Recently, there has been a lot of discussion on whether AI works or not, and there are even articles cue to me.

In fact, I had a meal with Zhu Xiaohu more than two months ago, and he said that half of their money would be invested in AI this year, and in fact they have invested in several very good projects in this field.

And that screenshot spread, many people are standing on the periphery of the issue, I look at it from a front-line perspective, in fact, two people said each have a point, then I will provide some of our own facts and opinions::

(1) The FA business of 42Chapter Scriptures has received more than ten AI projects so far this year, five or six of which have been closed or delivered, and three of which are still running at hand. As far as I know, this number is the largest in the industry (one of them?). But I do feel that in the last one or two months, there are a lot of projects in the pipeline. But I do feel that the market is getting colder in the last month or two. I understand that the fundamental reason is that the quality of big models is decreasing in the last month or two, so there are fewer new entrepreneurs and new stories to tell than at the beginning of the year.

(2) At present, almost all the dollar funds in the market are looking at AI, and some RMB funds are interested in AI, but there are only about 50 or 60 of them that we have contact with, including about ten pure RMB institutions. There are also a lot of people who are very active in AI, but not many of them will take the plunge in the end.

(3) in my sense (we and really in the hands of the fund daily to maintain high-frequency communication, so the sense should also be able to reference), since this year to get the money of AI projects in more than 100, of which the mainstream fund investment estimates dozens, there are a variety of underwater, or took a small angel, as well as the concept of rubbish together I will subjectively shoot a 100 - 200 this number.

(4) these projects from the stage, the vast majority are very early projects, most of the products are not online, basically no public financing reports, so may give many people a sense of this market is very poor, no one is investing, but in fact, there are still. In terms of direction, my sense is that 10%-20% of the projects that got money are doing the underlying model, 20%-30% are doing the infra/intermediate layer, and 60%-70% are doing the application layer. (If you add in the ones that haven't gotten money yet, the ones doing applications are estimated to be at least 95%+)

1. In terms of more specific tracks, in fact, most of the companies are talking about things that can be exhausted, no more than the bottom of the big model, multimodal big model, AI + various 2B SaaS (legal, marketing, customer service, CRM, BI, etc.), AI + personal assistant, AI + games, AI + social, AI + comics, AI + education, AI + travel, sound generation, 3D generation, video generation, Chinese version of Civitai, various middle layers, privatized models, vector databases, arithmetic acceleration, distributed computing, etc. ...... This time how to make differentiation is the problem that all teams have to overcome.

(6) At present, the projects that have demo or online account for about 10% of the total projects that have gotten money, so this thing is really a bit frustrating, but we have also contacted several companies that have several million or even higher revenue through AI landing products, and we have seen and heard some very innovative and promising products and ideas. I think it's a good thing for people who believe in it if some people in the market question it because they don't see good enough products.

1. Every track and hotspot has its ups and downs, which is normal. The next wave of the market AI hotspot I judge about two or three months later, because a lot of the first half of the project to get the money, how also have to do a few months to really get to the product online that step, at that time we can look at whether there are more and better killer app, but also to see who will be the leading brother of the application layer. In short, the next stage is to fight the actual landing data.

(8) At the same time, there are some things in the AI market that disappoint me, such as the more I understand the big model, the more I find the limited ability of the big model itself, which brings a lot of practitioners a great challenge to the ground, so I always think that we still overestimate the ability of the big model, underestimate the difficulty of engineering landing available. (We have been talking about this non-consensus for a few months, and it seems that the public's view is now changing)

1. So, I understand the real differentiation of the AI track at the moment is: execution and product implementation capabilities. On top of that is what many people are talking about: data closure, industry knowledge, underlying algorithms, etc.

(10) From the application of the model, we currently see the most typical landing practices is the combination of large models + open source models, I suspect that most companies will be successful in the future will more or less become the so-called "end-to-end" companies. In other words, we will first accept GPT and other models, then we will continue to accumulate data in operation and train our own models through open source models, so as to constantly adjust the proportion of models used, probably from 100% to GPT, gradually becoming, for example, 50% of generalized scenarios with GPT, and the other 50% of professional scenarios with their own models.

(11) Although the model capability is not satisfactory in the visible short term, most people think it is either too simple or too complex, and the actual things that can be done with large models are limited, but it is not necessary to make something very complex like multimodal capabilities. For example, most people are trying to use AI to do generation, to do something out of nothing, but relatively few people are using AI to do analysis, to do qualitative things, in fact, the latter is also able to immediately, the actual ground. So I believe that to use the existing AI is more of a test of the ability to define the product.

(12) The middle layer is necessary and will exist, and will also have value beyond everyone's imagination. If the future model side is open source, closed source multi-model coexistence, and different scenarios need to constantly switch and update, training, operation and maintenance, and the application side is a huge number of scenarios and options, then the middle layer may become a core entrance, even if it does not reverse the threat to the model layer, it may also affect the success or failure of an ecology.

(13) Most of the best companies in the next few years will come out in these two years, and will not wait until the market is completely clear and mature before there is an opportunity. History tells us that most of the great companies in the Internet era and 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, and we believe that the best product is to redefine the existing technology instead of trying to break the limit of technology constantly.

So we looked at it from the application layer, discovered the value of the middle layer, and then discovered the space for vertical domain models, instead of going in the opposite direction, which gave us a more unique perspective.

1. Another truth I have felt in recent years is that

If you see an opportunity, you may or may not be able to make it, but you can never make it without rushing. This truth has actually prevented many people from achieving great success. If something has formed a consensus, there are not too many real opportunities, and with the current trend of AI and the probability of success, it is always reasonable to rush first for respect, otherwise the meta-universe did not rush, web3 did not rush, AI is still not rushed, it may only be a coffee rush not rushed.

(16) Finally, put another egg, that day at the dinner Zhu boss also said a sentence, let me feel worthy of Zhu boss, he said:

Five years later, all companies are AI companies.

I also agree with this, as if almost all companies today are actually Internet companies. It just depends on which companies will take what path to reach this step.

Recently, there has been a lot of discussion on whether AI works or not, and there are even articles cue to me.

In fact, I had a meal with Zhu Xiaohu more than two months ago, and he said that half of their money would be invested in AI this year, and in fact they have invested in several very good projects in this field.

And that screenshot spread, many people are standing on the periphery of the issue, I look at it from a front-line perspective, in fact, two people said each have a point, then I will provide some of our own facts and opinions::

(1) The FA business of 42Chapter Scriptures has received more than ten AI projects so far this year, five or six of which have been closed or delivered, and three of which are still running at hand. As far as I know, this number is the largest in the industry (one of them?). But I do feel that in the last one or two months, there are a lot of projects in the pipeline. But I do feel that the market is getting colder in the last month or two. I understand that the fundamental reason is that the quality of big models is decreasing in the last month or two, so there are fewer new entrepreneurs and new stories to tell than at the beginning of the year.

(2) At present, almost all the dollar funds in the market are looking at AI, and some RMB funds are interested in AI, but there are only about 50 or 60 of them that we have contact with, including about ten pure RMB institutions. There are also a lot of people who are very active in AI, but not many of them will take the plunge in the end.

(3) in my sense (we and really in the hands of the fund daily to maintain high-frequency communication, so the sense should also be able to reference), since this year to get the money of AI projects in more than 100, of which the mainstream fund investment estimates dozens, there are a variety of underwater, or took a small angel, as well as the concept of rubbish together I will subjectively shoot a 100 - 200 this number.

(4) these projects from the stage, the vast majority are very early projects, most of the products are not online, basically no public financing reports, so may give many people a sense of this market is very poor, no one is investing, but in fact, there are still. In terms of direction, my sense is that 10%-20% of the projects that got money are doing the underlying model, 20%-30% are doing the infra/intermediate layer, and 60%-70% are doing the application layer. (If you add in the ones that haven't gotten money yet, the ones doing applications are estimated to be at least 95%+)

1. In terms of more specific tracks, in fact, most of the companies are talking about things that can be exhausted, no more than the bottom of the big model, multimodal big model, AI + various 2B SaaS (legal, marketing, customer service, CRM, BI, etc.), AI + personal assistant, AI + games, AI + social, AI + comics, AI + education, AI + travel, sound generation, 3D generation, video generation, Chinese version of Civitai, various middle layers, privatized models, vector databases, arithmetic acceleration, distributed computing, etc. ...... This time how to make differentiation is the problem that all teams have to overcome.

(6) At present, the projects that have demo or online account for about 10% of the total projects that have gotten money, so this thing is really a bit frustrating, but we have also contacted several companies that have several million or even higher revenue through AI landing products, and we have seen and heard some very innovative and promising products and ideas. I think it's a good thing for people who believe in it if some people in the market question it because they don't see good enough products.

1. Every track and hotspot has its ups and downs, which is normal. The next wave of the market AI hotspot I judge about two or three months later, because a lot of the first half of the project to get the money, how also have to do a few months to really get to the product online that step, at that time we can look at whether there are more and better killer app, but also to see who will be the leading brother of the application layer. In short, the next stage is to fight the actual landing data.

(8) At the same time, there are some things in the AI market that disappoint me, such as the more I understand the big model, the more I find the limited ability of the big model itself, which brings a lot of practitioners a great challenge to the ground, so I always think that we still overestimate the ability of the big model, underestimate the difficulty of engineering landing available. (We have been talking about this non-consensus for a few months, and it seems that the public's view is now changing)

1. So, I understand the real differentiation of the AI track at the moment is: execution and product implementation capabilities. On top of that is what many people are talking about: data closure, industry knowledge, underlying algorithms, etc.

(10) From the application of the model, we currently see the most typical landing practices is the combination of large models + open source models, I suspect that most companies will be successful in the future will more or less become the so-called "end-to-end" companies. In other words, we will first accept GPT and other models, then we will continue to accumulate data in operation and train our own models through open source models, so as to constantly adjust the proportion of models used, probably from 100% to GPT, gradually becoming, for example, 50% of generalized scenarios with GPT, and the other 50% of professional scenarios with their own models.

(11) Although the model capability is not satisfactory in the visible short term, most people think it is either too simple or too complex, and the actual things that can be done with large models are limited, but it is not necessary to make something very complex like multimodal capabilities. For example, most people are trying to use AI to do generation, to do something out of nothing, but relatively few people are using AI to do analysis, to do qualitative things, in fact, the latter is also able to immediately, the actual ground. So I believe that to use the existing AI is more of a test of the ability to define the product.

(12) The middle layer is necessary and will exist, and will also have value beyond everyone's imagination. If the future model side is open source, closed source multi-model coexistence, and different scenarios need to constantly switch and update, training, operation and maintenance, and the application side is a huge number of scenarios and options, then the middle layer may become a core entrance, even if it does not reverse the threat to the model layer, it may also affect the success or failure of an ecology.

(13) Most of the best companies in the next few years will come out in these two years, and will not wait until the market is completely clear and mature before there is an opportunity. History tells us that most of the great companies in the Internet era and 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, and we believe that the best product is to redefine the existing technology instead of trying to break the limit of technology constantly.

So we looked at it from the application layer, discovered the value of the middle layer, and then discovered the space for vertical domain models, instead of going in the opposite direction, which gave us a more unique perspective.

1. Another truth I have felt in recent years is that

If you see an opportunity, you may or may not be able to make it, but you can never make it without rushing. This truth has actually prevented many people from achieving great success. If something has formed a consensus, there are not too many real opportunities, and with the current trend of AI and the probability of success, it is always reasonable to rush first for respect, otherwise the meta-universe did not rush, web3 did not rush, AI is still not rushed, it may only be a coffee rush not rushed.

(16) Finally, put another egg, that day at the dinner Zhu boss also said a sentence, let me feel worthy of Zhu boss, he said:

Five years later, all companies are AI companies.

I also agree with this, as if almost all companies today are actually Internet companies. It just depends on which companies will take what path to reach this step.