The era of AI - a new productivity revolution

✏The author presses:

I am rather stupid. It was around March 9 that I suddenly realized that a new era of AI had arrived.

The full text is 7339 words, and it takes about 10 minutes to read

>>Preface

I came into contact with the knowledge of artificial intelligence and neural networks when I read the book "Artificial Intelligence Technology in Game Programming" in my junior year. Write a small unsupervised and supervised neural network model to implement an automatic obstacle avoidance minesweeper game and a gesture recognition game function. I was impressed by the very simple introduction to neural networks:

Cut from article P161

In short, a data structure (class) is used to simulate the structure of neurons in the brain. Dendrites correspond to input parameters, axon terminals correspond to output parameters, and nerve cell nuclei correspond to a function called an excitation function. According to different input parameters and The combination of corresponding weights sets a functional transformation relationship to determine the final output of the final neuron nucleus.

Above is the simple structure of a single digitized artificial neuron, corresponding to the human brain, where the dendrites of a single neuron nucleus are connected to more than 10,000 other neurons. In other words, the above-mentioned single class has more than 10,000 input parameters, and then how many instances of this class should be instantiated in the program? The book says it is 10 billion. These examples refer to each other, and the corresponding number of connections is 100 trillion. This is a huge network. To be honest, it is difficult for people to figure out how human characteristics such as consciousness, emotion, and learning are generated in such a large network, but there are some hard-headed people who think: whatever, I First use a computer to build a large network to see what the effect is! Then, it is still serious to come up with the model, the algorithm, the feedback, and the tuning.

Artificial intelligence in the early years was called artificial mental retardation. There is a picture that highly summarizes the working principle of artificial intelligence:

Although this is a joking picture, it explains to a certain extent the actual working principle of artificial intelligence. Now it is difficult to explain clearly with human cognition, and a few years ago, everyone still had doubts about whether artificial intelligence could be realized. attitude.

>>Bright future

Now those hard-headed pioneers have really figured it out! GPT-3 has 174 billion parameters, while GPT-4 has about 100 trillion parameters, which is equivalent to the number of human neuron connections! In other words, they really made a personal artificial brain.

In the early test paper "Sparks of Artificial General Intelligence: Early experiments with GPT-4" released by Microsoft Research Institute on GPT4 in March this year, the author talked about why GPT4 has such excellent intelligence performance despite its excellent performance. Still can't give a reasonable explanation.

"So what actually happened?

Our work on GPT-4 is entirely phenomenological: we focus on the surprising things that GPT-4 can do, but we do not address the fundamental questions of why and how such remarkable intelligence is achieved. How does it reason, plan and create? Why does it exhibit such pervasive and flexible intelligence when its core is just simple algorithmic components—gradient descent and large-scale transformers combined with extremely large amounts of data? These questions are part of the mystery and fascination of LLM, which challenge our understanding of learning and cognition, fuel our curiosity, and drive deeper research...  

Overall, elucidating the nature and mechanisms of AI systems such as GPT-4 is a daunting challenge that has suddenly become important and urgent. "

So the current results are more like an act of miracles. Even so, after I personally realized the great significance of this matter, I was still so excited that I stayed up all night and thought about it all night.

Why is this matter so important? I think there are two main points:

• This marks the gradual maturity of AI technology, making people realize that the road of super large neural network can go well

• In the current global environment full of haze, it lights up hope for people

Regarding the first point, if you haven’t had a deep understanding, you can read an article about artificial intelligence in February 2015, "How will super artificial intelligence lead to human extinction or immortality in our lifetime?" ", this article is equivalent to a summary of Ray Kurzweil's "The Singularity Is Near" published in 2005. At that time, I read it as a science fiction novel. Now I think the author has a foresight, and some predictions in the article are quite accurate. I suggest you read it. Referring to the figure below, the emergence of ChatGPT and GPT4.0 marks the imminent intersection of the intelligence curve in the figure below.

Although the number of neuron connections in the human brain is roughly equal to or even slightly more than the number of parameters in GPT4, it should be remembered that the human brain has only been developed by about 5%, and once GPT4 is fully fired, it will be the oligopoly in "Super Body". sister. 

ChatGPT is currently an all-round and all-knowing AI, and its outstanding performance has undoubtedly greatly enhanced people's confidence. A few years ago, the AI ​​myth, which was considered to be a fantasy, suddenly seemed not so far away, providing a guiding direction for the development of human science and technology. The dissipation of the epidemic did not bring the expected prosperity. On the contrary, the situation in the world today is quite hopeless: the economic situation is sluggish, and the bulk of social wealth flows to the major monopoly giants. The strong get stronger, winners take all, small and medium-sized It is difficult for enterprises to survive, and the social class is more solidified; the birth crisis, the demographic dividend is gradually disappearing; the lack of major progress in basic science has led to the continuous involution of various technological applications in the industry, and entrepreneurs can't figure out any new tricks; the haze of war Hovering over the world, the hard decoupling of China and the United States has begun.

Although the three years of the epidemic have been bitter, everyone still has some hope in their hearts. They all feel that as long as the epidemic dissipates, they will return to the past when the market was prosperous and the economy rose rapidly. However, the reality is that after the fig leaf of the epidemic has been ripped off, the world With his original skinny appearance exposed, we knew that the good old days in the past might really never come back together.

Back to me personally, I am thirty-four years old, and the company has only been independent for two years. In theory, it is at the peak of my life and career, but under such a big situation, my heart is inevitably beating, and the little flame of hope in my heart is swaying with the wind. . People are not afraid of being poor when they are alive, but they are afraid that they will no longer have hope in their hearts.

There is no doubt that the gradual maturity of AI artificial intelligence and ChatGPT's out of the circle really cleared the clouds and seen the sky, and re-lighted the hope of human civilization.

In the final analysis, the advancement of human civilization comes from the development of productive forces, which are the fundamental driving force of social development. The current level of productivity is no longer suitable for the development needs of society.

Taking the manufacturing enterprises that the author is more familiar with as an example, is the demand of enterprises not strong? Not necessarily, how many manufacturing enterprises are eager to place an order today, the equipment will be in place tomorrow, the system will be online, and the production capacity will be climbed quickly, the payment to Party B will be settled quickly, and the product will be produced quickly! But it’s unrealistic, because procurement has a process, infrastructure requires a process, equipment needs to be designed, assembled and debugged, production management and other software systems need to be researched, customized, and secondly opened, and none of them can be rushed. Although humans have developed various platform tools to empower manufacturing, the core engine that drives these tools is still humans.

A reality we have to admit is that people are the bottleneck that restricts the further improvement of social productivity.

The development of productivity needs new engines to drive. In the past, it was steam engines, internal combustion engines, and computers, and in the future it will be artificial intelligence engines. This brand-new engine for the development of human society will rapidly affect all walks of life and various positions of human beings in a short period of time:

• Various positions relying on a single skill: art, programming, mechanics, circuits, translation, marketing, operations, sales...

• Various positions based on past experience: middle and high-level management positions in various enterprises, various business consultants, doctors...

• Various positions according to social rules: lawyers, teachers, financial consultants...

I'm sure you've seen a lot of answers like this on the Internet.

After communicating with some people around me, I saw that many people only use ChatGPT as an after-dinner conversation, and they look like they have nothing to do with themselves. I am both anxious and excited:

What is anxious is that such a big opportunity is really fleeting. According to the author's judgment:

• In 2023, various common software applications will integrate AI assistants in different forms to complete the first step of intelligent upgrading, that is, shallow AI assistance. This is for everyone to see.

• Within two or three years, that is, by the end of 2025, most software applications will complete deep AI integration, but human participation and adjustments are still required.

• By 2030, AI will be able to complete work fully automatically without human intervention in most fields.

This is a bit abstract here, and everyone will question how I came to these time points of "within half a year" and "two or three years". Later, the author will talk about digitization in the field of production and manufacturing, combined with the Hanma low-code platform as a landing tool Let's talk about it in detail.

What is exciting is that many people have not yet realized the great significance of AI and the upcoming changes, which shows that we still have a chance. This is a poor cognition.

To realize the difference in cognition, you need to have landing scenarios, action plans, tools and strong execution. Let me take the Hanma Industrial Low-Code Platform as an example to describe the application process of AI in the digital transformation of manufacturing industry.

Hanma Technology’s low-code platform has actually done a lot of work before it is integrated into AI technology. The most important functional content is to dismember the production process of enterprise application software programs and form a set of standard steps, as shown in the figure below :

Among them, modeling, integration, and deployment have formed a zero-code solution, and the parts that require code development are mainly concentrated in the development process.

In the development process, the web pages and APP pages developed by the front-end have been low-coded. Only when you need to customize css and some front-end behaviors do you need to use javascript for customized development, and the custom logic of the back-end needs to use groovy scripts. Realization, this is one of the few places that still need to write code directly in the process of developing software based on the Hanma low-code platform.

Brothers who are not familiar with the low-code platform of Hanma can take a brief look at the overall capabilities of the existing low-code platform in 1 minute:

What is Hanma doing? video

In subsequent product iterations, Hanma Technology will gradually realize an artificial intelligence application generation platform that can directly generate enterprise applications according to user needs research documents or research interview recordings in the following steps :

【Future Yidao Cloud Product Planning】

How many steps does it take to put an elephant in a refrigerator?

1. Continuous integration of AI capabilities

• In the version on April 2, Hanma Technology's industrial low-code platform will soon integrate ChatGPT to improve the efficiency of groovy script development in the background logic part; [welcome to click the link at the end of the article to try it out ]

[One knife cloud AI assistant test page]

• In the future, we will apply AI capabilities to the automatic production of front-end scripts and css styles;

• We will support automatic generation of all models, fields, constraints and services based on detailed design documents;

• We will integrate AI drawing capabilities similar to Midjourney for making APP LOGO, icons, welcome pictures and other related materials;

• We will apply AI to the automatic generation of low-code web pages and APP pages from prototypes;

• We will integrate relevant industry knowledge intelligent assistants in the existing comprehensive industry-oriented solutions such as lithium battery solutions, SMT solutions, and injection molding solutions;

At this stage, humans will still participate in the development process of the project as the main program, and AI will act as an auxiliary programmer.

2. Automatic generation from detailed design documents to application skeleton

If the above is to use AI to help improve certain function points on the low-code development platform. Then in the second stage, we will pursue connecting all the functional points of the entire platform with a detailed design document.

By scanning and learning the detailed design documents, AI can call the AI ​​function points buried in various places on the platform in the previous stage to automatically produce the program skeleton and main logic.

At this stage, AI will replace humans as the main program, and humans will become assistants, mainly making local adjustments and supplements to the low-code program skeleton generated by AI.

3. Automatic generation from requirement documents to detailed documents

However, the writing of detailed design documents still has relatively high requirements for the professionalism of personnel. Therefore, in the third stage, AI will be used to complete the automatic production from requirements research documents to detailed design documents. This step will take place outside the low-code platform. It is the transformation of a loose, unorganized and even logically imperfect language fragment into another complete and comprehensive structural document, that is, from the requirement document or requirement research. Conversion of audio recordings into detailed design documents.

The AI ​​training method here can be routine. For example, some basic information such as the customer’s background information, process technology, product type, etc. is fed to the AI, which is used as the AI ​​​​training material in advance to fine-tune the basic model.

This step is to apply AI to the design of the program, beyond the stage of program implementation. If the above two stages will greatly reduce the workload of programmers, then this stage will greatly reduce the workload of architects/development representatives/product managers.

4. The whole process is connected in series

With the completion of the above stages and the increase in AI's familiarity with specific scenarios, combined with some existing industry templates to pre-train the AI ​​model, we can finally get an AI consultant who is very familiar with a certain industry.

Integrating such AI in the low-code development platform: On the one hand, it understands the business very well, on the other hand, it understands the use of the platform, and finally can truly achieve the effect of fully automatic generation of AI applications from requirements to finished products.

The effect diagram after realization

The time points mentioned in the above figure are the time points of each stage roughly estimated by our team based on the resource allocation of the existing low-code R&D team:

1. Most of the functions in the first step are expected to be launched in half a year

2. The effect of the second step, due to a large amount of model training and fine-tuning, is conservatively estimated to take about 1 to 2 years

3. The third and fourth steps can be carried out simultaneously, and the same progress depends on the quality and quantity of pre-training materials. The conservative implementation time should take about 3 to 5 years

4. As for the last step, the automatic generation of the whole process is estimated to take about 4 to 6 years.

I am also from a technical background, and I understand that developers have reservations in the evaluation of working hours, so there should still be a lot of water in it. In addition, foreign AI application manufacturers, such as DeBuild, actually demonstrated some of the above functions two years ago. Those who are interested can take a look at: [https://www.youtube.com/watch? v=WhPgZFsPLeE&ab_channel=SharifShameem] (Youtube link)

For a long time, the customization of 2B software has been a big problem that plagued the industry. There is even a saying that the success rate of 2B applications is only 30%. why? Because 2B software delivery is inherently a very complicated matter:

I think the core factor here is people. As long as there are too many people involved in one thing, the lower the efficiency of information transmission, the greater the loss in the process of information transmission, the higher the cost of information transmission, and the process is full of misunderstandings. Misunderstandings, speculations, etc., the final information is often distorted.

If humans look at it as an IO device, the IO efficiency is too low. The input efficiency through the five senses is: vision: 10MB/s, hearing: 1MB/s, smell: hundreds of bits/s, touch: dozens bit/s, taste: dozens of bit/s. The efficiency of human beings to output information through language and text is lower: language, the average person’s speech rate is 2 to 4 words per second, and it is about 4 to 12 bytes per second in terms of bytes; the fastest typing record in the world Calculated at 212 words/minute, it is almost 3 to 4 words per second, that is, the efficiency of a few bytes to a dozen bytes per second, which does not take into account that human beings are emotional. Multi-threaded, it is easy to misunderstand, and there is a situation where a lot of information is overthrown and restarted. [Data from ChatGPT] Although we have a very sophisticated brain, but limited by very inefficient IO, it is difficult for us to convey our large amount of thoughts.

Therefore, accelerating the efficiency of information processing by introducing a low-code platform + AI is the key to improving productivity. Especially in the link after business research, it is the main direction of the current stage to minimize the participation of personnel, improve the efficiency of the information transmission process, and save more energy to understand human needs.

In many cases, when it is necessary to collect requirements, human beings are often unable to explain the requirements for various reasons. At this time, the combination of AI knowledge reserves and historical application templates can be used to help people quickly build prototype applications. Then accumulate requirements in the process of use, and then fine-tune the AI ​​model to complete the digital construction in a dynamic manner.

>> How ordinary people can meet this wave

First of all, each of us must fully realize that a brand new era of AI has arrived!

In the process of writing this article, my mood has changed from the excitement at the beginning to now a little lost, because I found that most of the artificial intelligence-based application ideas generated during the night I thought about all night have already been realized by corresponding foreign manufacturers up. Here I list some:

1. Midjourney AI image and illustration generation tool [art drawing]

2. Bing New Bing 【Search Engine】

3. Jasper AI text content creation tool [text creation]

4. Copy.ai artificial intelligence marketing copywriting and content creation tool [Marketing, Content Marketing]

5. Copysmith enterprise-level and e-commerce copywriting [e-commerce copywriting]

6. ADEPT Artificial Intelligence RPA【AI+RPA】

7. SurferSEO AI SEO Outline and Content Optimization Writing Tool 【SEO】

8. Debuild AI Web program automatic generation tool [software generation]

9. Tabnine AI code auto-completion programming assistant [software programming]

... ...

I've tried a few of the tools here, and the results are really amazing.

I once thought about whether we should avoid talking about our product development route, but in this wave, our domestic manufacturers have obviously lagged behind. At this time, there is no need for everyone to hide and hide. We must work together. Make good use of the window period of industry and policy barriers that still exist, ALL IN AI, and strive to catch up to make up for the lagging distance.

Many people are worried about being replaced by AI. First of all, I want to say that AI does not replace human beings. AI is just a tool. It is designed to assist humans. It is humans themselves who really want to replace humans. I think the so-called replacement of human beings by AI is a typical capitalist thinking: it is a wrong idea to think that people have no residual value and can be kicked away, because human beings can also evolve continuously, and human beings will also evolve to adapt to this new world. era. Ordinary people in the new era of AI must take the initiative to learn to use AI tools. Lifelong learning is a precious quality that human beings cannot lose no matter what age they are in.

In addition to the types of occupations mentioned above that will be greatly affected, I think the following types of occupations and organizations will still be preserved:

1. Humanistic caring careers or positions: pension, psychological counseling, babysitter, pet shop, etc.

2. Public welfare undertakings or positions: volunteers, public welfare organizations, etc.

3. Networked organizations or positions: chambers of commerce, associations, key account sales, etc.

4. Some service-oriented organizations: health care centers, talent training, etc.

5. innovative creative jobs

In general, more people will go to the tertiary industry service industry. At the same time, many new occupations and positions will be born.

On the technical side, there will be many AI model trainers. AI model trainers are used to collect industry characteristics and knowledge corpus to produce pre-trained basic models with industry characteristics, and at the same time organize follow-up professional manual labeling and fine-tuning work.

On the user side, there will be a large number of prompt engineers called Prompt Engineers. After all, some people may lack basic expression skills, let alone how to scientifically communicate with AI to obtain ideal results. . Prompt Engineer will act as an early industry-oriented AI evangelist to summarize a set of prompt words and paradigms with industry characteristics, and provide enterprises with AI talent training services in a form similar to today's consultants, helping enterprises to establish themselves A set of industry-oriented AI model fine-tuning and communication methods to realize the final implementation of AI in all walks of life.

The prompt word is similar to the search engine keyword, which is a paragraph we provide to obtain information from the conversational AI model. You may think that sending a paragraph is not easy? However, since the content of the generative AI is randomly generated, and the generated content is different each time, in order to obtain as accurate an answer as possible, the dialogue with the AI ​​needs to follow a certain paradigm. The following is one of the typical paradigms that a prompt word needs to follow one:

In addition, it is also necessary to set the two parameters of the AI ​​model's Temperature accuracy or rationality Top_p core sampling to control the degree of certainty of the model when generating a response. In addition, it is necessary to classify the different scenarios of the dialogue: typed dialogue, summary summary dialogue, programming type, role-playing type, question and answer type, reasoning type, etc...

All in all, prompt word engineers need to play the role of a communication bridge between humans and AI. On the one hand, they need to gain in-depth industry knowledge and understand human language; Help AI complete the landing in specific industries. Such industry-oriented interdisciplinary talents will become shortage talents in various industries.

For various small and medium-sized software manufacturers, they must either fully open up their software products, empower AI, and become the hands and feet of AI; or sink into specific industries and become AI-based service providers for customers. Timely transformation and open cooperation are the way to survive in the new era.

>>People's Lifestyle in the AI ​​Era

The first is a huge increase in productivity.

The emergence of computers and a large number of mechanical equipment has transformed human society from a labor-intensive society to a brain-intensive society, and the emergence of AI will greatly reduce the intensity of human mental work, allowing human society to transform into a creative and leisure society. Like the current living conditions in Northern Europe.

The productivity released by AI allows people to have more leisure time, pay more attention to love and life experience, to think and meditate, and to explore the meaning of life. Many people don't work because of love, we just work for living, for social opinion, and for material things. I am born to be useful. With the help of AI, we can truly understand where the talent I am born with is intended to be used, so that we can not waste a trip in the world, so as to better pursue the meaning of life.

Our generation is lucky. Our ancestors spent decades building this country from a poor and white society to a materially rich society. We will no longer struggle to survive because of basic survival problems. But our generation is unfortunate. We are in an era of change. The rapid increase in material life will make people temporarily lost in hedonism. But in the end, we have to move forward again. For the double harvest of material and spirit, fully release the low-level mental work, and embrace the new era and life.

This is the worst of times, and it will be the best of times!

references:

"Artificial Intelligence in Game Programming"

"How will super artificial intelligence lead to human extinction or immortality in our lifetime?" "

"The Singularity Is Near"

《Sparks of Artificial General Intelligence: Early experiments with GPT-4 》

《Prompt Engineering Guide》

《Prompt Engineering Overview》

---

Communicate with Deputy General Manager Zhao Xiang of Hanma and contact Hanma's assistant: HANMAJISHU    

Everyone is welcome to click " Read the original text " at the end of the article to register on the Hanma Yidao Cloud platform. The Yidao Cloud AI assistant will be launched on April 2.

About Hanma

Suzhou Hanma Intelligent Technology Co., Ltd. (referred to as "Hanma Technology") is a subsidiary incubated by Suzhou Hanchuan Intelligent Technology Co., Ltd. (688022), the first batch of companies listed on the Science and Technology Innovation Board. As a PaaS platform focusing on the industrial field, Hanma Technology is committed to building an end-to-end enterprise operation support platform for the supply value chain.

Hanma Technology insists on deeply cultivating vertical industries, and has served many top companies in advanced manufacturing fields such as new energy lithium batteries, auto parts, 3C electronics, and life health. It is a rare company in the industry that has both deep industrial accumulation and advanced digital capabilities Enterprise digital new infrastructure technology service provider.

read the original text