large model can do are limited, but it is not necessary to do something Complex capabilities like multimodality. For example, most people are trying to use AI to generate and create things out of nothing, but relatively few people are using AI to do analysis and qualitative things. In fact, the latter can also be achieved immediately and actually implemented. So I believe that to make good use of the existing AI, it is the ability to define products that is more tested.

12. The middle layer must exist, will exist, and will have value beyond everyone's imagination. If open-source and closed-source multiple models coexist on the model side in the future, and different scenarios need to be constantly switched and updated, trained, and maintained, and the application side has a large number of scenarios and choices, then the middle layer may instead become the core entrance , even if it does not threaten the model layer in reverse, it may affect the success or failure of a certain ecology.

13. Most of the best companies in the next few years will come out in the past two years, and they will not wait until the market is completely clear and mature to have a chance. History tells us that most of the great companies in the Internet era and the mobile Internet era were born together with the beginning of the new era.

14. We believe that technology is for product service, and product is for user service. We believe that the best product is to redefine existing technology, rather than constantly trying to break through the limits of technology.

So we start from the application layer, discover the value of the middle layer, and then discover the space of the vertical domain model, instead of going in the opposite direction, which gives us a more unique perspective.

15. Another truth I have felt in recent years is:

If you rush to see an opportunity, you may or may not succeed, but you will never succeed if you don't rush. This principle actually hinders the success of many people. If there is a consensus on one matter, there are not many real opportunities. With the current situation of AI and the probability of success, it is always necessary to respect the situation and reason. If you don’t make the coffee or the AI, maybe you can only make the coffee or not.

16. Finally, put an easter egg. Boss Zhu said something at the dinner that day, which made me feel worthy of being Boss Zhu. He said:

Five years from now, all companies will be AI companies.

I also agree with this, as if almost all companies today are actually Internet companies. It's just to see which companies will use what path to achieve this step.