jakiro
Source: Investment Internship
After the rise of this wave of AI, we often compare it with the previous Internet wave, thinking that the rise of AI companies is like the rise of Internet companies.My feeling is that the two are still very different,AI feels more centralized, while the Internet has to be decentralizedEssence
Recently, the two founders of A16Z, Marc Andreessen and Ben Horowitz, just talked about this topic, and they believed that this kind of ratio was not completely applicable.In essence, in terms of technology,Marc Andreessen believes that the Internet is a network, and AI is more like a computerEssence
01. Differences in the nature of technical nature
The Internet is essentially a network connecting many existing computers and promoting the creation of new computers. It is dominated by network effects. As more people join, And prompt people to build more new computers to connect the Internet.
AI, especially large language models, is a new type of computer, a probability, neural network -based computer, which is very different from the previous von Normandan computers (certain computer).
The development of AI is more similar to the era of microprocessors or large machines. It handles data, learn from it, and generate output.It is an information processing system, not a network.
The original computer was very large and expensive, and people thought that only a few computers in the world had.But over time, the computer becomes smaller and cheaper, and eventually becomes everywhere.
The development of AI will also follow similar models. In the end, there will be AI models of various shapes, sizes and abilities. They will train based on different types of data and run on different sizes. They have different privacy and security policies.
02. The difference in the development level of the industry
In the Internet era, everyone focuses on building networks and applications that use network effects.The company strives to fight and maintain the user base to use these influences.
In the AI era, the core involved the construction of various AI models and applications, focusing on improving the functions of these models and integrating them into different fields.
03. Locking and availability
Ben Horowitz believes that unlike previous computers, AI is the most easy to use to date, because it can speak natural language like English, just like talking to people.
This triggers the problem of lock-in effect about AI, that is, whether the user completely chooses the size, price, selection speed of the AI model they needs, or whether it will be locked on a specific large modelEssence
In the Internet era, the lock is very important due to the complexity of the early computer and the network, and the conversion cost is high.
In the AI era, artificial intelligence is easier to use, because it can understand and generate human language, which reduces locking and provides greater flexibility when choosing an artificial intelligence solution.
04. Similarity of two waves
The first is speculative prosperity and depression: the two waves have gone through speculative investment cycle. The initial excitement triggered over -investment, and the collapse was expected to fail to achieve it later.This cycle is typical for new, transformer technology.
Economic and cultural influence: Both technologies have a profound impact on economy and culture.The Internet has completely changed communication, business and information sharing.Artificial intelligence is expected to change the industry through automated tasks, enhance decision -making and create new capabilities.
05. Lesson experience
Prosperity and depression: It is expected that excessive investment and correction cycle is expected. This is a natural component of transformer technology adopting curves.
Open system and proprietary system:The Internet starts with a proprietary network and then turns open, which has promoted its growth.The risk of artificial intelligence towards a more closed system may kill innovation and competition.
Speculation investment: speculative investment is a double -edged sword.It may lead to the rapid development and deployment of new technologies, but it can also lead to major financial losses when it meets expectations.
06. Future Outlook
Artificial intelligence model: The future of artificial intelligence may involve a variety of scale and models of models, which are similar to the evolution of computers from large machines to microprocessors.
Integration and application: The focus will be placed on applications that use artificial intelligence to various fields and create applications with unique artificial intelligence capabilities.
The following is a simple text translation version of the dialog video. You can watch the original dialogue video here (https://dub.sh/memo2):
Marc Andreessen:What is the strongest common theme between the current AI status and web 1.0?Ben, let me give you a theory first to see what you think.
Because of my role and your role in NetScape, we participated in the development of early Internet and often asked this question.Therefore, the Internet boom is a major event in the technical field and is still in the memory of many people.
People like to reason from analogy, so they think that the AI boom should be like the Internet boom. Founding AI companies should be like starting an Internet company.So, what are the similarities between the two?
In fact, I think this type ratio is not established in most cases.It may be established in some ways, but in most cases, it is not applicable.The reason is,The Internet is a network, and AI is a computerEssence
Let’s explain this view.PC boom, or microprocessor boom, I think the best analogy is microprocessor, which can even go back to the original computer, that is, the console era.
The reason is that the Internet is a network, which connects many existing computers. Of course, people have also established many new computers to connect the Internet.butFundamentally, the Internet is a networkEssence
Most of the industry’s industry dynamics, competitive dynamics, and entrepreneurial dynamics are related to the construction of the network or applications running on the network.In the Internet era, startups are very concerned about the network effects and various positive feedback cycles generated after connecting a large number of people, such as the so -called Metkov law, that is, the value of the network expands with the increase of users.
[MEMO Note: Metcalfe ’s Law is a network effect theory proposed by Robert Metcalfe.The law points out,The value of a network is proportional to the square of the number of network usersEssenceSpecifically, when the number of users in the network increases, each user can connect to more other users, so that the overall value of the network will increase significantly.Metkov’s law can be represented by a simple formula: v n², where (v) represents the value of the network, n represents the number of users in the network..
AI also has a network effect in some ways, but it is more like a microprocessor, more like a chip, and more like a computer.It is a system, data entry, data processing, data output, and then things happen.This is a computer, an information processing system, and a new computer.
We like to say that the computers so far are so -called von Nomann, that is, a certain computer, they are very strict. They do exactly the same thing every time.butThey are very limited in interacting and understanding the world in the worldEssence
We believeAI and large language models are a new type of computer, a probability computer, and a computer -based computerEssenceThey are not accurate, they will not give the same results every time, and may even argue with you without answering your question.
This makes them very different from old computers in nature, making the complexity of building large -scale systems, but their abilities are new, different, valuable, and important, because they can understand language and images, do it, do it, do it, do it, do it, do it, do it, do it, do it, and do it.Everything you see when using.
So I thinkAnalogy and lessons are more likely to come from the early stages of the computer industry or the early stages of the microprocessor, rather than the early stages of the InternetEssenceDo you think this is right?
Ben horowitz:I think it is right.Although this does not mean there is no boom and decline, because this is the essence of technology.People will be too excited and then frustrated.Therefore, there will definitely be some excessive construction problems, such as the excessive construction of chips and electricity.I agree that the network is essentially different from the computer in development, and the use of curves and all these things will be different.
Marc Andreessen:This also involves the best theory of the future development of the industry, that is, about how the industry will develop, such as whether there will be a few “god models” or a large number of different sizes.
In the early days of the computer industry, such as the original IBM host, the computer was very large and expensive, with only a few.For a long time, the general point of view is that this will be the entire computer.The founder of IBM, Thomas Watson, once said a famous words, and he believes that the world may only need five computers.
I think he means that the government will have two, three major insurance companies will have three, and then there will be no other places that need so much computing power.The computer is very large and expensive. Who can afford it?Who can afford the persons they need?
These computers are very large, require special buildings to accommodate, and personnel who need to wear white coats to maintain, because everything must be kept very clean, otherwise the computer will stop working.
Today we have the concept of AI’s “God model”, that is, a large basic model. At that time, we had the concept of God’s host, and there were only a few such computers.If you look at the early science fiction, there is almost always this setting, that is, there is a large super computer. It either does the right thing or do something wrong. If you do it wrong, the plot of science fiction movies is usually you wantGo in to repair or defeat it.
This top -down single concept was established for a long time, especially before the computer started to be smaller.Then the so -called small computer appeared, the price dropped from 50 million US dollars to $ 500,000, but even if it was $ 500,000, it was very expensive. Ordinary people would not put small computers at home. Therefore, medium -sized companies can buy small computers.But individual can’t.
With the emergence of the PC, the price dropped to $ 2,500, and the appearance of smartphones was reduced to $ 500.Today, you have various shapes, sizes, and described computers. It may only spend a penny, which may be some embedded ARM chips and firmware, with billions of such computers.
Buy a new car today, there may be 200 computers, even more.Today you assume that everything has chips, assume that everything needs battery or electricity, because they need to power the chip and assume that everything is on the Internet, because all computers are considered to be on the Internet, or they will be on the Internet or the Internet.superior.
Therefore, the computer industry is a huge pyramid today. There are still a few supercomputers or large hosts. They are god models, and then there are more small computers, more PCs, more smartphones, and then more smartphones. ThenThere are a lot of embedded systems.
It turns out that the computer industry is a collection of all these things.What kind of computer do you need depends on what you want to do, who you are, and what you need.
If this type ratio is established, this meansIn fact, we will have AI models of various shapes, sizes, descriptions, and abilities. Based on different data training, they operate in different sizes, have different privacy policies, different security policies, and there will be huge diversity and changes.It will be a complete ecosystem, not just a few companiesEssenceWhat do you think of this view?
Ben Horowitz: I think it is right.I also believe that another interesting thing about this era of calculation is that if you look at the previous computing era, from the host to smartphones, a huge source of locking is the difficulty of using them.
No one is fired because you have bought IBM, because you have trained people, knowing how to use the operating system, and due to the huge complexity of the computer, choosing IBM is a safe choice.
Even a smartphone,Why is Apple smartphone so powerful because the cost and complexity of switching from it is very highEssenceAI is the easiest computer to use. It speaks English, like talking to people.
Where is the lock?Do you completely choose the size, price, and speed that suits your specific tasks, or is it locked in the god model?I think this is an open problem, but it is very interesting, which may be very different from the previous period of calculation.
