Author | Mr.K Editor | Emma
Source | Technical Leadership (ID: jishulingdaoli)
Professor Andrew Feinberg, a famous technology philosopher, once mentioned a very creative concept, "Technology Democratization". The professor believes that technological democracy is to expand the free boundaries of social individuals, so that everyone can actively and effectively participate in technical design and technical decision-making, including "layman actors" of different identities and classes, so as to ensure their freedom. Interest demands are fulfilled. In short, it is to allow a wide range of people to participate in technical design, and ultimately achieve a greater degree of technical collaboration.
When I saw this statement before, I either felt that it was difficult to realize, or that it was still far away from us. It wasn't until the past six months that various real-world applications of AI large models continued to appear that I began to realize that the so-called "distant" might not be that far away. Recently, I witnessed that in the DingTalk group chat, just by taking a photo or writing a sentence, a business application can be automatically generated without a single line of code, and I realized that I am still superficial. How is technological democracy not far away? , it is clear that the future has come and is close at hand.
01
AI large model saves the life of "low code"
The debut of ChatGPT is the pinnacle. While constantly refreshing everyone's various understandings of AI, it is also constantly changing its own life in a self-subversive way. From GPT-3.5 to GPT-4, there are many unexpected function improvements. In the official promotional video of GPT-4, there is such an impressive scene: In the video, the tester draws a sketch of the website on a piece of white paper, uploads it to GPT-4, and then a surprising scene appears, just relying on A sketch, GPT-4 actually gave the HTML code of the website. Brother K believes that when this scene occurs, someone must be watching the fun and watching the doorway. Those who watch the excitement may marvel at the miraculous transformation from diagrams to codes; those who watch the doorway must be thinking, what is this code, this is clearly the "resurrection pill" that saves low-code from dying.
The so-called low-code (low-code) is a concept corresponding to all-code (All-code), which means that the required application can be developed with only a small amount of code through graphical drag and drop, parameterized configuration, etc. or digital tools. Since the emergence of low-code theory, it has obvious advantages such as low development threshold, high efficiency, and the ability to meet a large number of software development needs of enterprises in digital transformation. However, it has always been questioned and even complained that "it will only generate a bunch of shit". The same code" and "it is inconvenient to find bugs" are "pseudo-demand" and "industry cancer".
The emergence of large AI models represented by ChatGPT has become a timely rain to save low-code, and may even become the most powerful blow to eliminate doubts about low-code. We have seen that AI large models are widely used in image recognition, natural language processing, speech recognition and other fields. In addition to the familiar ChatGPT, there are also Google's BERT model and Facebook's Detectron2 model. They have also achieved great success in the field of natural language processing and computer vision. These models are based on the deep learning of algorithms, and summarize the laws and patterns through the training of massive data, so as to achieve very accurate recognition and prediction capabilities. And this perfectly solves the "business needs" problem that is the most troublesome for the technical side.
When the AI large model accurately identifies business needs, what low code presents is no longer the "negative energy" of "shit-like code", but the "great energy" of helping enterprises quickly build and deploy applications and improving the efficiency of digital transformation of enterprises. Guangzheng". The AI model makes the program more intelligent, and the low-code makes the development agile. The two are a perfect match. With the blessing of the AI model, the dying low-code is alive, isn't it surprising?
02
AI large model + low code:
Solve 90% of the software needs of small and medium-sized enterprises
Some experts speculate that AI low-code development will not only solve 90% of the software development needs of small and medium-sized enterprises, but is even considered a new development model after waterfall development and agile development.
1. Learn the software templates of various industries with large models and become an expert in business system design
In this era, software is defining everything. Most of the scenes in our lives have quietly completed the process of "informatization". It would not be an exaggeration to say that software has reshaped the world. But even so, people's demand for informatization is still breaking out in more and more detailed scenarios. The "supply" of software development cannot meet people's growing digital needs, and it is still the main contradiction that is difficult to solve in the technology industry .
If we say that the emergence of low-code, it has pointed out the direction for the rapid delivery of business applications and the use of platform tools to speed up software development on a higher abstraction dimension. Then the emergence of the AI large model provides a stronger guarantee for its landing. Through large-scale training of deep learning models, the models can learn software templates and business processes in various industries, thus becoming an expert in business system design.
It can generate high-quality business system design solutions and improve the efficiency and accuracy of digital transformation of enterprises by learning massive data, codes and documents, and best practices in various business scenarios. When there are new software design requirements , the large model can recommend high-quality design solutions and code implementations according to requirements, just like industry experts.
For example, if a hospital needs to develop a comprehensive management system, the AI large model can recommend a set of high-quality solutions based on the study of a large number of cases in the field of medical informatization, including functional module division, process design, prototypes with rich interfaces, and The underlying code framework, etc. The high intelligence of this software design can greatly shorten the development cycle, and the cost and risk are also greatly reduced.
To make a popular analogy, AI low-code software development is very similar to the automated production line in the manufacturing industry. It is no longer a point or individual change, but a systematic upgrade. What is improved is not just the technical framework, but the new software engineering thinking.
2. Generate requirement documents by chatting and interacting with customers
I believe that many people still remember the shock and excitement brought by the smooth and powerful interactive experience that almost subverted our cognition when they first used ChatGPT. Just through a simple dialogue, it can give us a lot of unexpected gains. In fact, the basic implementation logic to achieve this is not difficult to understand: the interaction of chatbots is based on some predefined templates and rules, allowing customers to express their needs through simple conversations, and the chat content is transmitted to the AI model. The model will use natural language processing technology to analyze and model the text, accurately extract the customer's demand information from it, and convert it into a text format. But what really surprises us is not just these, but the scenes and uses of this interaction, which seem to have infinite "extensibility".
Uipath, the global leader in the RPA industry, analyzes and evaluates customer feedback on specific products through interactive text between users and ChatGPT. For example, transmit a set of product feedback received from customers to ChatGPT, and enter a prompt such as "Please determine the sentiment of this feedback and assign it as positive, negative or mixed". These tones are then accurately identified by ChatGPT. Then these user emotional feedbacks are formed into documents and automatically sent to the product development team to strengthen product design and development.
3. The large model is based on documents, generates codes, and completes software development
Similar to production documents, AI-generated code also needs to go through a lot of program code training, so that AI can learn programming technology and master the syntax and structure of program code, so as to automatically generate program code that meets the requirements to assist programmers in implementing code complements. Alignment, algorithm logic generation, language conversion and application generation. This method can greatly improve the development efficiency of technicians, especially in tasks with high repetition and simple logic.
At the practical level, many large domestic and foreign companies such as Ali, Tencent, Microsoft, and Amazon are already using AI large-scale model algorithms to assist programmers in their work. It is not difficult to imagine that with the rapid development of IT productivity, large-scale production codes will become more and more deeply involved in the current programming work.
03
AI large model + low code, a new paradigm of software development
As soon as GPT came out, Bill Gates called it "the most revolutionary technological advancement in more than 40 years." Subsequently, another industry leader, Nvidia founder Huang Renxun also followed up and said: "We are in the iPhone moment of AI." Is this an exaggeration? Maybe not, on the contrary, more and more facts are constantly showing that in the era of AI, all industries are worth doing again with large models. At least, it is very obvious in terms of software development. Under the combination of AI large model + low-code "smartness", the software development model in the new era is becoming more and more clear.
1. The biggest headache for low-code companies is that customers do not know their needs
In the traditional software development process, it is often difficult for customers to clearly express their needs, resulting in extremely high communication costs and low communication effects during the development process. Low-code companies are faced with such a dilemma. The original intention is to create more value in less time, but the reality often backfires. Instead, users (especially those who are completely ignorant of technology) are scolded in front of the computer. "Garbage" and "useless at all", and the problems hidden behind "slots" and "infamy" are not low-code itself, but user needs. In the past, in the face of demand deviation, low-code businesses generally carried out two aspects of work for lay users who did not understand at all:
1) Conduct training
Through help documents, video guidance, etc., users are trained on the logic of the tool, the problems that can be solved, and the commonly used methods. Allow users to better solve problems when using or encountering problems.
2) Provide a template
Just like teaching a person to quickly improve the level of PPT production, the best way is to give him a template that can be used for reference and modification. Many low-code platforms also do this, templated and modularized functions commonly used in different fields. Allows users to use or edit more conveniently.
To be honest, these two ways of solving problems seem direct and simple, but they actually test the user's initiative, acceptance and patience. They originally wanted to use the tool conveniently and foolishly, but they found that the whole use process was not so "foolish", which would increase their extra time cost and learning cost, and the user experience would naturally be affected to varying degrees, which is also the case in the past One of the main reasons why low-code has been so controversial.
2. In the era of AI large-scale models, chat generation needs, drawing interface can generate software
In the age of AI, making everything more "stupid" is gradually turning imagination into reality. Today, when large-scale models are in full swing, due to the rapid development of natural language processing and image processing technologies, as long as you can chat and draw interfaces, you can generate requirements and software trends are gradually taking shape. The core idea of this trend is to allow the machine to automatically complete the requirements analysis and software development process by interacting with the AI large model. That is to say, the AI large model completes the following two steps in a silent manner:
Step1: Demand analysis
The AI large model can analyze the natural language input of customers and automatically generate requirement documents. This approach can greatly reduce the time and effort required to write requirements documents, while improving the accuracy and readability of documents. This ability has been fully demonstrated as early as the GPT-3.5 stage.
Step2: Software development
Users can draw interface prototypes or renderings by hand, and the AI model will automatically generate front-end interface codes, back-end logic codes, and database designs based on these drawings to form a software system that can be used directly. The whole process does not require the user to have any development skills, and can be done entirely by the automatic generation capability of the AI large model. And in this way, users only need to focus on whether the actual needs are consistent with the interface effect, and do not have to pay attention to the specific technical implementation. Whether it is the sketches in the GPT-4 promotional video that can be transformed into HTML codes in seconds, or the DingTalk "Photograph Generation Application" function experienced by Brother K, it is an intuitive embodiment of the AI era where everything can be generated like a fool.
3. Continuously tune the model to generate software that understands users better
During the development process of the AI large model, it can also continuously learn and optimize itself, so that the generated software can better meet the needs of users and improve customer satisfaction. To give a simple example, in March of this year, Ali launched Tongyi Qianwen, which has a cool little function, that is, "chat records do not need to be turned, and summaries are automatically generated".
Everyone knows that most of today's migrant workers are multi-line jobs. There are few or even dozens of groups on their mobile phones. All kinds of unread messages are constantly accumulating, and they can't be read when they are busy. Sometimes I spend a lot of time climbing stairs, only to find that the group is full of worthless and useless information. However, DingTalk, a new feature, can automatically generate chat summaries for users based on historical messages. It can not only track information, but also save time and effort to climb stairs.
Although this function seems inconspicuous, it well reflects the ability of AI to continuously optimize and generate software that understands users better according to various explicit and implicit needs of users. The Microsoft 365 Copilot launched by Microsoft in March, let us see the capabilities of Word, Excel, PPT, and Outlook beyond our imagination, which is actually what it means. Moreover, we have reason to believe that in the near future, these tools will be further optimized and upgraded, and you may even be surprised to find that it understands your needs better than yourself.
4. The generated code does not need to be tested
The AI large model not only has high working efficiency, but also has good stability and has quite good quality assurance. First of all, the ability of AI large model to generate code comes from the study and imitation of a large number of development cases. It can generate code that meets industry standards, is highly robust, and has excellent performance. Secondly, the AI large model automatically generates codes based on requirements documents and design schemes, which can also avoid various bugs caused by negligence and errors during the manual development process.
In fact, we can also see the clues from the name of the AI large model. The full name of the AI large model is "artificial intelligence pre-training large model". On large-scale datasets, pre-training is done. Therefore, it has sufficient stability and accuracy, and it can directly support various applications without fine-tuning, or only requires fine-tuning of a small amount of data, and it is easy to understand.
04
Low code without AI big model, no future
The AI large model greatly reduces or even completely destroys technical barriers. Only through chatting and drawing, software development and optimization can be realized, which in turn greatly reduces the overall cost of users using low-code. Let the low-code, which was once confused and has an unknown future, return to the wind.
Gartner predicts that by 2025, 70% of enterprise digital applications will be built with low-code, and even some experts regard low-code as "the core engine of enterprise digitalization." But it is not difficult for a discerning person to see that the reason why AI+low code and AIGC+low code are so popular is that the AI before the "+" sign is the key. In other words, the prevalence of low-code is inseparable from the detonation of AI. Without AI large models, low-code may not have a future, and even still cannot get rid of "low-code is only suitable for edge innovation and building long-tail applications", "low-code Just a toy" kind of fate.
1. The core value of the low-code platform is to allow non-technical personnel to easily develop software, but the quality of the development results is difficult to guarantee and cannot meet the digital needs of most enterprises . AI large models can generate high-quality codes and systems to solve this pain point.
2. Although the threshold for using the low-code platform is low, users face greater difficulties in expressing their needs, which affects the interactive experience and development efficiency . AI large models can fully obtain user needs through natural language understanding and interaction, and provide a more friendly interface for low-code development.
3. The increase in the popularity of low-code platforms requires the accumulation and optimization of a large number of templates, components, and frameworks . This requires a lot of manpower and time, and progress is difficult to guarantee. The AI large model can quickly learn and recommend high-quality design templates, accelerating the improvement of low-code development capabilities.
4. As a result of low-code development, the function and quality are difficult to meet the requirements of high-complexity business systems . This hinders the coverage of a wider range of customer groups. The AI large model can generate a more powerful, comprehensive and refined software system for the low-code platform, changing this situation.
5. The update and upgrade speed of the low-code platform is difficult to meet the continuous improvement and change of user experience . The AI large model can automatically optimize low-code templates and generate results based on user feedback, enabling low-code software to have stronger learning and iterative capabilities.
Babu, the person in charge of the DingTalk open platform, once said: "Deconstruct the complex software system, extend the rights of data to every individual, and combine the individual creation of employees with the effective governance of the organization." What he expressed Meaning, to a certain extent, it is similar to the theory of "technical democratization" mentioned at the beginning of the article. They all point to the fact that the current AI large model + low code is not only a trend, but also a subtle trend. Total revolution. We are born at the right time, optimistic about its success, and cheering for the times!
Profile
Profile of the account owner: Mr.K , Huang Zhekeng, expert in digital transformation of enterprises, founder of "Duwu Hill", technology blogger, former technical executive of Haier, Zhongtong, and 1 Yaowang, author of "The Way of Technical Personnel Practice" The Summit of Technology Management. Focus on: technology trends, business, personal growth.
- END -
Musk: I'm so excited to have found my mission in life...