[Generative AI Passport] What exactly is prompt engineering?

📱 When you consult Gemini or ChatGPT, the replies feel vague.
🎨 You asked an image generation AI for something, but the picture that came out was not what you imagined.
💰 I want to master AI and cut my work time in half!

If you feel that way, now is the time to learn “prompt engineering”!✨

If you master this, AI will evolve from “just a convenient tool” into “the ultimate right-hand assistant that perfectly understands your intentions.”🚀

1. Understand the basics

① [Prompt Engineering]

✅ Pronunciation: Prompt Engineering
✅ English name: Prompt Engineering
✅ Japanese name: Instruction Optimization (or Prompt Design)

[Related terms you should learn together]
・Prompt (instructions given to an AI)
・LLM (Large Language Model. The “brain” of AIs like ChatGPT)
・Hallucination (when an AI tells lies in a plausible way)

② Meaning (definition) and an analogy

[In a nutshell?]
✅ It is the “trick to giving instructions” to draw out the “best answer” from an AI!

[A relatable analogy]
Suppose you have a “super naive person who is the best in the world at knowledge and empathy, but sometimes cannot read the room” assigned to work next to you.

What would happen if you just told him, “Make some materials that look good”? He might bring you a blank sheet of paper with just the words “look good” written on it, or he might write a 10-page summary of today’s events 😱

But what if you told him this? “Create materials summarizing company rules for new employees. Please use bullet points, keep it to one page, and use a friendly tone!”

If you do this, he will do a perfect job, right? This “technique of giving specific instructions so the other person doesn’t get lost” is exactly what prompt engineering is!📱✨

[Relationship with AI]
AI is incredibly knowledgeable and empathetic, but it is bad at “reading the room.” By giving specific and structured instructions instead of vague ones like “make it look good,” you will be able to draw out 100% of its capabilities.

[Common misconceptions]
Misconception: It is a specialized technique that you cannot do without learning difficult programming languages.
Truth: It is a “communication skill” that uses everyday “language” to organize and convey information in a way that is easy for AI to understand.

2. Understand the history

Let’s trace the story of how giving instructions to AI has evolved.

①: “I want to talk to AI normally!” (The challenge at the time)
Old computers would not work unless you wrote special, difficult code. However, humans wanted to be able to ask for things by saying “do this” in more natural language.

②: “In that case, we should just have it learn ‘word connections’ from a massive amount of text!” (The emergence of new technology)
At one point, a “Large Language Model (LLM)” that had learned a vast amount of text from the internet appeared. Because of this, just by giving instructions in normal Japanese, AI became able to respond with words like a human.

③: “But the answers are vague or it tells lies!” (The exposure of the challenge)
It turned out that just talking to it normally would lead the AI to give random answers or forget instructions. We realized, “It understands language, but there is a trick to using it well.”

(4): “Can’t we structure instructions more strategically?” (Evolution toward a finished form)
This is where specific “instruction patterns”—such as “assigning a role,” “providing examples,” and “having the AI think step-by-step”—began to be researched. This was the birth of “prompt engineering.” It is about transforming a vague “request” into a “blueprint” that the AI can follow without confusion. This is the true value of prompt engineering.

3. Understanding through application

In real life and work, it is used in ways like this!

✅ Application Example 1: [Creating a travel plan]
Instead of saying, “Make a plan for a trip to Kyoto”…
You instruct it by saying, “You are a professional travel planner. Please propose a 2-day, 1-night Kyoto sightseeing course for two women in their 20s, with a budget of 30,000 yen, including three Instagram-worthy cafes, in a table format with time allocations.”
👉 With this, the perfect plan for you is created instantly.

✅ Application Example 2: [Creating an SNS post]
Instead of saying, “Write an apology email”…
“Write an apology email to a business partner informing them that the delivery date will be delayed by one day. The recipient has a very strict personality, so please use very polite business honorifics that emphasize sincerity and future countermeasures without making excuses!”
👉 By specifying even the recipient’s personality, you get a perfect message that won’t cause offense.

4. Understanding through practice

Question 1 [Most frequently asked question]

Which of the following is the most appropriate purpose of prompt engineering?

A: To rewrite all of the AI’s training data and add new knowledge.
B: To refine the prompts input into the AI to obtain the desired output.
C: To efficiently write the programming code used to operate the AI.
D: To completely eliminate the AI’s ability to lie (hallucination).

Question 2 [Trick question]

Which of the following statements regarding prompt engineering is inappropriate?

A: You can improve the accuracy of responses by assigning a specific “role” to the AI.
B: There is a method (Few-shot prompting) where you provide several examples of correct answers to guide the response.
C: To perform prompt engineering, it is mandatory to learn a programming language such as Python.
D: It is effective to make instruction sentences specific and set constraints (such as character count or format).

5. Understanding through explanation

Explanation for Question 1

[Correct Answer] ✅ B

・Reason why A is wrong: Rewriting training data is a different technique called “fine-tuning.” Prompt engineering only changes “how you communicate” with a pre-trained AI.
・Reason why B is correct: Exactly! The goal is to optimize the input (prompt) to get the desired result (output).
・Reason why C is wrong: It is not about writing programming code. It is a technique for giving instructions in natural language (such as Japanese).
・Reason why D is wrong: While you can “reduce” hallucinations, it is difficult to completely “eliminate” them with current technology due to the way AI works.

Explanation for Question 2

[Correct Answer] ✅ C

・Reason why A is correct: If you assign a role, such as “You are a professional marketer,” the AI will answer from that perspective, which improves accuracy.
・Reason why B is correct: Showing no examples is called “Zero-shot,” and showing a few examples is called “Few-shot”; both are very effective techniques.
・Reason why C is inappropriate (Correct answer): This is the trick! Prompt engineering is a “language” technique, so programming languages are not required. Anyone can try it.
・Reason why D is correct: By adding constraints such as “within 300 characters” or “in bullet points,” the AI will not get confused, resulting in a more usable response.

6. Understanding just before the test

Check this right before the exam! It’s a super condensed summary.🔥

[Most Frequent/Frequent Points]
Question 1: What is a prompt?
Answer 1: It is an “instruction” given to an AI.

Question 2: What is the purpose of prompt engineering?
Answer 2: To efficiently elicit desired responses from an AI.

Question 3: What is the basic method for improving response accuracy?
Answer 3: Providing specific instructions and setting constraints (format and character count).

Question 4: What is specifying a “role”?
Answer 4: Defining the AI as “You are an expert in XX” to set the direction of the response.

Question 5: What is Chain-of-Thought?
Answer 5: Telling the AI to “think step-by-step” to elicit a logical thought process.

[Trick Questions]
Question 6: Are programming skills required?
Answer 6: No. It is completed using natural language (everyday words).

Question 7: Can hallucinations be reduced to zero?
Answer 7: No. You can reduce them, but it is impossible to eliminate them completely.

Question 8: Is it about rewriting the AI model itself?
Answer 8: No. It is a technique for refining the “input” while keeping the model as is.

[Caution: Confusion/Comparison]
Question 9: What is the difference between a “prompt” and “fine-tuning”?
Answer 9: A prompt is “refining how you communicate,” while fine-tuning is “modifying the brain through additional training.”

[Parent and Child/Structural Relationship]
Question 10: What factors determine the output results of generative AI?
Answer 10: It is determined by the multiplication of “AI model performance” and “prompt quality.”

Question 11: What are the components of prompt engineering?
Answer 11: It consists of “role” + “background” + “instructions” + “constraints” + “output format,” etc.

7. General Review/Summary

Prompt engineering is not a difficult engineering skill, but rather something like “communication ability to refine how you convey things to others.”😊

Think of the AI as an “excellent assistant” and give it careful instructions. Just by adopting this perspective, your AI utilization ability will skyrocket!

In the exam, trick questions like “Is programming necessary?” or “Does it rewrite the model?” are common, so please be careful about those.

You are no longer the one being pushed around by AI, but the “pilot” who masters it. Grab your passing grade with confidence!❤️

If you found this article helpful, please support me with a “Like”!✨
Your support is the energy that creates the next explanation! ✨

Originally Appeared Here