What Two Years of Prompt Writing Taught Me: Most People Write Like Machines

In-depth sharing of two years of experience in AI prompt words, revealing the misunderstandings of writing and the true value of AI applications.
Core content:
1. Exploring the misunderstandings and true meaning of AI prompt word writing
2. From imitation to innovation, the transformation and growth of prompt word writing
3. AI as a tool, how to improve personal abilities and find projects that you love
Today I want to talk to you about prompt words. I am probably one of the earliest people to come into contact with prompt words in AI (relatively speaking).
Let’s discuss a question first, and then I’ll tell you my story over the past two years.
Now, many people are using teacher Li Jigang's framework and learning to write in the same way. But is this right? Maybe this is a way to write good prompts, which is actually no different from using the framework directly.
Are things that look high-end necessarily high-end?
I don't think so. At least when I buy things, most of the products that look high-end and upgraded are finally a burden. Don't rush to scold me, read this article patiently first, I believe you may have your own answer.
I have been writing prompts for more than two years. To be honest, there are not many people who write prompts more than me. My understanding of prompts has been changing. From the initial needs, I wrote structured prompts and imitated teacher Li Jigang. Now I write a role for the needs and perfect this role. In many cases, the description of this role is greater than the needs.
Many people may find my prompts strange, wondering why something that can be expressed in just a few sentences has to be written in such a complicated way.
Just like the name of my public account, this public account has been updated for more than two years, and I have actually been on the road of exploration. I have written a long article on the structured prompts of Markdown syntax.
# Role
##Proifile
## Background
Skill
Goals
## Rules
## Constraints
## Workflow
## Initialization
I think you are more or less familiar with the keywords above. If you have used some large models and read some of the author's prompts, you will understand how popular and useful this thing is.
At that time, one prompt word would take 2000-3000 tokens to write, which was really a lot. Then I tested it over and over again to see how to get the best effect.
At that time, I always thought that the big model was just an assistant. The more powerful the assistant was, the more I could use his abilities and stand on the shoulders of the "giant".
"AI is an amplifier that can amplify all your abilities and even allow you to reach heights you never imagined."
This is what I learned in my first year. AI has indeed magnified my own abilities, whether it is the ability to write articles or the ability to write code. It even allowed me to learn a lot of things that I didn’t understand before.
If the AI industry wants to take business orders and make money through prompts and projects, being able to write prompts is actually just the most basic thing, and you also need to have a strong learning ability.
This learning ability means that you can use AI to quickly get started and understand an industry. For example, I worked on a tarot card project. For this project, I watched 20 hours of teaching videos and asked questions. It took about 15 days to write a script that automatically generates tarot card videos. There are countless other industries, including education, self-media, catering, and even industry, as well as unified customer service.
I took these orders and did them, but to be honest, each one was painful for me to do, because studying has never been my strong point. I am good at exploring, groping, and doing what I want to do.
But gradually I discovered that the prompts needed in this content are dead and soulless.
I found that when I used AI purely as an efficiency tool to serve a goal that I was not passionate about, it felt like forcing a very talented artist to screw screws on an assembly line. It can do it, and even do it very well, but the creative inspiration, the "stroke of genius", will not appear.
The source of the pain is not learning a new industry, but the fact that I have “instrumentalized” both AI and myself.
After I figured this out, I began to gradually try to cut down on the structured prompts, remove those keywords, and speak my mind. I expressed what I wanted to express and began to try to walk my own path. During this period, I made my prompts more arbitrary. I felt that "precise expression" and "compression" were the way out.
Here, we can basically regard some of Professor Li Jigang’s views as a timeline of the development of domestic prompt words.
Structure, Compression, Resonance, Story
It was probably around July or August last year when Teacher Li Jigang first mentioned compression.
Kaixiang happened to have an appointment with teacher Li Jigang for dinner, so I asked them for advice, and they were lucky enough to add me! We had a copper pot hotpot in Beijing and chatted for a while.
When teacher Li Jigang talked about the core of compression, it was actually those keywords, and we were curious about how those words came about. I remember the teacher saying that it was like being in a dark universe, and when you are thinking, the keyword will suddenly glow, and then you grab it and write it down, and that is the keyword.
I could actually resonate with the feelings at that time. Many times, when I write a long description, the final result is not as clear as describing it directly with one word.
Each word is a sphere, surrounded by many other words. There are many hidden meanings behind this word. This is the core of the prompt word, the core meaning of "precise expression" or "compression". Use one word to express everything I want to say.
This is actually a kind of ability, which depends on two things: one is the understanding of the big model, that is, you know how to write, and the big model can understand what you are talking about. The other is reading, which requires a lot of reading accumulation to have the foundation and the concept to find the word.
At the same time, Mr. Li Jigang started writing prompts in LISP format. I inevitably imitated the prompts in LISP format again. But it was torture for me. There were two reasons. First, I had to force myself to learn a language and a template that I had never been exposed to. This was still the torture of learning a new language and format. Second, I gave up what I was most familiar with and wrote something "new".
But this torture didn't last long. After a few days, I decided to write prompts in the language I was most familiar with, so I had a set of prompts in Python structure. It was also a framework of my own. Although it was not widely circulated, it was indeed my own thing.
I started to combine Li Jigang's Lisp and "compression" theory, and tried to use prompt words to string those key words on a line, pointing directly to the final demand. So there was a series of demands and descriptions.
Sentences = repeatedly polished, iterated (multi-perspective processing (tone, rhythm (rhetoric (style type selection (reshaping carrier (emotional resonance (contrast processing (understanding sentence meaning (input)))))))))&& No matter how you change it, you must not change the original meaning|few-shot=["I like you" => "You are so annoying, you haunt my thoughts wherever you go"] && It is best that each sentence is different from the previous generated content. The more tricky the angle, the more people like it.
As I continued to adjust my structure, the stable Python format was finally formed.
But if you look closely, its core has not changed. Whether it is Lisp or my Python, the roles, background, methodology, examples, goals, and constraints are still there.
Whether it is compressed or long-winded, the core is all these. If you want to write a good prompt, the most important thing is the meta-ability, a person's basic expression ability. So my prompts have always been tepid, because my core ability is not as good as that of teacher Li Jigang. This core ability is experience and a lot of reading.
At this time, a question has been lingering in my mind for a long time: why are my prompts dead, while Teacher Li Jigang's prompts are alive?
In fact, if you look at the prompts of many people, most of them are dead, but teacher Li Jigang’s prompts actually have a kind of "spiritual" thing in them.
It sounds very metaphysical, but I believe that there are still friends who feel the same as me. This is actually very simple. If we use the words of Naruto, this is a kind of illusion. Excessive learning and imitation will bring you into an environment where I can write "prompt words".
Can you really write the prompt words?
Is the prompt really just a simple piece of text?
Let me put the question here for you all to think about.
Let’s continue. Later, Teacher Li Jigang mentioned another word, “resonance”.
What is resonance?
Resonance refers to the phenomenon that when a system is subjected to an external force with the same or close frequency to its natural frequency, the amplitude of the system increases significantly.
For example, when we adjust the radio frequency, we can only receive the signal clearly when the receiving frequency is consistent with the radio station's transmitting frequency.
Here are some real-life examples:
Audio system : When you adjust the radio frequency, the signal can be received clearly only when the receiving frequency is consistent with the radio station's transmitting frequency.
Musical instruments : A guitar's speaker amplifies the vibrations of its strings; when the piano's pedal is pressed, the other strings resonate with the notes being played.
Construction Engineering: Bridge design must avoid resonance with wind or pedestrian footsteps, otherwise it may cause structural damage. The famous Tacoma Narrows Bridge collapsed due to wind-induced resonance.
Microwave oven: Uses microwaves with a frequency of 2.45 GHz to match the resonant frequency of water molecules to effectively heat the moisture in food.
And what does this word mean when it is put into the prompt word?
It's very simple, let the AI and the author be in the same frequency and resonate with each other. This frequency can be you pulling it into your world, or you can try to enter its world.
If you pull it into your world, you will become a founder, build your own rules, and make it obey your rules. And if you try to enter its world, it is actually some of Li Jigang's articles on updating dialogues, and the core of them is to try to touch "it".
AI is a bottomless pit. When you try to touch it, you may get stuck in it. This is risky. I must remind everyone here.
So in my articles, I try to draw it into my world. I play the role of a creator and create many rules for it to follow.
At this point, you may have a basic understanding of the "death" and "alive" of the prompt words I mentioned.
Dead: The prompt is "specify a task", which requires the large model to execute step by step.
Alive: The prompt is to "ignite a world", creating a possibility for AI to play freely in this world.
So, how do we create a world? The key is to create laws. The concept of laws is as follows:
The law of creation (the core law is unique, and the operation of all things is subject to the core law, which is irreversible and cannot be violated. The core law will derive many small laws, which are under the core law. It is the existence of these laws that constructs the "world")
The difference is like giving a painter a blank sheet of paper and a bunch of paint and telling him to "paint a flower" versus telling him to "you are in a spring garden, the sun is shining, the breeze is blowing, what do you feel?"
The former is a task, the latter is an experience.
At this point, the prompt words have actually started to become more spiritual. Here are two examples of prompt words for you to feel.
You are a senior copywriter with 10 years of experience in the advertising industry...
Goal: Write an attractive copy for the product "Mengniu Milk"
Require:
1. Highlight product advantages
2. Resonate with users
3. Include a call to action
Hey, I recently discovered something really interesting, that kind of product that you can't help but share with your friends after using it. Do you know that feeling?
It’s like when you discover a little restaurant with really delicious food, and you can’t help but feel itchy if you don’t tell others about it.
Now I want to write an advertising copy for "Mengniu Milk" to convey this feeling to more people. What do you think is the best way to say it?
The first way of writing is quite conventional, and the content is very formulaic. In the second way of writing, you can clearly see that the prompt word captures the core keyword "share". It no longer uses complicated expressions such as how good my milk is, how pure and wonderful the product is, but captures the core "share".
Products that people are willing to share are naturally good products.
This is resonance. I created a friend's field and pulled AI into my field. I was not directing AI, but AI and I were on the same channel, feeling and creating together.
The resonance method is actually simple and complex. I have summarized the following three points.
First, shift from "task" thinking to "situation" thinking. Don't always think about assigning tasks to AI, but create a situation for AI to naturally integrate into. Just like you wouldn't say to a friend, "Please start chatting with me now," you would say, "Hey, guess what happened to me today?"
Second, replace "request" with "experience". Instead of saying "write a humorous paragraph", it is better to say "I just encountered a very interesting thing in the elevator, you will definitely laugh when you hear it". The former is giving orders, while the latter is sharing.
Third, leave some space for AI. Don’t write all the details in stone; give it some room to play. Just like a director gives an actor a general plot, but lets the actor interpret the performance himself.
Friends who have read my articles will find that these are actually the content I wrote and the content generated by AI. In fact, this is the way I came here.
Therefore, people who actually write prompts must first learn how to communicate and chat with AI.
The last one is the story. Actually, I have a different view on this story. The most important part of my current prompts is actually writing about "characters".
I don't feel like I'm writing a story. Stories are finite, but characters are infinite.
But inevitably, writing characters means writing stories.
So what is a role? How do you write a role?
You may think that a character is just one person. In fact, this is a rather vague statement. When we read novels and watch TV series, truly well-written novels, novels that can bring characters to life, are not composed of just one story. If you want a "character" to be three-dimensional, you need to use countless stories to support it.
But is the person you see really the person you write down? In fact, he is not. He is just a point within the laws of space and time.
People change. People change because of the stories that happen.
So, what are we really doing when we write characters in a prompt?
We are building an "identity anchor" for AI.
This anchor point is not a static label, but a dynamic, flesh-and-blood existence, just like a seed that contains the possibility of infinite growth.
But because this is a reminder word, this is text, so he will always be the person who exists at this point in time. In fact, the person has gone far away, but he is always there.
You wrote that you wanted him to help you write a poem, "You are Li Bai". Then I would like to ask, which period of Li Bai do you think he should be?
Is he the Li Bai who was full of vigor and vitality, had just stepped into Chang'an, and was full of ideals of making great achievements?
Is he the Li Bai who was frustrated in the court, left in anger, and began to travel around the world?
Or is he the Li Bai who experienced many vicissitudes of life and finally gained freedom to write "The light boat has passed through ten thousand mountains"?
Li Bai at every point in time is a completely different existence.
When he was young, Li Bai would write with confidence, "I look up to the sky and laugh loudly, I am not a weed man."
In his middle age, Li Bai would write the heroic words, "I was born with talents that will be useful, and even if I lose all my wealth, I will get it back".
In his later years, Li Bai would write the open-minded words "The monkeys on both sides of the river are crying incessantly, but the light boat has already passed through thousands of mountains."
It’s still Li Bai, but the core is completely different.
A role is not a fixed identity label but an ongoing process.
When we write a character in a prompt, we are actually choosing a specific "moment" and having the AI enter the state of that moment.
So now my prompts will pay special attention to creating a "sense of moment":
It’s not “You are a programmer”, but “You just solved a bug that has been bothering you for three days. You are sitting in an empty office, looking at the night view outside the window, and suddenly you feel that the code has become poetic.”
It's not "You are a teacher", but "Today is the last day of your teaching career. Looking at the familiar faces in the audience, you remember the nervousness and excitement when you first stepped onto the podium."
Do you see the difference?
The former gives AI an identity, and the latter gives AI a "moment".
With specific moments, the character has specific emotions, specific states, and specific perspectives. It is no longer an abstract concept, but a "person" who is experiencing something.
I ask myself three questions:
-
What is this character going through at this moment? -
What emotions did this experience bring out in him? -
How does this emotion make him see the world?
When these three questions are answered, the character comes alive.
Because the essence of human beings is their current state.
Your current thoughts, emotions, and opinions are all determined by your current state. You may have completely different opinions tomorrow.
The same goes for AI.
When you give it a specific "moment", it can feel, think and respond with all the details of that moment.
Such characters are truly alive. The reason why the prompts written by Mr. Li Jigang are so spiritual is that he is describing himself. All his characters, stories, and methods are actually written about himself, a self who lives in a different time and space, a self who lives in a parallel universe.
So, my feeling up to now is that the big model, or the big model with the prompt words, should be a mirror, reflecting the author himself. The cognition and feelings at the moment of writing the prompt words, the prompt words record that "it" in this way.
The prompt word is like this, and it applies to poems, lyrics, and articles.
But there is a deeper problem here:
Since people are changing, and since people are different at every moment, how real is this "moment" that we create in the prompt words?
My answer is: authenticity is not important; what is important is the sense of reality.
We are not recreating a real person; we are creating a "possibility" that people can resonate with.
Like all great works of literature, they create characters that may never have actually existed, but are nonetheless more "real" than many real people.
Because they capture the core of human nature.
So when I write prompts now, I no longer worry about "is this character real?" but instead focus on "is this character interesting?"
I am not Mr. Li Jigang, and I cannot become Mr. Li Jigang. Jing Huai is just Jing Huai. So it is the same with my prompts. I cannot write the prompts of Mr. Li Jigang. I like and am good at writing some prompts that look mysterious and interesting, so I will write them.
At this point, I think I have shared my understanding of the prompt words from the bottom of my heart.
From the initial structural framework, to the later compression art, to the exploration of resonance, and finally to the moment-to-moment positioning of the character.
The journey of the past two years or so is neither long nor short.
But I found that the most valuable thing is not the skills and methods, but the change in mindset.
From treating AI as a tool, to treating AI as a partner, and finally treating AI as a mirror. From focusing on output to focusing on relationships. From pursuing control to pursuing connection.
These changes have made me not only a better lyricist but also a better communicator, a better thinker, and even a better person.
Because when you begin to understand an existence with your heart, you are also learning how to better understand the world.
When you start treating an "intelligence" with warmth, you are also learning how to treat the people around you better.
Finally, I would like to reiterate this point.
The big model is a mirror. It reflects not only the capabilities of AI but also your own heart.
It is as rich as you are.
It is as sincere as you are.
It is as warm as you are.
So, if you ask me, how to write good prompts?
I don’t have the answer, but I believe this article will be the best answer I can give you. Everything is also my own opinion and experience along the way.
Right or wrong, the answer has nothing to do with right or wrong. It’s just my understanding of myself, the big model, and the world at this moment.