Chapter 1: Fundamentals of Prompt Engineering - Introduction to Prompt Engineering

What is Prompt Engineering

Prompt Engineering is the technique of designing effective inputs (prompts) to elicit desired outputs from Large Language Models (LLMs).

While LLMs like ChatGPT, Claude, and Gemini are extremely powerful, the appropriate "way of asking questions" and "giving instructions" is crucial to maximizing their capabilities. Even for the same question, the quality of the response can vary greatly depending on how the prompt is written.

🎯 Why Prompt Engineering Matters

Quality Improvement: Obtain more accurate and detailed responses
Efficiency: Reduce trial-and-error time
Reproducibility: Consistently obtain outputs of the same quality
Cost Reduction: Optimize API token usage
Expanded Applications: Handle more complex tasks

Six Characteristics of Good Prompts

Effective prompts have the following six characteristics:

1. Clarity

Eliminate ambiguity and clearly indicate what is being requested.

❌ Bad Example

Tell me about AI

Problem: Scope too broad, unclear what information is desired

✅ Good Example

Explain the differences between supervised learning and unsupervised learning
in machine learning, including concrete examples, in three key points.

Improvement: Topic, format, and level of detail are clear

2. Specificity

Specifically specify the expected output format, length, and style.

❌ Bad Example

Write an email

✅ Good Example

Create an email to a client regarding sending a proposal.

【Requirements】
- Recipient: Mr. Tanaka (Department Manager, ABC Corporation)
- Purpose: Send proposal for new system implementation
- Tone: Formal and polite
- Length: 200-300 words
- Include: Greeting, key points of proposal, next steps

3. Context

Provide background information and prerequisites for the task.

💡 Prompt Example: Including Context

You are an experienced Python programmer.
You are creating programming materials for beginners.

Explain Python list comprehensions in the following format:
1. Basic concept (within 100 words)
2. Simple code example
3. Common use cases (3 examples)
4. Common mistakes for beginners

4. Constraints

Explicitly state limitations and requirements for the output.

💡 Prompt Example: With Constraints

Create a product description with the following constraints:

【Constraints】
- Length: Within 150 words
- Tone: Casual and friendly
- Required keywords: "eco-friendly", "durability", "value"
- Avoid: Exaggerated claims, technical jargon
- Target audience: Working professionals in their 20s-30s

【Product】Stainless steel tumbler

5. Examples

Clarify format and quality standards by showing examples of expected output.

💡 Prompt Example: Few-shot Learning

Perform sentiment analysis following the examples below.

Example 1:
Input: Today was the best day ever!
Output: Positive (joy, satisfaction)

Example 2:
Input: Plans got cancelled due to rain, disappointing
Output: Negative (disappointment, sadness)

Example 3:
Input: I wonder if it will be sunny tomorrow
Output: Neutral (anticipation, uncertainty)

Now analyze the following sentence:
Input: A new project is starting but I'm anxious

6. Step-by-step Instructions

Break down complex tasks into step-by-step instructions.

💡 Prompt Example: Step-by-step Instructions

Create a data analysis report following these steps:

【Step 1】Review data overview
- Dataset size (rows, columns)
- Data types of each column
- Presence of missing values

【Step 2】Calculate basic statistics
- Mean, median, standard deviation for numeric columns
- Frequency of categorical columns

【Step 3】Detect anomalies
- Identify outliers
- Infer their causes

【Step 4】Conclusions and recommendations
- Key findings (3 points)
- Suggested next actions

Zero-shot and Few-shot Learning

Zero-shot Learning

Zero-shot is a method of directly instructing tasks without examples. It's suitable for simple tasks or general questions.

💡 Zero-shot Prompt Example

Translate the following sentence into English:

"Machine learning is the ability of computers to learn from data and perform tasks."

Few-shot Learning

Few-shot is a method of performing tasks after showing several examples. It's effective when formats are specialized or when you want to clarify quality standards.

💡 Few-shot Prompt Example (2-shot)

Convert product names to shortened forms following the examples:

Example 1:
Input: Super Premium High-Function Multi-Purpose Hybrid Vacuum Cleaner
Output: Premium Vacuum

Example 2:
Input: Ultra Slim Design Wireless Charging Compatible Smartphone
Output: Slim Smartphone

Now convert the following:
Input: Extra Long Battery Portable Bluetooth Speaker

One-shot vs Few-shot vs Many-shot

Method	Number of Examples	Advantages	Disadvantages	Best Suited For
Zero-shot	0	Simple, token-saving	May have lower accuracy	General tasks
One-shot	1	Can show format	Insufficient for pattern learning	Simple format specification
Few-shot	2-5	Pattern learning, high accuracy	Increased tokens	Complex tasks
Many-shot	6+	Highest accuracy	Heavy token consumption	When highest accuracy is essential

Chain of Thought (CoT) Prompting

Chain of Thought is a technique that encourages step-by-step reasoning in LLMs. By adding the instruction "think step by step," you can obtain more logical and accurate responses.

Basic CoT Prompt

❌ Without CoT

Q: There are 5 apples and 3 oranges.
We used 2 apples and 1 orange.
How many fruits are left in total?

Problem: Calculation process unclear, prone to errors

✅ With CoT

Q: There are 5 apples and 3 oranges.
We used 2 apples and 1 orange.
How many fruits are left in total?

Think step by step.

Expected Output:

Step 1: Check initial fruit count
- Apples: 5
- Oranges: 3
- Total: 5 + 3 = 8

Step 2: Subtract used fruits
- Apples: 5 - 2 = 3
- Oranges: 3 - 1 = 2

Step 3: Calculate remaining total
- Remaining: 3 + 2 = 5

Answer: 5 fruits

Few-shot CoT Prompt

💡 Few-shot CoT Prompt Example

Solve the problem following the example below:

Example:
Q: At a cafe, I ordered 2 coffees (450 yen each) and 1 cake (600 yen).
I'm paying with two 1000 yen bills. How much change will I get?

A: Let's think step by step.
1. Cost of 2 coffees: 450 yen × 2 = 900 yen
2. Total amount: 900 yen + 600 yen = 1500 yen
3. Payment amount: 1000 yen × 2 = 2000 yen
4. Change: 2000 yen - 1500 yen = 500 yen
Answer: 500 yen

Now solve this problem:
Q: At a bookstore, I bought 3 books (1200 yen each) and 2 magazines (800 yen each).
I'm paying with a 5000 yen bill. How much change will I get?

When CoT is Particularly Effective

Mathematical Reasoning: Calculation problems, logic puzzles
Complex Decision Making: Judgments considering multiple factors
Causal Analysis: Explaining cause-and-effect relationships
Multi-step Problem Solving: Tasks requiring multiple steps
Debugging: Identifying error causes in code or processes

Practical Prompt Templates

Here are ready-to-use practical prompt templates.

Template 1: Task Execution Type

【Role】You are an expert in [field of expertise].

【Task】[Specific task description]

【Input】
[Data or information to process]

【Output Format】
- [Format 1]
- [Format 2]
- [Format 3]

【Constraints】
- [Constraint 1]
- [Constraint 2]

【Example】(Optional)
[Example of expected output]

Template 2: Analysis Type

Analyze the following data:

【Data】
[Data to analyze]

【Analysis Perspectives】
1. [Perspective 1]
2. [Perspective 2]
3. [Perspective 3]

【Required Output】
1. Key findings (3 points)
2. Observable trends from data
3. Recommended next actions

Analyze step by step.

Template 3: Creative Type

Create content with the following conditions:

【Type】[Blog post / Email / Presentation, etc.]

【Theme】[Main topic]

【Target Audience】
- Age range: [Age range]
- Knowledge level: [Beginner / Intermediate / Expert]
- Interests: [Interests or challenges]

【Tone】[Formal / Casual / Professional, etc.]

【Structure】
1. [Section 1]
2. [Section 2]
3. [Section 3]

【Length】[Word count or character count]

【Required Elements】
- [Element to include 1]
- [Element to include 2]

Template 4: Code Generation Type

Create Python code with the following specifications:

【Function】[Description of function to implement]

【Input】[Function arguments or input data]

【Output】[Expected output]

【Requirements】
- Programming language: Python 3.8+
- Libraries to use: [Library name]
- Error handling: Required
- Comments: Japanese docstrings for each function

【Example】
Input example: [Concrete example]
Expected output: [Result example]

Provide code and usage example.

Common Failure Patterns and Improvements

Failure Pattern 1: Vague Instructions

❌ Before Improvement

Write a business email

✅ After Improvement

Create an email introducing a new product to a business partner.

【Requirements】
- Recipient: Sales Manager
- Purpose: Introduce new product "XYZ" and obtain business meeting appointment
- Tone: Formal yet friendly
- Length: About 300 words
- Include: Greeting, product features (3 points), specific date proposal

Failure Pattern 2: Multiple Tasks at Once

❌ Before Improvement

Review this code, fix bugs,
improve performance, and write documentation too

✅ After Improvement

【Task 1】First, identify bugs in this code:
[Code]

Once bugs are found, we'll proceed to the next task.

Then request tasks separately in sequence

Failure Pattern 3: Lack of Context

❌ Before Improvement

Improve this text:
"Developed a product"

✅ After Improvement

【Background】
Creating a press release for a startup.
Target audience is investors and media representatives.

【Text to Improve】
"Developed a product"

【Improvement Direction】
- More specific and impressive expression
- Emphasize product innovation
- Capture reader's interest
- Professional tone

Failure Pattern 4: Unspecified Output Format

❌ Before Improvement

Explain Python's main features

✅ After Improvement

Explain Python's main features in the following format:

【Output Format】
For each feature:
1. Feature name
2. Brief description (within 50 words)
3. Code example (within 5 lines)
4. Usage scenarios

【Target Features】
- List comprehension
- Decorators
- Generators

Explain in a beginner-friendly manner.

Best Practices for Prompt Design

💡 Practical Tips

Iterative Improvement: Don't expect perfection on first attempt, improve while reviewing outputs
Templatize: Create templates for frequently used prompts and reuse
Version Control: Record and manage effective prompts
A/B Testing: Try multiple prompts and select the best one
Leverage Feedback: Learn from LLM outputs and improve prompts

Prompt Design Checklist

☐ Is the task objective clear?
☐ Have you provided necessary context information?
☐ Have you specifically specified output format?
☐ Have you stated constraints?
☐ Have you broken down complex tasks into steps?
☐ Have you included examples when necessary?
☐ Have you specified tone and style?
☐ Have you clarified target audience and usage scenario?

Chapter Summary

🎯 Key Points

Clarity is paramount: Eliminate ambiguity and give specific instructions
Six characteristics: Clarity, specificity, context, constraints, examples, step-by-step instructions
Zero-shot vs Few-shot: Choose based on task complexity
Chain of Thought: Add "think step by step" for complex reasoning
Leverage Templates: Reuse effective patterns for efficiency
Iterative Improvement: Don't aim for perfection initially, improve while reviewing outputs

Exercises

Practice the knowledge you've learned with the following exercises.

Exercise 1: Improve Prompts (Difficulty: ★☆☆)

Problem: Improve the following vague prompt based on the six characteristics.

Tell me a recipe

Hint: Clarify dish name, number of servings, cooking time, difficulty, and output format.

Exercise 2: Create Few-shot Prompt (Difficulty: ★★☆)

Problem: Create a Few-shot prompt for sentiment analysis of product reviews (Positive/Negative/Neutral). Include at least 3 examples.

Exercise 3: Chain of Thought Prompt (Difficulty: ★★☆)

Problem: Create a CoT prompt to solve the following problem.

Problem: A company has 120 employees.
Of these, 40% work in sales, 30% in development, and the rest in administration.
If the development department increases by 10 people, what percentage will the development department represent?

Exercise 4: Task Execution Prompt (Difficulty: ★★★)

Problem: Design a comprehensive prompt template for "replying to customer complaint emails." Include role, task, constraints, and output format.

Exercise 5: Practical Application (Difficulty: ★★★)

Problem: Create one prompt that can be used in your actual work or studies, test it with an LLM, evaluate the results, and list three areas for improvement.

Next Steps

In Chapter 1, you learned the fundamentals of prompt engineering. In the next chapters, you'll learn more advanced techniques:

Chapter 2 (Coming Soon): Basic techniques like Role Prompting, Context Setting, Output Format
Chapter 3 (Coming Soon): Applied techniques like Tree of Thought, Self-Consistency
Chapter 4 (Coming Soon): Task-specific optimization methods (summarization, translation, code generation, etc.)
Chapter 5 (Coming Soon): Practical projects and prompt library development

← Series Top Chapter 2 (Coming Soon) →

References

OpenAI. (2023). GPT Best Practices
Wei, J., et al. (2022). Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
Brown, T., et al. (2020). Language Models are Few-Shot Learners
Anthropic. (2024). Prompt Engineering Guide

Update History

2025-12-01: v1.0 Initial release

What is Prompt Engineering

🎯 Why Prompt Engineering Matters

Six Characteristics of Good Prompts

1. Clarity

❌ Bad Example

✅ Good Example

2. Specificity

❌ Bad Example

✅ Good Example

3. Context

💡 Prompt Example: Including Context

4. Constraints

💡 Prompt Example: With Constraints

5. Examples

💡 Prompt Example: Few-shot Learning

6. Step-by-step Instructions

💡 Prompt Example: Step-by-step Instructions

Zero-shot and Few-shot Learning

Zero-shot Learning

💡 Zero-shot Prompt Example

Few-shot Learning

💡 Few-shot Prompt Example (2-shot)

One-shot vs Few-shot vs Many-shot

Chain of Thought (CoT) Prompting

Basic CoT Prompt

❌ Without CoT

✅ With CoT

Few-shot CoT Prompt

💡 Few-shot CoT Prompt Example

When CoT is Particularly Effective

Practical Prompt Templates

Template 1: Task Execution Type

Template 2: Analysis Type

Template 3: Creative Type

Template 4: Code Generation Type

Common Failure Patterns and Improvements

Failure Pattern 1: Vague Instructions

❌ Before Improvement

✅ After Improvement

Failure Pattern 2: Multiple Tasks at Once

❌ Before Improvement

✅ After Improvement

Failure Pattern 3: Lack of Context

❌ Before Improvement

✅ After Improvement

Failure Pattern 4: Unspecified Output Format

❌ Before Improvement

✅ After Improvement

Best Practices for Prompt Design

💡 Practical Tips

Prompt Design Checklist

Chapter Summary

🎯 Key Points

Exercises

Exercise 1: Improve Prompts (Difficulty: ★☆☆)

Exercise 2: Create Few-shot Prompt (Difficulty: ★★☆)

Exercise 3: Chain of Thought Prompt (Difficulty: ★★☆)

Exercise 4: Task Execution Prompt (Difficulty: ★★★)

Exercise 5: Practical Application (Difficulty: ★★★)

Next Steps

References

Disclaimer