Thank you very much for your support to the Global Product Manager Conference. It is your support that has made the Product Manager Conference last from 2009 to today. The topic of my speech today is "Product Layout and Paradigm in the AGI Era". My speech today is mainly divided into the following three parts:
In-depth understanding of the characteristics of the AGI era 2. The "paradigm transformation cube" of scientific and technological innovation 3. Six thoughts on the future development of AGI products
Part 1: In-depth understanding of the characteristics of the AGI era
First of all, let's get acquainted with the entire development of large-scale model technology through the following picture.
You can see that the earliest artificial intelligence was developed from machine learning. After the deep neural network on ImageNet shined in 2012, deep learning has become a prominent science of artificial intelligence. Later, RNN and LSTM were developed in the field of natural language processing. After seven Google research scientists published Transformer papers in 2017, the pre-trained language model gradually became the mainstream, and then the large language model (LLM) represented by GPT because ChatGPT The success of AI is considered by the industry to be the door to general artificial intelligence (AGI).
Due to my tracking and research on artificial intelligence-related technologies, I happened to have in-depth exchanges and discussions with several key figures on the above-mentioned development line. In 2018, we held the Global Machine Learning Conference in Shanghai, and invited Michael Jordan, the father of machine learning and a famous professor at UC Berkeley, as a keynote speaker, and had in-depth exchanges with him. In April 2021, when we held a machine learning conference in Beijing, due to the epidemic, we invited Lukasz Kaiser, one of the co-founders of the Google Transformer model, online. He told me shortly after that speech that he left Google to go to OpenAI, and later found out that he was investing in GPT 3.5. At that time, another deep learning veteran, Jurgen Schmidhuber, the father of LSTM, was invited. By April this year, I went to Silicon Valley and had a lot of in-depth exchanges with Ilya Sutskever, the chief scientist of OpenAI. In general, on the road of AI development, we have been maintaining in-depth discussions and exchanges with the frontiers of the industry, which have benefited me a lot.
First, let’s talk about the AGI technology stack, which is generally divided into three layers: the application layer, the model layer, and the infrastructure layer (of course, some people in the industry extract LLMOps and make a separate layer, called four layers). No matter the third floor or the fourth floor, looking at this picture, it is easier to understand, so I won't explain much. It is a basis for our understanding of large model technology.
But this understanding is easy to stay at the technical surface. Let's look at the deeper things behind the large-scale model technology. I think it can help us understand this wave of technological revolution led by the large-scale model. If we look at the history a little longer, there are two very important propositions in the entire technology industry, one is connection and the other is computing. They're like a pendulum that swings from connection to computing for a while, and computing for a while.
Let's take a look at the picture above. During the 100 years from 1840 to 1940, the entire technological revolution was dominated by "connection", including the telegraph, telephone, radio, and television. Connecting people and organizations from all corners of the earth has profoundly changed the social form at that time.
Then, from the birth of the first computer ENIAC in 1946, the technological revolution of mankind entered a pendulum of "computing". Including mainframes in the 1950s, minicomputers in the 1960s, minicomputers in the 1970s, and PCs in the 1980s. With the von Neumann architecture as the core, it is all carried out around the proposition of "computing". This process has been going on for about 50 years.
Next came the emergence of the Internet from 1994 to 1995, marked by Netscape and Yahoo. The WWW Internet pushed the human technology revolution to the "connection" pendulum, followed by Web 2.0 in 2004, and marked by the birth of the iPhone in 2007. Mobile Internet, Cloud Services in 2013. They are all greatly developing and enriching the proposition of "connection". This process lasted about 30 years.
Time came to 2017, the paper on the Transformer model was published, GPT 1.0 was born in 2018, and then 2.0, 3.0, and ChatGPT brought by GPT 3.5 at the end of last year came out, bringing the dawn of general artificial intelligence AGI to the entire human being. This round of large models has entered the pendulum of "calculation", and the time will start from about 2020.
It's very interesting, if you look at the first "connection" revolution, I named it "connection 1.0", it went through about 100 years; and the second "computing" revolution, I named it "computing 2.0" ", it went through about 50 years; then came the second "connection" revolution, so-called "connection 2.0", it went through 30 years. Do you see an exponentially decreasing trend in the time span? Therefore, many people believe that we may not be able to do so in the next 15 years, and by 2035, the "computing" pendulum brought about by this wave of large models may reach its peak.
With the understanding of the timeline of the industry, let's take a look at the difference between the underlying logic of the "connection" era and the "computing" era? This is actually very important. Because many people often use various paradigms of the mobile Internet during this period to deduce the paradigm of the big model era. I think this perception is wrong. Because the mobile Internet belongs to the logic of "connection", while the era of large models belongs to the logic of "computation". Let's take a look at the table below:
From the perspective of production transformation, connection logic dominates "production relations", while computational logic dominates "productivity". From the perspective of production-consumption relationship, we know that in the Internet age, under the connection logic, there is a very important effect called "two-sided market", including buyers and sellers on Taobao, passengers and drivers on Didi, and Douyin. listeners and presenters. One is a producer and the other is a consumer. It constitutes a very strong bilateral effect and is also an important moat for many products in the Internet era.
Under the calculation logic, there is actually no such a bilateral effect, and its core is a unilateral market. One end is algorithm computing power, and the other end is consumers. For example, in Midjourney, there is no network of designers, and the algorithm directly produces pictures to users; for example, in the era of autonomous driving, if Didi is to be subverted, the moat formed by Didi’s driver-passenger two-sided market will be useless at all, because there is no need for drivers. It's the algorithm that's driving the car.
Looking at the business model again, under the connection logic, its marginal cost is very low, so it is easy to do a free business model. However, under computational logic, the cost of algorithm computing power needs to be shared equally. So will most business models still be like "free is king" in the era of the Internet and mobile Internet? Most likely not, which is why everyone sees that Midjourney and ChatGPT Plus are charging. Many Silicon Valley investors have suggested that "directly charging users" will be the main business model in the era of large models. It is no longer a free model of "the wool comes from the pig" in the connection era.
And what about user experience? In the Internet age, the connection logic is that the more information the better, an inevitable problem of information overload arises. But the calculation logic, in fact, is that the fewer contacts the better, and efficiency comes first. That is, I will just give me my results in the next order, and don't let me participate in the intermediate process.
In terms of decision-making mechanism. The connection logic is "the machine will give me information, and I will make decisions". And computational logic is that people give information to machines, that is, data, and machines help me make decisions.
Through the above comparisons, you can see that these two logics are very different in our product thinking. Next, let’s talk about computing logic. What is the essential difference between computing 1.0 from 1940 to 1990 and computing 2.0 we are currently in? Look at the picture below:
In the computing 1.0 era, the left side is the human brain, which we call the biological neural network, and the right side is the digital logic circuit. All our traditional computers are essentially "and, or, and not" of digital logic circuits. It is a structured operation logic that can only accept structured data, which is essentially a 0-1 thinking. The result it leads to is to make people adapt to the logic of the machine.
So you will find that under Computing 1.0, whether it is PC or mobile software, there are many places where the user experience is a little careless, and it is easy for non-industry people, including the elderly and children, to fall into a predicament of mental burden. Why this dilemma?
Because the left side is the biological neural network of the human brain, and the right side is a digital logic circuit composed of "and, or, and not", you have to adapt it. Any input input must first be converted into something that a digital logic circuit can understand; any output from a computer must undergo a layer of conversion into something that a biological neural network can understand. This back-and-forth conversion is where the untrained average user gets into trouble.
But if we look at the computing 2.0 era dominated by the large model described in the figure below.
I drew the picture on the right as a brain, although inside it is a silicon-based chip, the so-called carbon-based intelligence and silicon-based intelligence.
When I communicated with OpenAI’s chief scientist Ilya Sutskever in Silicon Valley in April, I remember he repeated three times to emphasize that from the perspective of mathematical principles, the current digital neural network is no different from the carbon-based neural network of the human brain. Except for the different survival methods of biological organisms, the logic of operation is the same.
At this time, the interaction between humans and the large model is actually the interaction between the biological neural network and the digital neural network.
It is no longer the 0-1 thinking of digital circuits in the computing 1.0 era, but a probabilistic thinking. Everyone knows that you can adjust the temperature (temperature value) parameter of ChatGPT. Turn up the temperature to get creative. It is also very interesting to use the word temperature. A high temperature means that the brain is hot, and the brain is more creative when the brain is hot. You can also turn down the temperature to make the answer very mediocre. The essence of probabilistic thinking is the model of neural network, which is also the decision-making method of our human brain. Think about it carefully, everyone. In fact, every thought in our brain is calculating probability. The result of the entire orientation of Computing 2.0 is to allow machines to adapt to human thinking.
With the above in-depth understanding of the entire AGI technology and the logic behind it, let's talk about some methodologies on how to make products in the AGI era.
Part II: "Paradigm Shift Cube" of Technological Innovation
Let’s talk about paradigm shift first. It is called Paradigm Shift in English. It was first proposed by Thomas Kuhn in the book "The Structure of Scientific Revolutions". It refers to the fundamental changes in the basic concepts and practices in the field of science and technology. It breaks the original The laws and boundaries of human beings create a new world for people's thoughts and actions. The revolutions we often talk about in the field of technology are actually paradigm shifts. For example, from stand-alone to the Internet, from the Internet to the mobile Internet, are all paradigm shifts.
In the Chinese Internet field, many friends know that Wang Xing of Meituan once had a famous four-vertical and three-horizontal theory around 2009. At that time, many people on the Internet were very excited when they heard this theory. Because of Wang Xing's analysis, it seemed to be more in line with the current situation at that time. 2009 was when the mobile Internet just broke out. It is very similar to the current AGI outbreak.
Wang Xing pointed out at the time that human beings are basically dominated by four major needs in the technology industry: information, communication, entertainment, and business. The so-called four verticals. As for Sanheng, Wang Xing gave his own analysis: search, social, mobile. Then at the intersecting point of four vertical and three horizontal, a new opportunity was born. I agree with the "four verticals" very much, but for the three horizontals, I think this analysis method is problematic, because social networking and search are actually human needs, and they are not on the same dimension as the mobile Internet.
I guess Wang Xing put social networking on the same dimension as mobile at that time because the social network was so popular at the time that social networking became a huge traffic portal. Once social networking is encountered in many fields, there will be new ways to play. But traffic entry is not equal to technology. The same is true for search, because it also has a traffic entrance effect, so Wang Xing feels that it has also caused changes in many fields. However, Wang Xing did not give a convincing product model in the fields of communication, entertainment, and business other than the Google/Baidu model for search. So I think the "three horizontals" of Wang Xing's "four verticals and three horizontals" methodology are misplaced.
Next, let me talk about an analysis framework I proposed. I think this analysis framework is very helpful for us to analyze the product paradigm in the AGI era. I named the analysis framework I proposed "Paradigm Transformation Cube", and the English name is ParaShift Cube, where Para is the abbreviation of Paradigm.
Why is it called a cube, because we define three axes: x-axis, y-axis, z-axis.
The x-axis represents the technology axis, which represents the transformation of technological paradigms. There are connection 1.0, computing 1.0, connection 2.0 (including PC Internet, mobile Internet), computing 2.0, which is what we call the AGI intelligent era.
The Y axis represents the demand axis, covering all aspects of human needs, including the part Wang Xing talked about: information, entertainment, and business. At the same time, I also put social networking and search on the Y axis. Social and Communication, I combined them, and while they have slight differences, their commonalities outweigh their differences.
The Z axis represents the medium axis, because the expression of information requires a specific medium, and human beings also interact with the world around them through the medium. The media axes include: text, pictures, audio, video, and three-dimensional space calculation (this is Apple's latest proposal in VisionPro). I think the Z axis is a point that many people ignore. Including the GUI graphical user interface, which has a great influence on the development of computers, is also the result of interaction driven by events in a medium such as computer images.
These three axes form a cube structure. The three axes, the point of change on each axis, and the intersection of other axes are the places where "paradigm transformation innovation" occurs in products. Applying the Paradigm Shift Cube to analyze innovation opportunities in product areas is logically straightforward.
Next, we use the "Paradigm Transformation Cube" to analyze the connection to the PC Internet in the 1.0 era, connect to the mobile Internet in the 2.0 era, and predict the possible opportunities in the era of computing 2.0 AGI intelligence. As for the computing 1.0 and connection 1.0 era, in fact, you can also use the "paradigm transformation cube" analysis, but the age is relatively old, so I won't talk about it here, but if you are interested, you can go back to history for analysis, and you will find that it is also very logical .
Because the content filled in the cube is not very easy to express in PPT, so I used a table to express the technology axis and demand axis, and there is a blue axis below it, which represents media changes.
I split the demand axis in the table into two parts: red and green. The red part, in the era of AGI intelligence, is the part where the paradigm shift is more drastic, and there is a great chance of innovation and even overthrowing the giants; first look at the information demand, in the transformation from PC to mobile, portal models such as Sina and Sohu are being replaced by Toutiao, etc. Defeat, and the production method of information in the age of AGI intelligence will face huge changes. Now the information overload is very serious. Sometimes, after reading ten articles, you actually get only one piece of information. So can the big model help me integrate dozens of pieces of information from the previous day according to my preferences, and then give me some short and concise summaries, which is what the big model is very good at. This can easily lead to a paradigm shift in information products. Let’s talk about entertainment again. Entertainment needs content very much, and AIGC is very good at content generation. The form, carrier, supply and other chains of entertainment will be reshaped by AIGC, so there are also many opportunities for paradigm shift.
Finally, let’s talk about search. Search is a big industry in the Internet field, and it is also a big demand for human beings. But if you use ChatGPT and other large-scale model products, you will find that the proportion of search behavior will drop significantly. Why? Because many times, we search a dozen articles and turn several pages just to find an answer or a decision. Is this product any good? Is that attraction worth visiting? Does this medicine solve the problem? These decision-making and answering needs are what large-scale model products such as ChatGPT are very good at. Therefore, search will face a huge possibility of being subverted in the AGI era. Unfortunately, when I returned from Silicon Valley in April and May, I felt that Google was still counting money on the search credit book, and was numb to the threat of the huge paradigm shift brought about by ChatGPT.
Some people say, hasn't Google launched a lot of technologies and products related to large models? My answer is, don't look at the few products launched on the surface. These defensive actions are all drizzle when the paradigm shift comes. Think about how Yahoo collapsed that year? Of course, does Google have a chance? I think there is, but the premise is that Google must come up with determination similar to the war that Ma Huateng and Zhang Xiaolong used the power of Tencent's entire company to All in WeChat to fight Michat; or Bill Gates in 1995 The determination to fight the browser war with Netscape by using the strength of Microsoft's entire company to All in IE every year can prevent the subversive force of paradigm shift. And what did Google do? The founder lay on the beach in California and basked in the sun, and let the CEO who was hired to get the annual salary and OpenAI's Sam Altman and Ilya Sutskever and other co-founders with poetry and distance to fight, and then the technology master Jeff Dean was removed from the position of the boss of Goolge Brain, and Jeffrey Hinton, the father of deep learning, was tactfully dismissed. Do you want to win the AGI era by relying on a few large PPT models at the Google I/O conference? By the way, I would like to criticize the impetuous style of Google's Daxing PPT products in recent years (many of which have not been officially released after that), and let's see when startups such as OpenAI need a development conference? A product release is a press conference, with at most a blog explaining it. Talking about products (rather than PPT) is the king of innovation!
Finally, let me say that it is not another portal that overturns Yahoo; it is not another search that overturns Google, but the structural subversion brought about by the paradigm shift, and this is the chess game that the search field is facing in the era of large models.
After talking about the red part, let's talk about the green part. Compared with the red part, the green part still has the "two-sided market" effect of the connection era, so the moat is relatively deep. Even when the technology undergoes a paradigm shift, I think it is not There are great opportunities. For example, in the transformation from PC Internet to mobile, even though Michat, Momo and other companies attacked social networking at that time, Tencent survived (although it changed its name from QQ to WeChat), because the social networking established by Tencent in the QQ era The moat is too deep. The entire social relationship in China is not a simple bilateral structure, but a network structure, which is in the hands of Tencent.
Many people also asked some time ago that in the AGI era, will WeChat be subverted? We use the above paradigm transformation cube analysis, the conclusion is not. First of all, in the AGI era, will people’s social needs change? Won't. Secondly, will the social model between people and people become the social model between people and virtual people? I don't think so either. There is a saying in Silicon Valley that robots in the future will be slaves to humans, and people use robots for a clear purpose of "making it work." The need for social interaction comes from the structure of human social relationships: friends, relatives, classmates, colleagues... unless there are no such social relationships in the human social structure in the future. Nobody wants to socialize with a robot. Therefore, the advantage of WeChat in China, and more precisely, it should be the social network of Chinese people all over the world, will not be subverted even in the era of AGI's major paradigm shift. As for WeChat, there are some specific interactive updates, such as text and image generation, and smarter information optimization. These are small interactive improvements. These things can be done in minutes with WeChat’s ability, and it’s not up to any start-up company to do it. Subvert it.
I remember that on the same day in January 2019, Byte launched Duoshan, Luo Yonghao launched Chat Bao, and Kuaibo founder Wang Xin launched Toilet MT. They wanted to cooperate to overthrow WeChat’s status as the social king, but they died within two months. . If the founders had read my "Paradigm Shift Cube" at that time, they would have understood that this matter cannot be done. What's more, 2019 is not an era of technological paradigm shift. Even in the era of PC Internet to mobile paradigm shift, Lei Jun's Michat, international WhatsApp, Talkbox and other applications have not subverted Tencent, because the social moat is the deepest. In the AGI era, everyone should not think about using WeChat, there is no door.
Let’s talk about business, which actually refers to e-commerce. There are not many opportunities for major disruption, and the reason is that the moat of the two-sided market in the "connection era" still exists in the e-commerce field. Any business form is inseparable from buyers and sellers. The AGI large model can only optimize some small links in the closed business loop (such as Taobao pictures, virtual fitting rooms, etc.), but these can never be separated from buyers, sellers, logistics, etc. Basic e-commerce attributes, and the advantages of traditional giants in these fields are still great. That is, in the transformation from the PC Internet to the mobile Internet, the advantages of Taobao and JD.com are basically translated.
However, one thing needs to be pointed out. In the field of mobile Internet, Pinduoduo and Douyin e-commerce have emerged halfway, but this is brought about by the traffic entrance effect of WeChat and Douyin. It is a good supplement, but it does not constitute a subversion. Moreover, Pinduoduo hits the sinking crowd, which is brought about by the new Internet demographic changes (this has the opportunity to re-analyze). My judgment is that as large-scale models may bring new traffic entrances, new opportunities similar to Pinduoduo and Douyin e-commerce may emerge in the e-commerce field, but the existing advantages of Taobao and JD.com will still exist. After all, the plate of e-commerce is too big. The Pacific Ocean can accommodate China and the United States, and the e-commerce field can accommodate many small giants.
The blue axis at the bottom is mainly the media we mentioned earlier: text, pictures, audio, video, and three-dimensional (spatial computing). It is easy for many people to overlook this part. The power of products behind different media is very different. When we make any product, we must think about what kind of media carrier we want to focus on.
In terms of media logic, pictures are better than text, video is better than pictures, and audio has its own special scene. It is estimated that many of the note-taking and text-based products in the early mobile field have disappeared. A mobile phone camera feeds a lot of picture products. Abroad, there are Instagram, Pinterest, etc. What about domestic photo products? In fact, several of my friends used to make photo products, but they didn’t do much. Why? The largest photo product in China is actually WeChat Moments.
Many people actually don’t know how to post plain text in WeChat Moments, they only know how to use it to post photos (If you don’t believe me, ask ten people around you to see how many of them know? In fact, WeChat Moments has this function, but it’s hidden , you need to press and hold the camera button). The default button in Moments is a camera button. Why did WeChat do this? Why didn't Zhang Xiaolong put the button for posting text and the button for posting photos side by side under the appeal of many people? It's very simple, if you randomly circle 100 people in a subway station in Shanghai, how many of these 100 people can post pictures and text? I think the gap will be quite astonishing. 99% of WeChat users will use Moments to post pictures. But I don't think more than 10% of users can send text. Writing words requires skills, but posting pictures is known to all women and children, and there is no threshold. This is the product power of different media.
Let’s talk about video again. When I was in Silicon Valley, several Google friends mentioned that TickTok is a greater threat to Google than ChatGPT. Why? Because TickTok users spend a lot of time online, it is squeezing Google's user time. The domestic version of Douyin is now the king of traffic. If WeChat does not have a video account in the past few years, it will not be a circle of friends in the early days. I think the territory of China's Internet is definitely not what it is today. As a comparison, if you look at WhatsApp today, its value may not even be one-twentieth that of WeChat. At the right time, WeChat seized the dividends of pictures (Moments) and videos (Video ID), and stepped on the banner every step of the way, so Zhang Xiaolong deserves the title of China's No. 1 product manager.
So the medium is also a very important axis of change. So what about after the video? It is the VisionPro released by Apple two weeks ago, the so-called three-dimensional (spatial computing). Text is a one-dimensional medium, pictures and videos are two-dimensional medium, and spatial computing is three-dimensional. Two-dimensional must surpass one-dimensional, and three-dimensional must surpass two-dimensional, which is why I am optimistic about VisionPro.
Therefore, the two biggest powers of paradigm shift in the future, the first is the paradigm shift brought about by the AGI large model on the technology axis; the second is the paradigm shift brought about by spatial computing on the media axis. The first one has already arrived, and the second one is estimated to be 2 to 3 years later according to the maturity speed of VisionPro products. The intersection of the two axes, I think the next 5 to 10 years will be very exciting.
But the big model does not have a huge product paradigm shift in all fields, let's look at the picture below
I divide the entry points of large-scale model products into two categories. One category belongs to the so-called AI-Native native applications that will undergo drastic paradigm shifts. The characteristic of this category is that there is no need for a large model, and the product cannot be made at all. Representative product categories in this category include: intelligent assistants (such as ChatGPT), AIGC generative products (such as Midjourney), and embodied intelligence (such as Tesla’s Optimus Prime). These products have huge disruptive opportunities and are new species in the AGI era.
The other type belongs to the progressive enhancement type, the so-called AI-Copilot co-pilot mode. Such as productivity tools (Office Copilot), code generation (Github Copilot), design enhancements (Adobe Firefly), and so on. This category is an enhancement of the original product using a large model in a certain link.
The paradigm shift and progressive enhancement brought about by AI-Native and AI-Copilot have many different features. I listed them in the picture above. Everyone should choose according to the characteristics of their business. What I need to emphasize here is the issue of organization. If you choose AI-Native paradigm-shifting innovation, you must reorganize the organization. It is easier to run a new startup company. If it is an internal innovation of a large company, you must also build a new team, otherwise it is likely to be unreliable. Because of many things, it is the reason of the organization. History has proved countless times that doing new things requires a new team structure. Because its organizational process is different, and its business model is also different. This is why the logic of the connection era and the logic of the computing era I mentioned earlier are different. If you choose AI-Copilot, you must have relatively mature existing products, so at this time, the old organization only needs to do some enhancements and fine-tuning. It is completely different from AI-Native's requirements for the team. This is also evident in many companies in Silicon Valley.
By the way, let me mention Microsoft. After the arrival of AGI, Microsoft has been waving the flag and shouting. Many people think that Microsoft is taking the path of paradigm shift. But I want to say that from the perspective of the big picture, Microsoft actually chose the progressive enhancement mode of AI-Copilot. Because Microsoft is already a $2.5 trillion company. It doesn't need a revolution, it just needs to ride the east wind of AGI and grow slowly. Let its Office, Windows, and Azure core products grow by about 30%, and in a few years it can become a $4 trillion or even $5 trillion company. It does not need to gamble its life and wealth to All-In paradigm-shifting AI-Native products, because although paradigm shifting has the possibility of high returns, it is also high-risk. Microsoft's product strategy choice in the AGI era is also worthy of reference for many mature companies in the track.
Part III: Six thoughts on the future development of AGI products
**The first thought is 2C or 2B. **Because 2C is an end-to-end product, the user path is usually relatively short, which is very suitable for AI-Native paradigm shift. However, due to the long customer path of 2B, the ability of AI is only a part of the entire business closed loop, so it is suitable for the gradual enhancement of AI-Copilot.
Focus on sorting by value. For 2C products, the content is greater than the service, and the service is greater than the tool. This logic is not only applicable in the era of PC Internet and mobile Internet, but also in the era of AGI. Think about it, when the mobile Internet era first started, many hot tools, such as image editing tools, anti-virus tools, note-taking tools, etc., were not very popular, but where are they all today? Even if they are alive, the overall value is far lower than all content companies, such as Xiaohongshu, Zhihu, and Douyin. Services mainly refer to products connected to services such as 58 and Ctrip. Their value is greater than tools, but lower than content. Why? Quite simply, the time spent by users on content is much longer than that on services, and the time spent on services is much longer than tools, which are not in the same order of magnitude at all. The user duration basically determines the value range of the product. Therefore, in the AGI era, don't be obsessed with those cool-looking tools, its value is far less than content or services.
In the 2B field, however, the value logic has completely changed. Customers' decision-making is greater than efficiency, and efficiency is greater than content. Enterprise products do not pursue a lot of content, but instead require maximum efficiency. It is best to assist me in making quick decisions.
The second consideration is whether to be a platform or an application. As soon as the big model became popular, many people's platform dreams were ignited. I gave cold water to my friend who cried and shouted to be a large-scale model platform as soon as I came up. I'll just quote a recent quote from Midjourney founder David Holz. He said that the biggest lesson he learned from his previous company, Leap Motion, was that everyone started with the platform dream of an ecosystem, and then failed. But he learned this lesson when he was working on Midjourney. He must make a product that users really like, and make it a super app first.
In fact, OpenAI is the same. It first has a super application like ChatGPT, and then builds the ecological construction of Plugin and API. The same is true for WeChat. It first has super application portals such as chat, Moments, and official accounts before building an ecology such as mini programs. Even when Jobs was working on the iPhone, he had a lot of good songs on iTunes first, and he went to Time Warner, Disney, and the New York Times to beg his grandparents to make apps on the iPhone and persuade many websites to do well. H5 is adapted to make Safari browse websites easy to use, and the iPhone has the status of an ecological platform.
This path is very important to our product people. Recently, I often meet some people who say that our goal is to build a XXX large-scale model platform. I can’t help but want to vomit. What is your product that has not been used? No one has ever used it. How can I have the courage to make a platform? Did you give it? No one will support you just because you claim to be a platform. Only when you have a super app and a huge user pool, you will have the appeal of the platform, and everyone will support you as a platform.
The third thought is UGC VS. AIGC. UGC used to be a very important strategy in Web 2.0. But the big model brings AIGC's capabilities.
UGC is a typical two-sided user network, while the content provider of AIGC is not a user, but a model + computing power + data. This is a data flywheel. The cold start of UGC is difficult, because you have to gather many creators. But AIGC puts computing power first and model first. UGC is low in cost because it takes a lot of wool from content providers. However, the cost of AIGC computing power is relatively high, especially at the time of startup. On the other hand, UGC has a relatively high social moat, because sometimes users are not just because they like the content, but because they are fans of the creator. But for AIGC, users simply consume content, and the switching cost is very low. If a big company does something similar to yours, but at a lower cost, users may switch over immediately because it is cheaper.
The fourth thought is the innovator's dilemma: Innovation vs. Conservative
Every time the paradigm shifts in history, the established giants will face the innovator's dilemma. For example, taking Google's current situation as an example, OpenAI has cleverly used the "innovator's dilemma" that Google faces.
In fact, many people in Google still don't pay much attention to large models such as GPT, because they have internally calculated that compared to index-based search queries, ChatGPT's neural network training is too expensive. Search is 90% of Google's revenue. How much revenue can a large model bring in? Still unknown. Therefore, it is difficult for Google to All in large models now, which gives OpenAI a very good period of strategic opportunities. In addition, OpenAI has cleverly used strategic cooperation to empower Microsoft's Bing to attract Google's firepower in the search battlefield through GPT, and it has the opportunity to run blindfolded in the field of large-scale ecological platforms. When the opportunity for paradigm shift arises, entrepreneurs must make good use of the "innovator's dilemma" to prevent giants from confronting themselves head-on.
To give a counter-example, when Lei Jun was doing Michat, Tencent was shocked by all kinds of earth-shattering publicity, but he did not prepare a strategic design to deal with giant competition in advance, so that Tencent was united and united in WeChat. In just over a year, It basically wiped out Michat. Of course, conversely, from the perspective of Tencent, when the paradigm shift occurs, if the established giants can pay enough attention and have enough All in, they can also get rid of the fate of the "innovator's dilemma".
The fifth thought is how AGI can cross the chasm
"Crossing the Chasm" is also a very famous book. The multiple-stage gaps it describes require special attention for many innovative products.
At present, ChatGPT has obviously crossed the first gap (that is, the small gap between Innovators and Early Adopters), and its users have exceeded 100 million. But the next gap is the biggest gap between Early Adopters and Early Majority, and ChatGPT is still trying to bridge it. Personally, I am more optimistic, although its recent data shows that Plugin has not yet passed PMF (Product Market Match). But according to my communication with OpenAI people in Silicon Valley, they still have a lot of big killers inside, especially its multi-modal ability is extremely powerful, far surpassing the brainless Midjourney. It's just that it still needs to do a lot of alignment compliance work. Of course, each product has its own gaps that need to be bridged.
The sixth thought is that the large model is just a door to AGI
If we look back at the history of the entire technology industry, when many technologies first came out, we felt that the technology was very powerful. Flock to this technology, and forget the paradigm shift in various fields brought about by this technology. For example, when browsers and web servers first came out, many people rushed to be browsers and web servers, because many technical people believed that browsers and web servers represented the Internet. The most tragic war is that Microsoft used the whole company to develop IE and Netscape to grab the browser market. I know this history quite well, because Marty Cagan, an old friend of our global product manager conference and the author of "Revelation", was the senior vice president of products at Netscape. I have invited him to China many times before, and I often talk about this period. history.
Microsoft and Netscape are both losers in this matter, because when they were fighting, they completely ignored the greater strategic opportunities after human beings entered the Internet. Web server market.
Now that many companies are flocking to large models, they are likely to repeat the mistakes of browsers and web servers back then. And completely forget that the big model brings a huge opportunity for industry-level paradigm shift. In this sense, the large model is just a door. After the door is opened, a more exciting AGI world is waiting for us.
Well, this is the end of my speech today. I hope that the whole content, especially the analysis of "Paradigm Transformation Cube", that is, ParaShift Cub and "Connection and Computing Era" will be helpful to everyone's innovation and entrepreneurship in the AGI era. Thank you Everyone!
View Original
The content is for reference only, not a solicitation or offer. No investment, tax, or legal advice provided. See Disclaimer for more risks disclosure.
Good in-depth article: Product layout and paradigm in the AGI era
Source: Research by Li Jianzhong
Part 1: In-depth understanding of the characteristics of the AGI era
First of all, let's get acquainted with the entire development of large-scale model technology through the following picture.
Due to my tracking and research on artificial intelligence-related technologies, I happened to have in-depth exchanges and discussions with several key figures on the above-mentioned development line. In 2018, we held the Global Machine Learning Conference in Shanghai, and invited Michael Jordan, the father of machine learning and a famous professor at UC Berkeley, as a keynote speaker, and had in-depth exchanges with him. In April 2021, when we held a machine learning conference in Beijing, due to the epidemic, we invited Lukasz Kaiser, one of the co-founders of the Google Transformer model, online. He told me shortly after that speech that he left Google to go to OpenAI, and later found out that he was investing in GPT 3.5. At that time, another deep learning veteran, Jurgen Schmidhuber, the father of LSTM, was invited. By April this year, I went to Silicon Valley and had a lot of in-depth exchanges with Ilya Sutskever, the chief scientist of OpenAI. In general, on the road of AI development, we have been maintaining in-depth discussions and exchanges with the frontiers of the industry, which have benefited me a lot.
Then, from the birth of the first computer ENIAC in 1946, the technological revolution of mankind entered a pendulum of "computing". Including mainframes in the 1950s, minicomputers in the 1960s, minicomputers in the 1970s, and PCs in the 1980s. With the von Neumann architecture as the core, it is all carried out around the proposition of "computing". This process has been going on for about 50 years.
Next came the emergence of the Internet from 1994 to 1995, marked by Netscape and Yahoo. The WWW Internet pushed the human technology revolution to the "connection" pendulum, followed by Web 2.0 in 2004, and marked by the birth of the iPhone in 2007. Mobile Internet, Cloud Services in 2013. They are all greatly developing and enriching the proposition of "connection". This process lasted about 30 years.
Time came to 2017, the paper on the Transformer model was published, GPT 1.0 was born in 2018, and then 2.0, 3.0, and ChatGPT brought by GPT 3.5 at the end of last year came out, bringing the dawn of general artificial intelligence AGI to the entire human being. This round of large models has entered the pendulum of "calculation", and the time will start from about 2020.
It's very interesting, if you look at the first "connection" revolution, I named it "connection 1.0", it went through about 100 years; and the second "computing" revolution, I named it "computing 2.0" ", it went through about 50 years; then came the second "connection" revolution, so-called "connection 2.0", it went through 30 years. Do you see an exponentially decreasing trend in the time span? Therefore, many people believe that we may not be able to do so in the next 15 years, and by 2035, the "computing" pendulum brought about by this wave of large models may reach its peak.
With the understanding of the timeline of the industry, let's take a look at the difference between the underlying logic of the "connection" era and the "computing" era? This is actually very important. Because many people often use various paradigms of the mobile Internet during this period to deduce the paradigm of the big model era. I think this perception is wrong. Because the mobile Internet belongs to the logic of "connection", while the era of large models belongs to the logic of "computation". Let's take a look at the table below:
Under the calculation logic, there is actually no such a bilateral effect, and its core is a unilateral market. One end is algorithm computing power, and the other end is consumers. For example, in Midjourney, there is no network of designers, and the algorithm directly produces pictures to users; for example, in the era of autonomous driving, if Didi is to be subverted, the moat formed by Didi’s driver-passenger two-sided market will be useless at all, because there is no need for drivers. It's the algorithm that's driving the car.
Looking at the business model again, under the connection logic, its marginal cost is very low, so it is easy to do a free business model. However, under computational logic, the cost of algorithm computing power needs to be shared equally. So will most business models still be like "free is king" in the era of the Internet and mobile Internet? Most likely not, which is why everyone sees that Midjourney and ChatGPT Plus are charging. Many Silicon Valley investors have suggested that "directly charging users" will be the main business model in the era of large models. It is no longer a free model of "the wool comes from the pig" in the connection era.
And what about user experience? In the Internet age, the connection logic is that the more information the better, an inevitable problem of information overload arises. But the calculation logic, in fact, is that the fewer contacts the better, and efficiency comes first. That is, I will just give me my results in the next order, and don't let me participate in the intermediate process.
In terms of decision-making mechanism. The connection logic is "the machine will give me information, and I will make decisions". And computational logic is that people give information to machines, that is, data, and machines help me make decisions.
Through the above comparisons, you can see that these two logics are very different in our product thinking. Next, let’s talk about computing logic. What is the essential difference between computing 1.0 from 1940 to 1990 and computing 2.0 we are currently in? Look at the picture below:
So you will find that under Computing 1.0, whether it is PC or mobile software, there are many places where the user experience is a little careless, and it is easy for non-industry people, including the elderly and children, to fall into a predicament of mental burden. Why this dilemma?
Because the left side is the biological neural network of the human brain, and the right side is a digital logic circuit composed of "and, or, and not", you have to adapt it. Any input input must first be converted into something that a digital logic circuit can understand; any output from a computer must undergo a layer of conversion into something that a biological neural network can understand. This back-and-forth conversion is where the untrained average user gets into trouble.
But if we look at the computing 2.0 era dominated by the large model described in the figure below.
When I communicated with OpenAI’s chief scientist Ilya Sutskever in Silicon Valley in April, I remember he repeated three times to emphasize that from the perspective of mathematical principles, the current digital neural network is no different from the carbon-based neural network of the human brain. Except for the different survival methods of biological organisms, the logic of operation is the same.
At this time, the interaction between humans and the large model is actually the interaction between the biological neural network and the digital neural network.
It is no longer the 0-1 thinking of digital circuits in the computing 1.0 era, but a probabilistic thinking. Everyone knows that you can adjust the temperature (temperature value) parameter of ChatGPT. Turn up the temperature to get creative. It is also very interesting to use the word temperature. A high temperature means that the brain is hot, and the brain is more creative when the brain is hot. You can also turn down the temperature to make the answer very mediocre. The essence of probabilistic thinking is the model of neural network, which is also the decision-making method of our human brain. Think about it carefully, everyone. In fact, every thought in our brain is calculating probability. The result of the entire orientation of Computing 2.0 is to allow machines to adapt to human thinking.
With the above in-depth understanding of the entire AGI technology and the logic behind it, let's talk about some methodologies on how to make products in the AGI era.
Part II: "Paradigm Shift Cube" of Technological Innovation
Let’s talk about paradigm shift first. It is called Paradigm Shift in English. It was first proposed by Thomas Kuhn in the book "The Structure of Scientific Revolutions". It refers to the fundamental changes in the basic concepts and practices in the field of science and technology. It breaks the original The laws and boundaries of human beings create a new world for people's thoughts and actions. The revolutions we often talk about in the field of technology are actually paradigm shifts. For example, from stand-alone to the Internet, from the Internet to the mobile Internet, are all paradigm shifts.
In the Chinese Internet field, many friends know that Wang Xing of Meituan once had a famous four-vertical and three-horizontal theory around 2009. At that time, many people on the Internet were very excited when they heard this theory. Because of Wang Xing's analysis, it seemed to be more in line with the current situation at that time. 2009 was when the mobile Internet just broke out. It is very similar to the current AGI outbreak.
Wang Xing pointed out at the time that human beings are basically dominated by four major needs in the technology industry: information, communication, entertainment, and business. The so-called four verticals. As for Sanheng, Wang Xing gave his own analysis: search, social, mobile. Then at the intersecting point of four vertical and three horizontal, a new opportunity was born. I agree with the "four verticals" very much, but for the three horizontals, I think this analysis method is problematic, because social networking and search are actually human needs, and they are not on the same dimension as the mobile Internet.
I guess Wang Xing put social networking on the same dimension as mobile at that time because the social network was so popular at the time that social networking became a huge traffic portal. Once social networking is encountered in many fields, there will be new ways to play. But traffic entry is not equal to technology. The same is true for search, because it also has a traffic entrance effect, so Wang Xing feels that it has also caused changes in many fields. However, Wang Xing did not give a convincing product model in the fields of communication, entertainment, and business other than the Google/Baidu model for search. So I think the "three horizontals" of Wang Xing's "four verticals and three horizontals" methodology are misplaced.
Next, let me talk about an analysis framework I proposed. I think this analysis framework is very helpful for us to analyze the product paradigm in the AGI era. I named the analysis framework I proposed "Paradigm Transformation Cube", and the English name is ParaShift Cube, where Para is the abbreviation of Paradigm.
The x-axis represents the technology axis, which represents the transformation of technological paradigms. There are connection 1.0, computing 1.0, connection 2.0 (including PC Internet, mobile Internet), computing 2.0, which is what we call the AGI intelligent era.
The Y axis represents the demand axis, covering all aspects of human needs, including the part Wang Xing talked about: information, entertainment, and business. At the same time, I also put social networking and search on the Y axis. Social and Communication, I combined them, and while they have slight differences, their commonalities outweigh their differences.
The Z axis represents the medium axis, because the expression of information requires a specific medium, and human beings also interact with the world around them through the medium. The media axes include: text, pictures, audio, video, and three-dimensional space calculation (this is Apple's latest proposal in VisionPro). I think the Z axis is a point that many people ignore. Including the GUI graphical user interface, which has a great influence on the development of computers, is also the result of interaction driven by events in a medium such as computer images.
These three axes form a cube structure. The three axes, the point of change on each axis, and the intersection of other axes are the places where "paradigm transformation innovation" occurs in products. Applying the Paradigm Shift Cube to analyze innovation opportunities in product areas is logically straightforward.
Next, we use the "Paradigm Transformation Cube" to analyze the connection to the PC Internet in the 1.0 era, connect to the mobile Internet in the 2.0 era, and predict the possible opportunities in the era of computing 2.0 AGI intelligence. As for the computing 1.0 and connection 1.0 era, in fact, you can also use the "paradigm transformation cube" analysis, but the age is relatively old, so I won't talk about it here, but if you are interested, you can go back to history for analysis, and you will find that it is also very logical .
I split the demand axis in the table into two parts: red and green. The red part, in the era of AGI intelligence, is the part where the paradigm shift is more drastic, and there is a great chance of innovation and even overthrowing the giants; first look at the information demand, in the transformation from PC to mobile, portal models such as Sina and Sohu are being replaced by Toutiao, etc. Defeat, and the production method of information in the age of AGI intelligence will face huge changes. Now the information overload is very serious. Sometimes, after reading ten articles, you actually get only one piece of information. So can the big model help me integrate dozens of pieces of information from the previous day according to my preferences, and then give me some short and concise summaries, which is what the big model is very good at. This can easily lead to a paradigm shift in information products. Let’s talk about entertainment again. Entertainment needs content very much, and AIGC is very good at content generation. The form, carrier, supply and other chains of entertainment will be reshaped by AIGC, so there are also many opportunities for paradigm shift.
Finally, let’s talk about search. Search is a big industry in the Internet field, and it is also a big demand for human beings. But if you use ChatGPT and other large-scale model products, you will find that the proportion of search behavior will drop significantly. Why? Because many times, we search a dozen articles and turn several pages just to find an answer or a decision. Is this product any good? Is that attraction worth visiting? Does this medicine solve the problem? These decision-making and answering needs are what large-scale model products such as ChatGPT are very good at. Therefore, search will face a huge possibility of being subverted in the AGI era. Unfortunately, when I returned from Silicon Valley in April and May, I felt that Google was still counting money on the search credit book, and was numb to the threat of the huge paradigm shift brought about by ChatGPT.
Some people say, hasn't Google launched a lot of technologies and products related to large models? My answer is, don't look at the few products launched on the surface. These defensive actions are all drizzle when the paradigm shift comes. Think about how Yahoo collapsed that year? Of course, does Google have a chance? I think there is, but the premise is that Google must come up with determination similar to the war that Ma Huateng and Zhang Xiaolong used the power of Tencent's entire company to All in WeChat to fight Michat; or Bill Gates in 1995 The determination to fight the browser war with Netscape by using the strength of Microsoft's entire company to All in IE every year can prevent the subversive force of paradigm shift. And what did Google do? The founder lay on the beach in California and basked in the sun, and let the CEO who was hired to get the annual salary and OpenAI's Sam Altman and Ilya Sutskever and other co-founders with poetry and distance to fight, and then the technology master Jeff Dean was removed from the position of the boss of Goolge Brain, and Jeffrey Hinton, the father of deep learning, was tactfully dismissed. Do you want to win the AGI era by relying on a few large PPT models at the Google I/O conference? By the way, I would like to criticize the impetuous style of Google's Daxing PPT products in recent years (many of which have not been officially released after that), and let's see when startups such as OpenAI need a development conference? A product release is a press conference, with at most a blog explaining it. Talking about products (rather than PPT) is the king of innovation!
Finally, let me say that it is not another portal that overturns Yahoo; it is not another search that overturns Google, but the structural subversion brought about by the paradigm shift, and this is the chess game that the search field is facing in the era of large models.
After talking about the red part, let's talk about the green part. Compared with the red part, the green part still has the "two-sided market" effect of the connection era, so the moat is relatively deep. Even when the technology undergoes a paradigm shift, I think it is not There are great opportunities. For example, in the transformation from PC Internet to mobile, even though Michat, Momo and other companies attacked social networking at that time, Tencent survived (although it changed its name from QQ to WeChat), because the social networking established by Tencent in the QQ era The moat is too deep. The entire social relationship in China is not a simple bilateral structure, but a network structure, which is in the hands of Tencent.
Many people also asked some time ago that in the AGI era, will WeChat be subverted? We use the above paradigm transformation cube analysis, the conclusion is not. First of all, in the AGI era, will people’s social needs change? Won't. Secondly, will the social model between people and people become the social model between people and virtual people? I don't think so either. There is a saying in Silicon Valley that robots in the future will be slaves to humans, and people use robots for a clear purpose of "making it work." The need for social interaction comes from the structure of human social relationships: friends, relatives, classmates, colleagues... unless there are no such social relationships in the human social structure in the future. Nobody wants to socialize with a robot. Therefore, the advantage of WeChat in China, and more precisely, it should be the social network of Chinese people all over the world, will not be subverted even in the era of AGI's major paradigm shift. As for WeChat, there are some specific interactive updates, such as text and image generation, and smarter information optimization. These are small interactive improvements. These things can be done in minutes with WeChat’s ability, and it’s not up to any start-up company to do it. Subvert it.
I remember that on the same day in January 2019, Byte launched Duoshan, Luo Yonghao launched Chat Bao, and Kuaibo founder Wang Xin launched Toilet MT. They wanted to cooperate to overthrow WeChat’s status as the social king, but they died within two months. . If the founders had read my "Paradigm Shift Cube" at that time, they would have understood that this matter cannot be done. What's more, 2019 is not an era of technological paradigm shift. Even in the era of PC Internet to mobile paradigm shift, Lei Jun's Michat, international WhatsApp, Talkbox and other applications have not subverted Tencent, because the social moat is the deepest. In the AGI era, everyone should not think about using WeChat, there is no door.
Let’s talk about business, which actually refers to e-commerce. There are not many opportunities for major disruption, and the reason is that the moat of the two-sided market in the "connection era" still exists in the e-commerce field. Any business form is inseparable from buyers and sellers. The AGI large model can only optimize some small links in the closed business loop (such as Taobao pictures, virtual fitting rooms, etc.), but these can never be separated from buyers, sellers, logistics, etc. Basic e-commerce attributes, and the advantages of traditional giants in these fields are still great. That is, in the transformation from the PC Internet to the mobile Internet, the advantages of Taobao and JD.com are basically translated.
However, one thing needs to be pointed out. In the field of mobile Internet, Pinduoduo and Douyin e-commerce have emerged halfway, but this is brought about by the traffic entrance effect of WeChat and Douyin. It is a good supplement, but it does not constitute a subversion. Moreover, Pinduoduo hits the sinking crowd, which is brought about by the new Internet demographic changes (this has the opportunity to re-analyze). My judgment is that as large-scale models may bring new traffic entrances, new opportunities similar to Pinduoduo and Douyin e-commerce may emerge in the e-commerce field, but the existing advantages of Taobao and JD.com will still exist. After all, the plate of e-commerce is too big. The Pacific Ocean can accommodate China and the United States, and the e-commerce field can accommodate many small giants.
The blue axis at the bottom is mainly the media we mentioned earlier: text, pictures, audio, video, and three-dimensional (spatial computing). It is easy for many people to overlook this part. The power of products behind different media is very different. When we make any product, we must think about what kind of media carrier we want to focus on.
In terms of media logic, pictures are better than text, video is better than pictures, and audio has its own special scene. It is estimated that many of the note-taking and text-based products in the early mobile field have disappeared. A mobile phone camera feeds a lot of picture products. Abroad, there are Instagram, Pinterest, etc. What about domestic photo products? In fact, several of my friends used to make photo products, but they didn’t do much. Why? The largest photo product in China is actually WeChat Moments.
Many people actually don’t know how to post plain text in WeChat Moments, they only know how to use it to post photos (If you don’t believe me, ask ten people around you to see how many of them know? In fact, WeChat Moments has this function, but it’s hidden , you need to press and hold the camera button). The default button in Moments is a camera button. Why did WeChat do this? Why didn't Zhang Xiaolong put the button for posting text and the button for posting photos side by side under the appeal of many people? It's very simple, if you randomly circle 100 people in a subway station in Shanghai, how many of these 100 people can post pictures and text? I think the gap will be quite astonishing. 99% of WeChat users will use Moments to post pictures. But I don't think more than 10% of users can send text. Writing words requires skills, but posting pictures is known to all women and children, and there is no threshold. This is the product power of different media.
Let’s talk about video again. When I was in Silicon Valley, several Google friends mentioned that TickTok is a greater threat to Google than ChatGPT. Why? Because TickTok users spend a lot of time online, it is squeezing Google's user time. The domestic version of Douyin is now the king of traffic. If WeChat does not have a video account in the past few years, it will not be a circle of friends in the early days. I think the territory of China's Internet is definitely not what it is today. As a comparison, if you look at WhatsApp today, its value may not even be one-twentieth that of WeChat. At the right time, WeChat seized the dividends of pictures (Moments) and videos (Video ID), and stepped on the banner every step of the way, so Zhang Xiaolong deserves the title of China's No. 1 product manager.
So the medium is also a very important axis of change. So what about after the video? It is the VisionPro released by Apple two weeks ago, the so-called three-dimensional (spatial computing). Text is a one-dimensional medium, pictures and videos are two-dimensional medium, and spatial computing is three-dimensional. Two-dimensional must surpass one-dimensional, and three-dimensional must surpass two-dimensional, which is why I am optimistic about VisionPro.
Therefore, the two biggest powers of paradigm shift in the future, the first is the paradigm shift brought about by the AGI large model on the technology axis; the second is the paradigm shift brought about by spatial computing on the media axis. The first one has already arrived, and the second one is estimated to be 2 to 3 years later according to the maturity speed of VisionPro products. The intersection of the two axes, I think the next 5 to 10 years will be very exciting.
But the big model does not have a huge product paradigm shift in all fields, let's look at the picture below
The other type belongs to the progressive enhancement type, the so-called AI-Copilot co-pilot mode. Such as productivity tools (Office Copilot), code generation (Github Copilot), design enhancements (Adobe Firefly), and so on. This category is an enhancement of the original product using a large model in a certain link.
By the way, let me mention Microsoft. After the arrival of AGI, Microsoft has been waving the flag and shouting. Many people think that Microsoft is taking the path of paradigm shift. But I want to say that from the perspective of the big picture, Microsoft actually chose the progressive enhancement mode of AI-Copilot. Because Microsoft is already a $2.5 trillion company. It doesn't need a revolution, it just needs to ride the east wind of AGI and grow slowly. Let its Office, Windows, and Azure core products grow by about 30%, and in a few years it can become a $4 trillion or even $5 trillion company. It does not need to gamble its life and wealth to All-In paradigm-shifting AI-Native products, because although paradigm shifting has the possibility of high returns, it is also high-risk. Microsoft's product strategy choice in the AGI era is also worthy of reference for many mature companies in the track.
Part III: Six thoughts on the future development of AGI products
**The first thought is 2C or 2B. **Because 2C is an end-to-end product, the user path is usually relatively short, which is very suitable for AI-Native paradigm shift. However, due to the long customer path of 2B, the ability of AI is only a part of the entire business closed loop, so it is suitable for the gradual enhancement of AI-Copilot.
In the 2B field, however, the value logic has completely changed. Customers' decision-making is greater than efficiency, and efficiency is greater than content. Enterprise products do not pursue a lot of content, but instead require maximum efficiency. It is best to assist me in making quick decisions.
The second consideration is whether to be a platform or an application. As soon as the big model became popular, many people's platform dreams were ignited. I gave cold water to my friend who cried and shouted to be a large-scale model platform as soon as I came up. I'll just quote a recent quote from Midjourney founder David Holz. He said that the biggest lesson he learned from his previous company, Leap Motion, was that everyone started with the platform dream of an ecosystem, and then failed. But he learned this lesson when he was working on Midjourney. He must make a product that users really like, and make it a super app first.
This path is very important to our product people. Recently, I often meet some people who say that our goal is to build a XXX large-scale model platform. I can’t help but want to vomit. What is your product that has not been used? No one has ever used it. How can I have the courage to make a platform? Did you give it? No one will support you just because you claim to be a platform. Only when you have a super app and a huge user pool, you will have the appeal of the platform, and everyone will support you as a platform.
The third thought is UGC VS. AIGC. UGC used to be a very important strategy in Web 2.0. But the big model brings AIGC's capabilities.
The fourth thought is the innovator's dilemma: Innovation vs. Conservative
Every time the paradigm shifts in history, the established giants will face the innovator's dilemma. For example, taking Google's current situation as an example, OpenAI has cleverly used the "innovator's dilemma" that Google faces.
To give a counter-example, when Lei Jun was doing Michat, Tencent was shocked by all kinds of earth-shattering publicity, but he did not prepare a strategic design to deal with giant competition in advance, so that Tencent was united and united in WeChat. In just over a year, It basically wiped out Michat. Of course, conversely, from the perspective of Tencent, when the paradigm shift occurs, if the established giants can pay enough attention and have enough All in, they can also get rid of the fate of the "innovator's dilemma".
The fifth thought is how AGI can cross the chasm
"Crossing the Chasm" is also a very famous book. The multiple-stage gaps it describes require special attention for many innovative products.
The sixth thought is that the large model is just a door to AGI
If we look back at the history of the entire technology industry, when many technologies first came out, we felt that the technology was very powerful. Flock to this technology, and forget the paradigm shift in various fields brought about by this technology. For example, when browsers and web servers first came out, many people rushed to be browsers and web servers, because many technical people believed that browsers and web servers represented the Internet. The most tragic war is that Microsoft used the whole company to develop IE and Netscape to grab the browser market. I know this history quite well, because Marty Cagan, an old friend of our global product manager conference and the author of "Revelation", was the senior vice president of products at Netscape. I have invited him to China many times before, and I often talk about this period. history.
Microsoft and Netscape are both losers in this matter, because when they were fighting, they completely ignored the greater strategic opportunities after human beings entered the Internet. Web server market.
Well, this is the end of my speech today. I hope that the whole content, especially the analysis of "Paradigm Transformation Cube", that is, ParaShift Cub and "Connection and Computing Era" will be helpful to everyone's innovation and entrepreneurship in the AGI era. Thank you Everyone!