From hearing about it to using it for myself, this is how I got started with AI

Matrix Home Page Recommendation

Matrix It is a minority writing community. We advocate sharing real product experiences, practical experiences and thoughts. We will select the best articles from Matrix from time to time to show the most real experiences and opinions from users.

The article represents the author’s personal views, and the minority has only made minor modifications to the title and layout.

AI is developing so fast, and there is a lot of new information every day and every month.

When you read public account articles, browse Xiaohongshu, and watch short videos every day, do you also see a lot of AI-related push notifications: yesterday Musk open-sourced Grok, today Alibaba started the internal test of Tongyi Lingma, and tomorrow KimiChat will start the internal test of the 2 million context model.Suno.ai 3.0 is online again, and the quality of the generated music has been greatly improved… But because there are a lot of information gaps, and I don't know how to learn AI, I am left with only a bunch of questions:

If I don't learn AI, will I be left behind? How can AI help my life? I want to learn AI, how should I learn it? These problems eventually turned into information anxiety.

So today I will talk about how I learned AI.

What is AI useful for?

Artificial Intelligence (Artificial Intelligence) is a very young technology that can be traced back to the 1950s. However, it was not until scientists proposed the concept of deep learning in 2006 that artificial intelligence really entered people's field of vision.

The AI we are talking about today is specifically the AI craze triggered by ChatGPT starting in November 2022. More specifically, what we are actually discussing is “generative AI”.

Generative AI is an artificial intelligence technology that can create or generate new content based on existing data. Just like a painting robot, it can draw a new painting on its own after learning thousands of paintings. This kind of AI can work in fields such as music, text, pictures and videos. It understands the patterns and rules of data and then applies this knowledge to create new things. Simply put, generative AI is an intelligent system that can “imagine” and create new things.

Generative Artificial Intelligence (GenAI)

Generative AI can be understood as AI generating content under the prompt of text. According to the different content carriers, it can be divided into the following directions: text-generated text, text-generated images, text-generated videos and text-generated audio.

Wenshengwen: ChatGPT, Claude, KimiCChat, Wenxinyiyan, Tongyi Qianwen, etc. are all large language models, which can be understood as text-generated text. As long as you provide a prompt word, it will follow your requirements. Generate text. Expanding the scenarios of generated text include writing, summarizing, dialogue, emotional analysis and other scenarios.
Vincent pictures and pictures: Stable Diffusion and Midjourney that are often used are based on diffusion technology. By inputting a prompt word, AI will generate the corresponding picture.
Vincent video: Sora, which became popular some time ago, is the Vincent video technology of OpenAI. This technology has exploded around the world just by relying on the video preview released by OpenAI without any public or internal beta. In addition, Stable Video Diffusion (SVD) based on diffusion technology is also under development.
Vincent Audio: TTS technology can be used for timbre reproduction, speech synthesis, music synthesis, etc. I have not systematically studied these technologies, I only know that they have these functions.

If we look at “Wen Shengwen> Wen Sheng Audio> Wen Sheng Picture> Wen Sheng Video” according to the maturity of technology. The AI technologies being discussed now are all centered around the direction of generative AI. If you don’t need these in your life or work, then maybe you don’t need AI and your life can be very good; and if you are interested in AI If you are interested and want to step into this torrent, let us start learning AI.

How to enter the new paradigm of GenAI

In my opinion, generative AI is undoubtedly a new “paradigm”.

A paradigm is a set of scientific theories, experimental methods, instruments and equipment, and rules and standards for explaining natural phenomena that are generally accepted by the scientific community during a certain period of time.

From today on, all human activities will be inseparable from generative AI, so how to enter this new paradigm?

How to enter a new field

Before talking about how to learn AI, let’s talk about a bigger question – how to enter a new field.

Get a map

When entering a new field, the most important thing is to have a map in hand. – Gan Jiawei

Entering a new territory is like stepping into a completely unfamiliar territory. The most important thing is to only get a map of this area. And for those who have just entered the field of AI, what is the map of AI?

On the timelineyou need to understand the general development history of AI. Here is a recommended book called “Deep Learning Revolution”, which talks about the past of deep learning before GPT3.5, which can give you a preliminary understanding of the short history of deep learning.

Of course, the development of AI is very short. Starting from 2020, the development of AI seems to have broken through a “singularity”. Suddenly, both academia and business circles flourished, and many directions and knowledge in many sub-fields emerged.

on the horizonthe current development enthusiasm of AI, as mentioned above, probably has the following directions:

Wen Shengwen
Vincentian picture
Vincent Audio
Vincent Video

You need to have a basic understanding and try out the popular products in each field.Of course, if you want a more comprehensive understanding of a field in academia, such as the field of learning large language models, you can read[LLM_Survey]Overview of large language models。

Minimum necessary knowledge

When we enter any new field of study, we start from the basics to the depths. At the beginning, we may learn some concepts, some basic frameworks and principles, and then learn more in-depth and cutting-edge knowledge.

Continuing this idea, every field has“Minimum necessary knowledge”.

When a certain skill is required, you must find a way to figure out the minimum necessary knowledge in the shortest possible time. Then master them quickly.

With the minimum necessary knowledge, you can defeat the other 80% of people. Just like in ball sports, if you practice basic skills well, you can defeat 90% of humans. Mastering the “minimum necessary knowledge” in any field can quickly open up the gap with others.

For another example,

The minimum necessary knowledge for design is to organize information architecture and color selection. Organizing information architecture means that you can abstract and organize complex information into structured information and express it. Color selection is the key to making the design look good.

The minimum necessary knowledge of photography is picture composition and imaging principles. Understand the imaging principles of the camera, and know how to adjust ISO, exposure duration, aperture size and shutter speed. If you know how to construct a picture, you can easily take pictures that will lead your circle of friends.

In the field of AI, the minimum necessary knowledge that needs to be mastered is the principles and use of generative AI.. As mentioned above, if you understand the most popular generative large language models and stable diffusion technology and apply them in your work and life, you can defeat 80% of people.Recommend my friends here – feel free to search for Newsletter host Jimmy'sintroductory course。

From this page, you can learn:

How to write ChatGPT prompts in a structured way
The principle of Stable Diffusion
Generating images using Midjourney
How to draw with Stable Diffusion

Monster-fighting missions in Novice Village

When you have mastered the minimum necessary knowledge, you have mastered the method of killing monsters and leveling up. If you want to fully master a skill, you must practice it.

So after completing the minimum necessary knowledge, you need to complete the monster-killing mission in the Novice Village:

Use large language models to generate content
Master the ability to write structured prompts and use ChatGPT to meet content production needs
Generate images using Midjourney
Use Stable dDiffusion to generate images
use Suno.ai Generate a piece of music

First use of a large language model

This step is like someone who is learning code for the first time. print("Hello World") Again, welcome to the world of AI.

There are many choices of large language models now. Domestic ones include Wenxin Yiyan, Tongyi Qianwen, KimiChat, Doubao… Foreign ones include ChatGPT, Claude, Gemini… Open the website or APP of these large model applications and issue your first command. This will enter the new paradigm of generative AI question and answer.

Structured prompt ability learning

Prompt means prompt word. This word will be used frequently in the future. In scenarios such as Vincent pictures and Vincent videos, prompt means prompt word.

You provide the model with a prompt word, and it generates content based on your prompt word. In the field of large language models, good prompts can help you significantly improve the generation effect of the model. If you want to have text generation scenarios, learning structured prompts is a must.

For learning methods, please refer to my previous article “ChatGPT prompt writing guide: Let AI work for you》。

Midjourney generated images

Mastering the ability to use Mjidouney will allow you to draw beautiful pictures even if you are not good at drawing. I have written beforeThis articleexplained in detail how I used Midjourney to draw my own red envelope cover.

Enter some keywords in Midjourney to generate nice pictures. For example, the prompt of my red envelope cover is:

chinese dragon, cute, anime, flying， chinese spring festival atmosphere,chinese dragon, cute, anime, flying， chinese spring festival atmosphere --ar 3:4 --niji 5

This may seem like a bunch of English words put together, but Midjourney's prompt words also have their own techniques. You can read Midjourney'sOfficial DocumentationCome and learn. There are also relatively good free courses on the Internet. There are many such video sites on B. You can search for them.

Generate images using Stable Diffusion

In theory, both Midjourney and Stable Diffusion use diffusion technology. So why do you want to experience Stable Diffusion after experiencing Midjourney?

That’s because although the two are based on the same technology, their routes are completely different.

Midjourney's vision is to reduce the difficulty of producing good-looking images, and it takes a closed-source route. Through a large number of artists' image training, users can easily create good-looking images. Stable Diffusion's vision is to make diffusion technology more stable, and it takes an open-source route. Because of its open-source ecosystem, many designers and developers have built a very powerful model ecosystem based on Stable Diffusion.

In other words, Midjourney can only let you experience the image generation model, but to really learn and research, you have to learn Stable Diffusion.

Stable Diffution requires a lot of computing power to be deployed locally. You can experience it in some online ways. For example stability.ai The official experience website of Dream Studio, just enter the prompt and select different models to generate pictures.

Use SunoAI to generate a piece of music

recently sun The v3 version of the model has been updated, and the quality of the generated music is much better, so it has also received a wave of widespread attention. You can ask ChatGPT to generate a piece of lyrics for you, and then give the lyrics to Suno to generate a piece of music to try.

Generate a video using Pika or Runway

Pika has a similar function to Runway and can animate a static image. Some up owners have begun to use these two tools to create videos. However, it is still at a relatively early stage and can only be used to perform some simple actions.

You can use the image generated by Midjourney earlier, upload it to Pika or Runway and enter some descriptors to generate a video.

I once tried to animate the red envelope cover generated by Midjourney, but it seems that Pika's support for two-dimensional images is still relatively poor, and the generated quality is unsatisfactory.

Find a direction and keep learning

Characteristics of the AI field

The biggest feature of the generative AI field is “fast”. It has developed in a short time, at a fast pace, and is changing rapidly.

For example, the Stable Diffusion technology we mentioned above is only a technology that has only been available since 2020. This characteristic means that the knowledge we learn is cutting-edge and fragmented. There is a lot of knowledge that you may not be able to learn directly from books.

A lot of knowledge comes from the Internet, update announcements from large model manufacturers, video tutorials from UP owners, etc. This relies heavily on personal information processing capabilities: information retrieval capabilities, information filtering capabilities, information processing capabilities, fragmented learning capabilities, the ability to organize information…

Find your interests

There are many subdivisions of AI development, and each direction is currently developing rapidly. There are many different technical routes and different application scenarios in each of the four major categories (text, text, audio, and video).

There are different technical routes, such as Chatbot, RAG (search enhanced generation), Agent (agent, etc.). Agent is further divided into single agent, multi-agent, and auto-agent. Different technical routes have very different applications in different scenarios.

So when you face such a novel situation with so many directions, to be honest, your ability and energy cannot support you in covering everything. In this case, the best way is “T-shaped learning”, which means to conduct in-depth research and study in an area that interests you most, while maintaining the most basic learning activities in other areas.

At this time interest is the best teacher. Find the direction you are interested in and study it in depth, learn the most cutting-edge knowledge, and find actual implementation scenarios. The time you spend will definitely be rewarded. At this very early stage, the ROI on investment of time is very high.

Where is the quality information?

You will definitely encounter this problem when studying: facing so much information every day, and a lot of new information, how to filter out high-quality information?

A basic formula is: primary information > second-hand information > third-hand information.

first hand information

Primary information refers to information obtained directly from primary data sources, which may be direct observations, experiments, surveys, interviews, or other first-hand experiences. This information is usually original and has not been processed or interpreted by others.

Primary information is most commonly found in:

Academic Papers
Websites of large companies

Friends who have the ability to read papers can pay attention to papers in their fields of interest. Here are a few recommended websites.

The first website is LetterGPT, the positioning is to translate the paper into Chinese and then push it. Paper inquiries and subscriptions can be made within the website.

The second website is Arxiv, an overseas paper website where most AI-related papers are published.

The third website is a website that helps you read papers – Connected papers. As the name suggests, after you search for a paper, it can recommend more papers based on the citation relationships between the papers.

On the websites of large companies, common large model manufacturers, such as OpenAI, Claude, Midjourney, etc., all provide official documentation. You can often read a lot of good content in the documentation.

For example, OpenAI provides it on its websiteBest PracticesClaude also provides instructions for writing prompts in Claude in its own documentation.Best Practices. There are many similar manufacturers. If you want to know something, just read their official documents first.

Second hand information

The definition of second-hand information refers to information that has been published or exists. After another transmission of information, the second-hand information has been sorted, analyzed, explained or summarized. Because there is too much and too new AI information now, the existence of second-hand information is necessary. Through good secondary information creators. It can help us obtain better information.

The source of the most second-hand information is “netizens”. This kind of behavior can be compared to curation, which is the same as the effect of writing a newsletter, organizing good information and sharing it. Netizens will share on the Internet:

New technology, new tools
Some experiences and methods of using AI myself

These are all good learning materials. Common places for gathering second-hand information in the AI industry are some social media, such as Moments, Twitter, some KOL newsletters, websites, etc. Here are some information sources that I often check:

The first is Immediate (the community with the highest concentration of AI in China). Generally speaking, you can see a lot of high-quality AI information and usage by following these two circles:[AI Exploration Station]and[Artificial Intelligence Discussion Group]. Of course, I also recommend some users that I often read:

@Guizang: The first to synchronize AI consulting, share the use of LLM and Stable Diffusion, and is also the host of the AIGC Weekly Newsletter, publishing the latest AI news every week.
@JimmyWang: Founder of the learning prompt website and the creator of the “Random Search” Newsletter
@吕立青_JimmyLv.eth: BibiGPT (can summarize B station videos and podcasts) is the person in charge of this product
@宇一.Dev: Share how to use large language models
@海丝Hyacinth: Has powerful Stable Diffusion refining ability
@Simon Awen: Has powerful Stable Diffusion refining ability
@idoubi: Former Tencent senior engineer and AI application developer. Very strong. He has developed mainstream AI products in almost all tracks.
I would like to recommend another wave of my own account @草草士: I will no longer share AI consultation immediately, but will update useful AI usage methods.

There are many excellent creators in Jiji Town. After being active in Jiji Town for a period of time, you can naturally find the people you want to follow.

Twitter It is also the community with the highest concentration of overseas AI, and many excellent bloggers in Chinese and English are here. I would also like to share some Twitter bloggers that I often follow:

@dotey: Teacher Baoyu, shares course translation, prompts, development methods, etc. in the field of AI.at the same timeblogManager. Teacher Baoyu will be updated simultaneously on Weibo, just search Baoyu on Weibo.
@WaytoAGI is the account of the best Chinese AI knowledge base. It is a community formed by a group of AGI enthusiasts.
@hanqing_me history, focusing on exploring AI short video creation, AI Talk is the project he is currently creating
@oran_ge Orange, a practitioner in the AI industry, an excellent product manager, a very frank and interesting person.
@vista8 Xiangyang Qiaomu, sharing efficiency tools, how to use LLM

There are also many high-quality English bloggers on Twitter, but due to my limited English proficiency, I don’t read them very often.

@sama OpenAI CEO Sam Altman’s account, also recommend his blog blog.samaltman.com
@thesephist Linus, how should I introduce him? He is a developer who combines art and technology.

Reposting between bloggers can help you discover more interesting bloggers.

The AI industry newsletter also has some recommendations:

The AIGC Weekly by @Guizang teacher mentioned earlier,Subscription address
Ben's Bite: Sharing AI consultation every day,Subscription address
You can also pay attention to Product Hunt Weekly. After all, almost half of the new products are AI-related nowadays. You can see what new AI products are launched.
I also recommend my Newsletter —— Draft Supplement

Third-hand information

The definition of third-hand information is information that has been simplified and modified by someone who does not understand it. It is usually information compiled for a specific purpose. We should avoid and discard this type of information.

I strongly advise against learning common third-hand information, such as AI information videos on short video platforms. Short videos are a media form that is short, flat, and fast. The medium itself determines that they are more suitable for carrying entertainment content rather than knowledge. Furthermore, I think it is too easy to be misled when learning knowledge through videos.

There are several reasons:

You can use beautiful images/tones of language to incite emotions to cover up logical loopholes.
Watching videos is mostly a passive learning scenario, and few people will think critically after watching the videos.
In video communication, knowledge is greatly simplified for the audience to understand and accept easily.

If you really want to learn seriously, the best way is to choose to read the text. If you can't read the text, you can choose a systematic video tutorial.

The advantage of video learning is that it is more suitable for imitation learning. For example, if you want to learn a sport, watching a video is much better than reading a book. Or if you are a novice who has never been exposed to code or deep learning, watching videos + imitation will also be more effective than reading a book.

PS In 2020, I made a point of not recommending learning through videos.

Extra: The information war between fragmentation and system

Learning can be divided into systematic learning and fragmented learning according to the length of time spent on the learning content.

Fragmented learning, such as reading a short article, looking at a note on Xiaohongshu, watching a short video, etc.
Systematic learning refers to studying a course, reading a non-fiction book, watching long videos, etc.

Generally speaking, in a mature field, we believe that the information quality and information density of systematic information will be much higher than those of fragmented information.

The knowledge learned in fragments is fast, but often scattered, and you need to organize the fragmented learning into a knowledge system yourself. The role of fragmented knowledge for competitiveness is relatively weak. It is difficult for individual knowledge points to play a role. Only by connecting knowledge points into lines, forming a network, and organizing them into a system can the linkage between knowledge and knowledge play more roles. effect.

The content learned through systematic learning is more systematic and logical, and can form a complete knowledge system. But it requires more time and effort. Older knowledge systems may not be able to keep up with the latest content.

In mature fields, we recommend using the “barbell reading method”:

barbell reading method

In Nassim Taleb's series of books on randomness (“The Random Walking Fool”, “Black Swan”, “Antifragility” and “Asymmetric Risk”), a reading method is mentioned: the barbell reading method. The barbell reading method means that when reading, only read both ends of the timeline. On one end are classic books in the field, and on the other end are the most cutting-edge research results, cases, and ongoing happenings.

Read the classics because they have stood the test of time. The Lindy Effect says that the longer something exists, the longer its life expectancy. Classics like “The Analects of Confucius” will be passed down forever, and useless books that have just been put on the shelves will most likely be eliminated by history.

Read the leading edge because you can get the relative time difference.

Information screening

In the field of AI, all information is still too new, so the barbell in the field of AI may be unbalanced, requiring you to have your own information screening capabilities.

There is some systematic information, such as Andrew Ng's artificial intelligence course mentioned above, which is relatively systematic and complete. But most of the information is still fragmented:

The papers are fragmented, but there are citations between them, and the citation network will be more systematic.

The dynamics are fragmented and you need to identify high-quality bloggers yourself.

Articles are fragmented, and you need to evaluate the quality of the articles (of course you can also use AI to evaluate, refer to my previous article “Incremental reading method based on KimiChat》。

Finding best practices for GenAI in your field

Learning and thinking complement each other. Only by obtaining feedback and thinking through actual actions can we truly learn in depth.

Learning by doing is the fastest and the best effect. Learn and apply AI to your work and life events. In the process of doing it, what I learned became more solid and profound. You can learn more by doing it first and reviewing it later.

Regarding the practice of AI, after understanding the learning field of AI, you can find a best practice in the field you are familiar with.

“Best practices” refer to technologies, methods, processes, activities or mechanisms that have been tested and proven to have excellent results in a specific field. They are called “best” because these practices demonstrate advantages over average or existing practices in improving efficiency, reducing costs, improving quality, ensuring safety, enhancing customer satisfaction, and more.

One of the characteristics of general artificial intelligence is “generality”. Taking ChatGPT as an example, I believe that the way ChatGPT is used in each industry and each position may be different.

For example, I wrote in “What is ChatGPT good for the average person?The example given in this article: Teachers may use ChatGPT to complete some tasks such as student evaluation, but as a product manager, I may not be able to understand what problems civil servants or teachers will use generative AI to solve in their productivity scenarios. Only through interviews can I understand.

Because the application scenarios of generative AI are very scattered and vertical, you can definitely come up with your own “best practices” by combining your industry/professional methodology and your knowledge system with GenAI in the field you are familiar with.

Be a developer

Call the LLM interface to develop an AI conversational robot

The most entry-level call can connect to the interface of any LLM (Large Language Model) vendor, and then combine it with some local visualization WebUI projects, such as Built or ollama webuiimplement a conversational robot that can communicate locally.

Use rich Hugging Face to call large models

Hugging Face is a company focused on the field of natural language processing (NLP). It provides an open source platform of the same name to promote the development and application of deep learning and natural language processing technologies.

Hugging Face provides a model center on the website where users can find and share various pre-trained models and datasets, as well as an online demonstration platform where most models can be tested directly on the web page. You can choose the model to call according to your needs.

In addition, Hugging Face is even more powerful because it has developed a Transformer library. You can install the Hugging Face library to complete pre-training, fine-tuning, and inference of large models.

Develop a RAG conversational bot

The next challenge on the image route is RAG (Retrieval Augmented Generation), which translates to “search augmented generation” – the content is vectorized and stored, combined with vectorized search each time it is generated, and the results closest to the search problem are passed as context. Give large language models to assist large language models in generating better content.

You can implement this function based on Langchain. LangChain is an open source programming framework specifically designed for developing applications driven by large language models (LLM). By using the modular design function of langchain, developers can more easily implement chain calls, memory mechanisms, loss processing, etc., to complete AI development work.

Deploy Stable Diffusion yourself

If you choose the image generation route, the most important thing is to deploy Stable Diffusion locally. You can use web-UI or compose UI to build it. B station UP host @秋枼aaaki shared the entire web-UI project, which can be found in hisvideofound in.

Using open source diffusion model

exist Civitai.com There are many models shared by users on this website. You can download some models to try them out locally. This website is a bit like Hugging Face in the LLM field, but since I am not an image expert and don’t know much about it, I won’t go into details.

Of course, there are more defense directions that can be developed, but because the author is interested in the direction of LLM, there is relatively little research on other fields, so I cannot share more.

be a creator

The next area of learning AI is creation.

Of course, this step requires stronger professional knowledge and basic knowledge of deep learning, natural language understanding, image understanding, etc. Because I don't have it yet, so I'm still learning, so I can only give you some direction.

Fine-tune the large language model to solve your own problems

If you find that the existing large model is not performing well when doing some fixed text generation tasks, you can use fine-tuning technology to fine-tune the large language model.

Fine-tuning, the most commonly used method now is Lora, which simply means fine-tuning of the generation style, correcting the generation effect of the large language model by inputting QA pairs. The fine-tuned model will be more in line with the generation effect you want.

At present, all major language model manufacturers have improved fine-tuning interfaces, which can be fine-tuned with microdata; if it is a locally deployed model, you can also use the Transformer library provided by Hugging Face mentioned earlier for fine-tuning.

Train your own Stable Diffusion model

Stable Diffusion also supports fine-tuning, and training is much simpler than text. All I know is to input a large number of pictures, label them and submit them for training.

Also, because I am not taking the image generation route, I cannot expand on this here.

other

There is not much I can write about the Creator. Mainly because of my limited abilities.

Because knowledge grows fractal, new knowledge always appears at the boundaries. Any knowledge has boundaries. When you are close enough to a piece of knowledge, you will know its usage boundaries, and you can often discover new knowledge by exploring the boundaries of knowledge. , able to create new connections.

When it comes to the field of creation, there are definitely more directions, many of which are the latest academic results. It’s just that my current superficial knowledge can only be written here. I will share it with you when I learn more.

Communicate with people in this field

Human information sources can be divided into three categories: reading, talking, and doing. We talked about reading and doing, and the remaining one is communication.

Communication plays a very important role. Finding like-minded friends not only helps you find fellow travelers, but also increases the chances of a collision of ideas, which can often produce sparks.

You have friends in the field of what you do. To learn AI, of course you need to find friends with the same goal to discuss and learn together. In addition, you need to communicate with people who are better than you, so that you can gain knowledge beyond books. The content of books is generalized, and the content of conversations with people is specific.

Many important knowledge and profound experiences are difficult to describe in words, and therefore cannot be recorded in books. “It's not what you know, but who you know that matters.” Knowledge can only be acquired through your own efforts, while connections can help you acquire the knowledge that others have.

If you can use human leverage, you can outsource your study time and let others study for you.

open learning

One way to communicate with others is to make your learning process public.

As mentioned above, you can disclose your learning process on platforms such as Twitter or Instant and share your unique methods of using AI. This will attract friends who share your interests and hobbies. At present, there are still very few people in China who are willing to devote themselves to learning AI, so the circle should not be too big.

Conclusion

This article took a long time to write and edit.

Because AI learning is a big topic, I tried to write the article more simply, but it was not deep enough. But I tried to write more deeply. Except for the areas I am good at, I couldn’t write well in many areas, and the length was not enough. After continuous adjustments, it was finally written like this.

No matter what stage you are currently at, I hope you will gain something from reading this article.

If this article is helpful to you, you can like the article, or forward it to someone you like, or buy me a cup of coffee, or you can follow my official account and learn with me. Your support means I will persist. Motivation to go on.

> Follow Minority Little Red Bookexperience the wonderful digital life🍃

> Practical and easy to use Genuine softwarethe minority presents you 🚀