An article by Zhang Xiaojun, a reporter from Tencent Technology Channel"Lu Qi’s latest speech record - My big model world view" Popularized the circle of friends. The article is rich in knowledge, and there are entrepreneurial methodologies and AI application business opportunities flowing between the lines. It is extremely valuable!
PS: It’s a family’s opinion and no one agrees. If you are unhappy with anything, you are welcome to come to me.
Tencent News original text:
"Even Lu Qi said that he could not keep up with the rapid speed of the big model era. He asked his subordinates to do a "big model daily report", which on the one hand facilitated him to keep up with papers and information updates, and on the other hand, it was shared with Qiji eco-entrepreneurs . He used three "realities" to express this point. "I really can't do it anymore, I really can't keep up with the paper, and I really can't keep up with the code." Just too much. "Lu Qi said at a recent sharing event. "
Interpretation:
Recently, LLM large model-related technology releases have been too intensive. Before I finished digesting ChatGPT, the following ones immediately popped up. It is really fascinating!
Tencent News original text:
Lu Qi, the founder and CEO of Qiji Chuangtan, is China’s AI evangelist and one of the most vocal people in China regarding large models. He once held important positions in global giants, including IBM, Yahoo, Microsoft, and Baidu. He was once the most authoritative senior Chinese person in American technology companies, serving as executive vice president of Yahoo and Microsoft. He returned to China to join Baidu as group president and COO. Lu Qi is known in the technology circle for his hard work—getting up at 4 a.m. every morning, running 5 miles, and arriving at the office on time at 6 a.m.
At the same time, he has a deep relationship with OpenAI. The predecessor of Qiji, led by Lu Qi, is YC China, which is the Chinese branch of the famous American startup incubator YC (Y Combinator). He is also the director of YC Global Institute. OpenAI CEO Sam Altman is YC’s second-generation successor and current president. Although the two are 24 years apart, they have known each other for more than 18 years. It was Sam Altman who repeatedly invited Lu Qi to join YC. Therefore, Lu Qi has long-term close observation of YC, Sam Altman and OpenAI. I met Sam Altman in 2005. He was less than 19 years old at the time, and I was already in my 40s. We have been friends for many years. He is a very kind and strange child. Today I am very happy that he can change the world in this way. Not long ago, I was in the United States for three months during the Spring Festival, and I went to OpenAI to chat with Sam.
Interpretation:
To put it simply, Dr. Lu Qi has a close relationship with Sam Altman, the founder of OpenAI, and may be one of the few Chinese people who can talk to him and understand the inside story of ChatGPT (this is crucial). This means that his views have very important reference value, because most people who do not know the inside story of OpenAI basically rely on guesswork to analyze and evaluate it, and do not have such a good reference value.
Tencent News original text:
Lu Qi hopes to help Chinese entrepreneurs recognize this historic turning point, locate the coordinates of today's era, and find their own position. "This era is very similar to the gold rush era," he said. "If you went to California to dig for gold at that time, a lot of people would die. But people who sell spoons and shovels can always make money."
Interpretation:
I agree with Dr. Lu Qi’s view of the “AI gold rush era”. He himself is inextricably linked to Sam, so he naturally does not need to end up “gold rush” and help OpenAI invest and incubate upstream and downstream related AI ecological projects “selling shovels” "That's it.
Tencent News original text:
First of all, how to understand the new social paradigm brought by AI? This picture can clearly explain everything brought about by ChatGPT and OpenAI. Then, based on first principles, you will naturally deduce the opportunities and challenges of the track you are in.
This picture is the "Trinity Structural Evolution Model" including:
"Information" system (subsystem of information), obtains information from the environment;
"Model" system (subsystem of model) expresses information for reasoning and planning;
"Action" system (subsystem of action), we ultimately interact with the environment to achieve the goals humans want to achieve.
Interpretation:
The new turning point and new paradigm discussed by Dr. Lu Qi are the core essence of the entire article. Friends, if you understand this paragraph, it is better than thousands of other words! Please read the following content carefully, it is full of useful information!
Tencent News original text:
Most digital products and companies today, including Google, Microsoft, Alibaba, and Byte, are essentially information handling companies. We must remember that everything we do, everything, including most of the companies here, are moving information. Nothing more than that, You just move bytes. But it was good enough to change the world.
The reason is that the marginal cost of acquiring information starts to turn into a fixed cost.
Any change in society or industry will always be a structural change. This structural change tends to be a large category of costs that now move from marginal costs to fixed costs.
For example, a map cost $3 back then. Today I want a map. Google pays an average of US$1 billion a year to make a map, but the basic cost for each user to use the map is 0. In other words, when the cost of obtaining information becomes 0, it must change all industries.
Why is Google great? It turns marginal costs into fixed costs. Google's fixed costs are high, but it has a simple business model called advertising. It can become one of the most profitable companies in the world and change the world. This is the key to an inflection point.
Interpretation:
The new turning point Dr. Lu Qi talks about here: "The marginal cost of obtaining information becomes a fixed cost!" We can simply understand it as: AI makes it extremely easy for us to obtain information, making the information itself worthless, but the traffic is valuable. ! Back then, Google and Baidu provided free search services to aggregate traffic and made more money by selling ads! The same goes for today's AI explosion. All the knowledge in human history is in the "brain" of ChatGPT. If you want to know something, you don't need to search, just ask it.
Tencent News original text:
What is the inflection point in 2022-2023? AI, ChatGPT!
It’s overwhelming and why? The cost of the model is moving from marginal to fixed because of something called a large language model.
The cost of models has begun to shift from marginal to fixed, and large models are the core of technology and the basis of industrialization. OpenAI has built the large model of ChatGPT, and its development speed has increased rapidly. Why is the large model so important and this inflection point so important? Because there is an inherent relationship between models and people. Each of us is a combination of models. There are three models of people:
Cognitive model, people can see, listen, think, and plan;
Task model, people can climb stairs, move chairs and peel eggs;
Domain model, some people are doctors, some people are lawyers, and some people are programmers.
All our contributions to society are a combination of these three models.
Just think about it, what would happen if you don’t have much insight, but all the big models have your model capabilities, or the big models will gradually learn all your capabilities?
——In the future, the only thing of value is how much insight you have!
Interpretation:
No matter how powerful AI is, it is still just a "brain in a vat", and human value cannot be replaced by AI in the short term. Dr. Lu Qi’s saying, “The only value is how much insight you have”, I think includes two levels:
1. Your knowledge and opinions on AI;
2. Your knowledge and insights into the objective world;
Simply put, you have to spend time understanding AI, learning AI, using AI, and knowing what capabilities it has and what problems it can help you solve. Secondly, you must be good at observing the world and people's hearts, and know what the underlying needs of all living beings are. Then, combine the two, and business opportunities arise!
Tencent News original text:
Human society is driven by technology. With mechanization, electrification, and electronics, human physical labor has declined. Since the information age, people have been mainly engaged in mental work, and the economy has shifted from a commodity economy to a service economy - coders, designers, and analysts have become typical professions of our era.
This big model inflection point will basically affect all people in the service industry and blue-collar workers, because they are just models. Unless you have unique insights, the big service models you are engaged in today can all be replaced. We believe that the typical professions of the next era are entrepreneurs and scientists.
Therefore, this AI revolution affects everyone. It affects society as a whole.
Interpretation:
Dr. Lu Qi believes that the typical professions of the next era are entrepreneurs and scientists. I deeply believe that, and I believe that in the AI era, people who are good at using tools such as ChatGPT will become "super individuals" in the new era - everyone will be Iron Man in the future, and AI will be our "Jarvis"! With the assistance of AI, the gap between people will become smaller and smaller, and everyone will be "omniscient and omnipotent". For example, use AIGC tools to create content and solve problems for people or companies on the other side of the world (similar to the enhanced version of Bangalore coders who provide technical service outsourcing for Silicon Valley IT companies). Relatively higher economic income and relatively lower cost of living can allow them to have a better quality of life. Is this also using AI technology to promote common prosperity and rural revitalization?
Tencent News original text:
Which companies can reach the next turning point? I think Tesla has a high probability. Its autonomous driving and robots are very powerful now. Microsoft is climbing the ladder with OpenAI today, but there are still variables in how to reach the next turning point. Where will Microsoft go after reaching the Open AI contract?
Next, let’s talk about the three inflection points we saw:
① Today, information is everywhere. In the next 15-20 years, models are knowledge and will be everywhere.
② In the future, automated and autonomous actions can be everywhere.
③ People and digital technology evolve together. Sam has often said recently that it must co-evolve to achieve general intelligence (AGI). The four major elements of general intelligence are: emergence + agency + affordance + embodiment.
To sum up, we analyze the future from the fundamental trinity structure, and we can clearly see the turning point we are facing today from the past historical turning point. The essence is that the model cost will move from marginal to fixed, and one or even more great companies will be born. There is no doubt that OpenAI is leading the way.
Although it’s a bit early to talk about it, I personally think that OpenAI will definitely be bigger than Google in the future. It's just 1x, 5x or 10x bigger.
Interpretation:
As a stakeholder of Sam, Dr. Lu Qi is definitely going to make OpenAI famous, and it is completely understandable to be badmouthing Google and Microsoft. After Google was taken over by Indians, they brought the third brother's method of forming gangs and engaging in office politics into the company, which created a chaotic atmosphere. As a result, capable people could not compete with those with backing. The internal friction in the team was serious, people's mentality changed, and productivity reduce. The result is: Thousands of people at Google have been engaged in AI for many years ahead of OpenAI, and even the Transformer paper was published by Google. Now he can't defeat ChatGPT's team of less than 300 people. Bard was beaten to the ground by GPT, which is very embarrassing! What seems to be a technical gap is actually due to differences in human factors, management and vision, and the results are completely different! In comparison, is the domestic environment better or worse? I believe everyone has their own opinions.
Tencent News original text:
Next, let me talk about OpenAI from a technical perspective. How did it usher in the era of large models?
Why talk about OpenAI, not Google or Microsoft? To tell the truth, because I know that thousands of people at Microsoft also do this, but they are not as good as OpenAI. At first, Bill Gates didn't believe in OpenAI at all. He didn't believe it about 6 months ago. I was stunned when I saw the demo (product prototype) of GPT-4 4 months ago. He wrote an article saying: It’s a shock, this thing is amazing (This is too shocking, this thing is amazing). Inside Google, too, were dumbfounded.
The key technologies that OpenAI has developed along the way:
GPT-1 is the first time to use pre-training methods to achieve efficient language understanding training;
GPT-2 mainly uses transfer learning technology, which can effectively apply pre-training information in a variety of tasks and further improve language understanding capabilities;
DALL·E goes to another mode;
GPT-3 mainly focuses on generalization ability and few-shot (small sample) generalization;
GPT-3.5 instruction following and tuning are the biggest breakthroughs;
GPT-4 has begun engineering.
The Plugin in March 2023 is ecological.
Why is OpenAI’s financing structure designed this way? It is inseparable from Sam's early goals and judgment of the future. He knew that he would need to raise a lot of money, but there was a big challenge in equity design—it was easy to mix returns and control—so he had to design a structure so that it would not be restricted by any shareholders. Therefore, OpenAI investors have no control and their agreement is a debt structure. If you make 2 trillion yuan, then you will be non-profit (no longer profitable), and everything will return to society. This era requires new structures.
Interpretation:
Dr. Lu Qi talked about the technological development history of OpenAI here. I would like to share the governance structure of OpenAI with community friends and give some inspiration and reference to students who want to start a business: OpenAI has designed a very unique equity structure, which fully embodies the traditional ownership structure. Business operations with public welfare as the original intention, and the vision of returning to public welfare after successful commercialization. This design is not achieved overnight. According to the development stage of OpenAI, it can be briefly summarized into the following four steps:
In the first stage, in 2015, artificial intelligence wizard Sam Altman, Elon Musk and others announced that they would invest US$1 billion to establish OpenAI, a non-profit research organization.
In the second phase, OpenAI’s structure was adjusted in 2019 and transformed into two organizations: the for-profit organization OpenAI LP and the original non-profit organization OpenAI, Inc.
At present, when we mention OpenAI, we usually refer to a limited partnership called OpenAI LP. The organizational form is a limited partnership, which is a profit-making company.
The non-profit organization OpenAI, Inc is its general partner (GP) responsible for investment management, a limited liability company registered in Delaware, USA. In other words, OpenAI LP is controlled by OpenAI, Inc.
In phase three, starting in 2019, Microsoft has established a strategic partnership with OpenAI no less than three times. It has invested a total of US$13 billion in OpenAI and has become OpenAI's largest limited partner. The VCs who have invested in OpenAI LP since its inception have also become limited partners.
In the fourth stage, after OpenAI makes profits in the future, unlike ordinary companies that allow investors to get returns through listing, OpenAI has chosen a new equity investment agreement model: replacing the level of return on investment with the speed of return on investment.
After making profits, OpenAI will gradually return investors at its own pace:
Priority will be given to ensuring that OpenAI’s first investors receive their initial capital back;
After Microsoft’s investment is completed and OpenAI LP’s first batch of investors recover their initial investment, Microsoft is entitled to receive 75% of OpenAI LP’s profits;
After Microsoft recovered its US$13 billion investment and received US$92 billion in profits from OpenAI LP, its share of profits dropped from 75% to 49%;
After OpenAI LP generates $150 billion in profits, shares from Microsoft and other venture investors will be transferred to OpenAI LP's general partner at no cost: the nonprofit OpenAI, Inc.
As can be seen from the above four stages, OpenAI is essentially lending the company to Microsoft. The length of the loan depends on how quickly OpenAI makes money.
Directly replacing the level of return on investment with the speed of return on investment is based on the strong belief that OpenAI's vision of eventually returning to public welfare will inevitably be realized, and the full trust that the management team can steer OpenAI to success.
Under this belief, OpenAI has attracted many top AI researchers, promising absolutely competitive salaries and absolute business freedom. They resolutely left their original technology companies and academic institutions to join OpenAI, hoping to concentrate on thinking about the ultimate problems of mankind. Compared with such a novel organizational structure and Google, which is slightly stale and has a strong curry flavor, which one do you think is more competitive?
Tencent news original text
Ilya Sutskever (co-founder and chief scientist of OpenAI) firmly believes in two things.
The first is the model architecture. It has to be deep enough. As long as it reaches a certain depth, bigness is betterness. As long as there is computing power and data, the bigger the better. They started with LSTM (long short term memory), and later used Transformer when they saw Transformer.
The second thing OpenAI believes is that any paradigm, a paradigm that changes everything, always has an engine, and this engine can continue to advance and generate value.
The final core is GPT, which is the Transformer after pre-training, which can highly compress information. Ilya has a belief: If you can compress information efficiently, you must have gained knowledge, otherwise you cannot compress information. Therefore, if you compress information efficiently, you got to have some knowledge.
Ilya firmly believes that GPT already has a world model in it. Although what you are doing is predict next word (predict the next keyword), this is just an optimization method. It has already expressed the information of the world, and it can continue to improve the model capabilities, especially in sub-sections where there is currently a lot of research. Make generalizations in the conceptual space. Knowledge graphs really don’t work. If any classmate makes a knowledge map, I tell you seriously, you should not use a knowledge map. I myself have been doing knowledge graphs for more than 20 years, just don’t do that. Just pretty bad. It does not work at all. You should use Transformer. (Don't do that. It's terrible. It doesn't work at all. You should use a Transformer instead.)
What’s more important is to use reinforcement learning, coupled with human feedback, to align with human values. Because GPT has been in use for more than 4 years and the knowledge has been encapsulated in it, it was really impossible and difficult to use it in the past.
It is difficult to make a large model, and the biggest difficulty is infra (infrastructure). When I was at Microsoft, each of our servers did not use network cards, but all had FPGAs. The bandwidth speed of network IO is unlimited bandwidth technology (Infiniband), and memory is directly accessed between servers. Why? Because Transformer is a density model, it is not only a matter of computing power, but also has extremely high bandwidth requirements. Just think that GPT-4 requires 24,000 to 25,000 cards for training. Just imagine how many people in the world can do this kind of system.
The above is the engine. How did it reach the turning point? ChatGPT can reach 100 million active users in two months for the first time in history. There is no way to stop it. Why?
① It encapsulates all the knowledge in the world.
② It has strong enough learning and reasoning abilities. GPT-3 ability is between high school students and college students. GPT-4 is not only for entering Stanford, but also for people ranked very high at Stanford.
③ Its field is wide enough, the knowledge is deep enough, and it is easy to use. The biggest breakthrough of natural language is its ease of use. The scalability is also good enough. Of course it is still very expensive, as it is such a big project to require more than 20,000 cards and several months of training. But it’s not that outrageously expensive – Google can do it, Microsoft can do it, several large Chinese companies can do it, and startups can do it by raising money.
Taken together, the paradigm tipping point has been reached. The turning point has arrived.
Interpretation:
Dr. Lu Qi is here to give a reminder to those working on AI technology: Give up the knowledge graph and embrace the Transformer! As an expert who has been working on knowledge graphs for 20 years, I believe in his professional judgment when he comes to this conclusion. Maybe Transformer is really like what Google said in the paper [Attention is all we need]. Looking back on the success of OpenAI, I think the key is that they can keep pace with the times, have no idiosyncratic views, use whatever works well, and then dare to invest a lot of resources in trial and error! Successful results are directly related to its lofty vision, innovative organizational structure, the elite talents it attracts, and a sufficiently open and inclusive research and engineering atmosphere!
Tencent News original text:
What OpenAI will do in the next 2-3 years is to make the model sparser, with more modalities, more token spaces, more model stability, more latent spaces (such as Latent Space alignment), and more Compute, more infrastructure tools. The schedule is basically full in 2-3 years. In other words, we roughly know what is needed to continue to make this engine bigger.
However, the main reason for this flywheel to start is the large inflow of capital. From January to March 2023 in the United States, there is no way to stop it, all the money has gone in, and every month is growing compared to the last month. The same is basically true in China. The business model and profit model have preliminary scale, the development of infrastructure, platform applications, and ecology is accelerating, and start-ups and large enterprises are entering.
Of course, social security, supervision, and a lot of issues—these are the biggest headaches for OpenAI right now—Sam spent a lot of energy in the United States to get society to recognize this technology. What the core of OpenAI is doing now is to slow down the advancement speed. Every time a new version is promoted, there will be enough time for users to give them enough feedback, find potential risk points, and have enough time to make up for it. But taken together, the prototype of a growth flywheel has basically taken shape.
Interpretation:
ChatGPT's current initiative to slow down the development speed of new models seems a bit like Bolt's intention to slow down and look back at his opponent when he was running 100 meters or Wang Meng was speed skating while being absolutely ahead.
But in fact, this represents a kind of absolute confidence: OpenAI once publicly stated that if any team in the world proves that it has the ability to achieve close to 1/10 of the AGI technical capabilities, OpenAI will stop its own research and development to fully support its opponent! This is the flywheel effect mentioned by Dr. Lu Qi: ChatGPT is now used by hundreds of millions of active users every day. The more people use AI, the better it will evolve and its capabilities will be stronger! The stronger the ability, the more people will use it, thus forming a positive spiral flywheel effect. How does this allow opponents to compete? I'm afraid I can't even see the taillights!
Tencent News original text:
How will the AI model world develop in the future? The first is that there will be more big models coming out. More complete modalities and more complete knowledge of the world are here. You have a lot of knowledge and more modalities, and your learning ability, generalization ability and generalization mechanism will definitely be enhanced.
Additionally, there will be more alignment work to do. What will OpenAI focus on now? Today's alignment is basically done. Some people can accept it, but you also offend many people, and many people criticize GPT every day. They want to achieve an alignment that is broad enough and hope to have a result like the United States Charter. Although ChatGPT is not recognized by everyone, it is stable and comprehensive enough that most people can accept it. This is an alignment project. Whether it is natural language, code, mathematical formulas, or forms, there is a lot of alignment work to be done.
There are more modal alignments. Let’s talk about the modality of human scale first. It mainly describes people and focuses on human language. Its modalities are currently language and graphics, and more modalities will be added in the future. This is the big model level.
Interpretation:
The main pressure on ChatGPT now comes from people's panic about the quasi-AGI capabilities displayed by GPT-4. They are worried that it may cause many people to lose their jobs, and that AI hallucinations (that is, GPT's serious nonsense) may have negative effects on the existing social order. Influence. These additions may bring a lot of trouble to the development of OpenAI. Therefore, it is a prudent and wise decision to slow down, or even stop temporarily, do a good job of aligning the model, and let the surrounding soft environment improve before starting.
Tencent News original text:
More models are built on top of larger models. I judge that there are mainly two types of models and their combinations.
The first is the model of things. Every type of human need has a domain/working model, including structural model, process model, demand model and task model, especially memory and prior.
Second, the human model includes cognitive/task models, which are individual, including professional models, cognitive models, motor models, and human memory priors. People are basically a combination of these types of models, whether they are lawyers or doctors, there will be a large number of models moving forward in many fields.
There is an essential difference between the model of people and the model of things. This is what I have gained personally in the past 1-2 months.
There are many models today, such as digital twins, that are difficult to use. Because the physical world is always changing, this model is rigid and unchanging, so it cannot be used. Especially the models built using knowledge graphs, which I have been working on for decades, are super difficult to calculate and the function structure is a mess. Therefore, the human model has advantages and is highly professional, but it also has major shortcomings.
The model represented by ChatGPT and the human model complement each other and can be integrated in the long term. The future we see is an ecology of more models, new fields, new majors, new structures, new scenarios, and new adaptability, forming a closed loop to continuously enhance cognitive and reasoning capabilities. Of course, in the end, it still needs what is called grounding, which requires grounding in perception, and the ability to access actions to form real intelligence.
In a sense, 20-30 years from now, this model world will have many similarities with the biological world. I think big models are like genes, which come in different types and then evolve. We can currently see the future core technology model world, which uses this method to drive forward.
We basically have a structural understanding of the paradigm of this era. So next, how do we embrace this era?
Interpretation:
What are the main models of people and things mentioned here? The so-called human model is to standardize information such as people's relevant behavioral needs into data and feed it to AI to build a model. When the data is sufficient and accurate enough, then the human model can basically map the thoughts and needs of real people in reality. For example: the recently popular AI model of Steve Jobs. After the author cleaned, annotated and fed information about Steve Jobs' life story to ChatGPT and made Finetun, the AI can be made to look like Steve Jobs and use his tone and preferences to communicate with people. The models are such as the popular Mid Journey, Stable Diffusion and other tools that specifically help us solve the specific needs of painting. Human model + event model + AI proactively sets tasks based on human needs and puts them into action to help us solve problems and create value. This is an important direction for the future development of AGI!
Tencent News original text:
Personally, I've been reading a lot every day for the past 10 months, but recently I just can't stand it. I really can’t keep up. The pace of development is very, very fast. Recently we started publishing the "Big Model Daily", but I really can't do it anymore. I really can't keep up with the papers and the codes - there are just too many.
The world is changing. I once said that I felt this way in 1995-1996, but it is even stronger now than it was then. Why? The cost of models has shifted from marginal to fixed. Knowledge creation is the acquisition of models and knowledge, and its structure has evolved.
Production capital is comprehensively improved from two levels. First, all brainwork can reduce costs and increase production capacity. We currently have a basic assumption that the cost of code farmers will decrease, but the demand for code farmers will increase significantly, and code farmers do not need to worry. Because the demand for software will increase significantly, even if this thing is cheap, just buy it. Software can always solve more problems, but some industries may not. This is a widespread increase in productive capital.
Second, production capital has been deeply improved. The nature of productive capital in some industries is model-driven. For example, medical care is a model industry, a good doctor is a good model, and a good nurse is a good model. The nature of the medical industry is strongly model-driven. Now that the models have improved, so has the science. In the core gaming industry, our production capacity will be substantially and deeply improved. The speed of industrial development will accelerate, because the speed of scientific development has accelerated, the speed of development has accelerated, and the heartbeat of every industry will accelerate. Therefore, we believe the next inflection point will be accelerated. Use large models to build robots, automation, and self-driving, and you can't stop them.
It will have profound and systemic consequences for everyone. Our assumption is that everyone will soon have a co-pilot, not just one, but maybe five, six. Some co-pilots are strong enough to become full-pilots, and they can automatically help you with things. Longer term, we each have a team of drivers serving us. The future human organization is a real person, plus his co-pilot and real pilot working together.
There will undoubtedly be structural impacts on every industry and systemic restructuring. Here's a simple formula. How many hours a day do people who use their brains earn on average today? Excluding ChatGPT, the current average price is about 15 US dollars/hour. In 3 years, it may be less than 1 US dollar, and in 5 years, it may be tens of cents. Then multiply by how many there are. Reduce costs or increase efficiency so that programmers can become super programmers and doctors can become super doctors.
You can calculate it according to this formula. If you were a hedge fund on Wall Street, you could short a whole bunch of industries.
To give a simple example, the average price of a lawyer in the United States is US$1,500 per hour. I have seen this kind of information on the Internet every day - if you want a divorce, don't find a divorce lawyer, ChatGPT divorce is very cheap! (Everyone laughs)
Developers, designers, coders, and researchers are all the same. Some have more needs, and some have lower costs. Especially the core industries, science, education, and medical care, these are the three industries that OpenAI has been paying most attention to for a long time, and they are also the most fundamental to the entire society.
Especially medical. In China, demand far exceeds supply. Moreover, China is a market economy driven by a big government, and the government can play a larger role because the government can bear the fixed costs.
The most important thing is education. If you are in college, your first worry is, how to take the exam? There is no way to take the test. As soon as he asked ChatGPT, he knew everything. More importantly, how will we define a good college student in the future? Suppose there is a college student who doesn’t understand anything, neither physics nor chemistry, but he knows how to ask ChatGPT, is he a good college student? Opportunities and challenges coexist.
To sum up, this entire era is proceeding at a high speed, getting faster and faster. It is structurally determined. Unstoppable.
Interpretation:
With the explosive evolution and development of AI, the first thing you do when you open your eyes every day is to open the news and see what new technologies, new tools, and new models have been released by the AI companies, celebrities, and KOLs that you follow... and a few of them accidentally Hours passed and I only read half of the new messages. Then, I opened the website, tested these new products and technologies, and some large open source models had to be downloaded and deployed by myself, and even finetune debugged. The whole day was spent in such a tense and busy state. Then, I sighed, this industry is developing too fast, I really can't keep up with the pace, and I fall asleep like a dead pig. Tomorrow will be almost another cycle...
Tencent News original text:
Now, I will give you a structured thinking framework. In a sense you can kind of plug in, knowing I'm here, how I think about today's opportunities.
This picture is the entire entrepreneurial innovation driven by human technology, and the opportunities for all things lie in this picture.
First of all, the bottom layer is digital technology, because digital is an extension of human beings. The foundation of digitalization includes a platform and a development foundation, including open source code, open source design, and open source data; the platform includes front-end, back-end, etc. There are tons of opportunities here.
Second, use digital capabilities to solve people’s needs. We put digital applications entirely on this table.
1) On the C side, all people are divided into groups. Each group has 24 hours. What does he spend his time doing? There are communication, social networking, content, game consumption, travel, fitness... There is a special type of people on the C-side. These people are coders, designers, and researchers who change the world. They create the future. A company as big as Microsoft is based on a simple concept: at Microsoft we just want to write more software and help others write more software, because writing software is the future.
2) On the B side, the needs of enterprises are the same, to reduce costs and increase efficiency. It needs production, supply chain, sales, customer service... After meeting these requirements, there are six visible digital experience structures: two-dimensional is enough to give you information; three-dimensional interactive experience is enough for you in games, meta The universe; the abstract relationship between people, including trust relationships, Web 3; people driving autonomously, robots, etc. in the physical world; the inner part of people is implanted into it with carbon machines. Today it is a brain-computer interface, and in the future there will be More, you can use silicon base in the future; finally, I will give you a model.
Interpretation:
The development of AI gives us all the opportunity to become people who change the world. There are so many open source projects. As long as we are willing to learn and ask ChatGPT if we don’t understand, we will achieve something if we persist.
So there are basically three types of startups: digital foundation, using digital to solve people's needs, and changing the physical world. With this large framework, we can look at the situation systematically: Where am I? If I were in this position, what would I need to focus on?
Let’s talk about the foundation of digitalization first. It has a stable structure. No matter how it develops, the structure will always be like this. In the past 30 years or so, I have come across most systems to some extent, and this structure is indeed quite stable.
The core is the front-end and the back-end - the front-end is a complete and extensible experience, and the back-end is a complete and extensible capability. There is the device side, such as computers, mobile phones, glasses, cars, etc., and the device side is filled with chips, modules, etc. on the operating system. Trillion dollar companies are on this tier.
The second is the container of experience, a two-dimensional container, a three-dimensional container, and an internally embedded container.
On top of the container, everyone who writes code knows about the canvas. The canvas can be a document, a chat, a code, a space, a world, a digital person, a protein in a carbon base, etc. This is the front end.
The same goes for the back end, the underlying equipment, servers, switches, data centers, etc. are also chips, modules, and operating systems.
The middle layer is very important, network data stack, distributed system, blockchain, etc.
The top is the cloud, which is the supply of capabilities. Capacity supply is like a natural water source, which means computing power, storage and communication capabilities. In today's model era, opening is a model.
Here are the digital basics. Symbolic computation, or so-called deep learning, floating-point computation of superposition vectors, silicon-based, carbon-based.
If you are an entrepreneur here, where are the opportunities?
① First of all, transfer information. There is still a lot that can be done in this era.
If you are making models, I now judge that you have to redo everything. Big models first. Many devices also need to be redone, you need to support large models, and containers need to be redone. There are opportunities for these. Cloud, intermediate infrastructure, underlying hardware, including the core foundation of digital development, especially the open source system, there are truly a lot of opportunities here.
Interpretation:
Studying large open source models may seem like a waste of reinventing the wheel for a long time to come, but this step is also the basis for future competition barriers. If all entrepreneurs call ChatGPT’s API and develop plugins, how can the startup team defeat those traditional Internet giants? Even if you come up with some new tricks, and the other party is richer than you, and has stronger technical, operational, and market capabilities than you, you may end up just helping others with trial and error proofing for free. Therefore, friends who are engaged in AI technology should not be anxious to make money. The accumulation of underlying technologies for large models really requires being able to withstand loneliness. Over time and accumulation, isn’t this how OpenAI has come? The entrepreneurial atmosphere in China is not bad compared to overseas. As long as you keep your mind down and accumulate in a down-to-earth manner, you will eventually make a difference!
③ The third generation system has begun to develop robots, automation, and autonomous systems. Son is all in today. This can also be done with large models. Musk also sees this opportunity. They are all opportunities that startup companies can seize at the next turning point of the third generation.
④ At the same time, parallel, I call it the "third generation++ system", which is carbon-based biological computing. This type of company has a lot of quantum computing and there are many opportunities. The Metaverse and Web 3 are a bit cold today, but from a historical perspective, it is only a matter of time, because these technologies can truly bring future human value.
So if it is this entrepreneurial project, the basic layer opportunity is here. This is the best business. Why? This era is very similar to the gold rush era. If you went gold mining in California at that time, a lot of people would die, but the people selling spoons and shovels would always make money. The so-called shove and pick business.
Interpretation:
Big models are platform opportunities.
According to our judgment in a few days, a model-first platform will be larger than an information-first platform. The platform has the following characteristics:
① It is available out of the box;
② There must be a business model that is simple and good enough. The platform is a platform where developers can live and earn enough money to support themselves. Otherwise, it is not called a platform;
③ He has his own killer application. ChatGPT itself is a killer application. Today’s platform companies are you in the Apple ecosystem. No matter how good you are, as long as you become a big Apple, they will confiscate you because it needs to use your underlying things, so you are the platform. The platform generally has its anchor point and strong support, and there are many long-term opportunities for OpenAI equipment - it is possible that this is the first $10 trillion company in history.
This is a fierce battle for competitive platforms, and it will be a very large company in the future. Competition in this field is extremely fierce. The price is too big (the price is too big), it would be a pity to miss it. No matter what, you have to give it a try.
Innovation, especially the implementation of start-up companies, is always a combination of technology push and demand pull. In the process of implementation, the most important thing is to understand the needs and master the methods to meet the needs. In the long run, it must be technology-driven, but when it comes to implementation, dismantling, analyzing, sorting out and controlling the needs is everything.
There is a secret that everyone knows today - OpenAI uses GPT-4 to make GPT-5, and every coder is a coder who can amplify his capabilities. Its scale effect is different, the Matthew effect is different, the barriers and competition pattern are different, the intellectual property results are different, and the internationalization pattern is also different. China clearly has a chance.
Interpretation:
The existence of the wall between the East and the West is both a challenge and an opportunity for domestic AI entrepreneurs. The challenge is how to avoid policy risks and develop steadily. The opportunity is that potential opponents at home and abroad are blocked by walls, and it is impossible to spy on the opponent's lineup in the short term. Being good at using AI tools is the basis of all competitiveness, and everyone must remember this!
The inner structure of a startup is a combination of people and things. A person is the founder/founding team at the beginning; he has original intention, internal driving force, and external driving force; he can think independently and judge the future; he can be action-oriented and solve problems; he can be demand-oriented and find value; and finally through Communicate to obtain resources. Next is product-market matching, which involves developing technology, developing products, and delivering products. The business model is to receive money, grow more, reach more customers, raise more money, and reach future value. Organizationally, through system construction, we develop future-oriented talents, organizational structure, cultural values, etc. It all adds up to a company.
Our advice to every student is not to act rashly, but to think first.
① Don’t be exaggerated and don’t take advantage of others. Personally, I am most opposed to the hype. You need to make a big model and figure out exactly what to do. What the big model really is is essentially related to the dimension or dimensions of your entrepreneurial direction. Taking advantage of others is the worst behavior and will waste opportunities.
② Be diligent in learning at this stage. The new paradigm has multiple dimensions and is quite complex. You should read the papers you should read, especially now that the development is too fast and the uncertainty is very high. My judgments all have a certain grayscale. I can’t say I see them clearly, but I roughly see this result. Take the time to learn, I highly recommend it.
③ After thinking clearly, be action-oriented and take decisive and planned action. If this change has a structural impact on your industry, if you don’t advance, you will retreat. If you don't move forward, there will be no retreat, and you won't be able to hold on to your position today. You can only take action if your industry is directly affected.
Interpretation:
Dr. Lu Qi’s views are all good advice. As a serial entrepreneur and angel investor, I would like to support them with both hands! Everyone must remember to always check themselves
Next I want to talk about several dimensions - each company is a combination of capabilities.
① In terms of product development capabilities, if your company is mainly software-based, there is no doubt that it will have an impact on you, and the long-term impact will be huge. Especially if you are doing C-side, the design of user experience must have an impact, and you must seriously consider what to do in the future today.
② If your company develops its own technology and has local and indirect impacts in the short term, it can help you think about the design of the technology. The research and development of long-term core technologies will also be affected. Today's chip design requires a large number of tools. In the future, large models will definitely affect chip development. Similarly, proteins are protein structure designs. No matter what you do, it affects future technology. There is no direct impact in the short term, but there may be significant impact in the long term.
③ Ability to meet demand. Meeting demand basically requires reaching users, and the supply chain or operation and maintenance will definitely be affected. GPT can help you with the operation and maintenance of software, but not the supply chain of hardware. In the long term, there are opportunities for change because the upstream and downstream structures will change. You have to judge whether your structure in this industry will change.
④ It can help you think and iterate on business value exploration, reaching users, and financing.
Finally it’s about talent and organization.
① Let’s talk about the founder first. Today’s founders have strong technical skills and seem to be very good and important, but they really won’t be important in the future. Technical ChatGPT can help you do it in the future. As a founder, what becomes more and more important and valuable is your willingness and hard work. Willingness is a unique judgment and belief in the future, persistence and strong tenacity. This is an increasingly important core quality for future founders.
② For start-up teams, tools can help explore directions, accelerate the iteration of ideas, products, and even resource acquisition.
③ To cultivate future talents, on the one hand, we need to learn tools, think and explore opportunities, and cultivate our own prompt engineers at appropriate times over the long term.
④ Finally, when it comes to building organizational culture, we need to think more deeply, prepare early, and seize the opportunities of the times. Especially considering that many functions already have co-pilots, whether they are writing code or doing design, how can they be coordinated?
We face opportunities in this era. It is both an opportunity and a challenge. We recommend you think through this opportunity.
Summarize:
The tide of the AI era is unstoppable. Everyone must not be short-sighted. AI is a generational development opportunity that is far superior to the industrial revolution and information revolution. We must aim high and not rush for short-term arbitrage. We must study technology, products and content carefully, build walls high, accumulate food widely, and accumulate more and more!
Come on, let’s focus together.
Interpretation of Dr. Lu Qi’s “Large Model World View”_NaiveCode’s Blog-CSDN Blog