Chapter 3: Content Structure - Techniques

In the previous chapter, Content Structure - Introduction, we learned the basics of a prompt (Instruction and Context) and how to adjust settings like Temperature.

Now, we are entering the "Advanced Kitchen."

If Chapter 2 was about learning how to turn on the stove, Chapter 3 is about specific recipes. In the context of AI, we call these recipes Techniques. These are proven strategies to make the AI solve complex logic, math, and reasoning problems.

The Motivation: When the AI "Hallucinates"

Imagine you are asking the AI to solve a tricky riddle.

The Problem: You ask: "I have 3 apples. I eat 2. I buy 5 more. How many do I have?" A basic, un-tuned AI might rush to answer and say: "You have 8 apples." (It simply did 3+5, missing the subtraction).

The Solution: You need a technique to force the AI to slow down. You don't just want an answer; you want the reasoning. This chapter introduces techniques like Few-Shot Prompting and Chain-of-Thought that force the AI to show its work, drastically increasing accuracy.

Key Concepts

This section of the repository (pages/techniques/) covers a spectrum of strategies, moving from simple to complex.

1. Zero-Shot: asking the AI to do something without any examples.
2. Few-Shot: Giving the AI a few examples of what you want before asking your question.
3. Chain-of-Thought (CoT): Asking the AI to "think step-by-step" to solve logic puzzles.
4. Advanced (RAG, ReAct): Methods where the AI looks up external information or uses software tools.

Use Case: Solving Logic Puzzles

Let's stick with our math problem. We want to ensure the AI gets the answer right every time.

Goal: Accurately solve a multi-step word problem.

How to use the Guide:

1. Navigate to the Techniques section.
2. Find the guide on Few-Shot Prompting.
3. Find the guide on Chain-of-Thought.

Technique 1: Zero-Shot (The Default)

This is what most beginners do. You just throw the question at the model.

Q: Roger has 5 tennis balls. He buys 2 more cans of tennis balls. 
Each can has 3 balls. How many tennis balls does he have now?
A:

Result: The AI might guess 7 (5 + 2) because it didn't pay attention to the word "cans."

Technique 2: Few-Shot Prompting (The Upgrade)

The guide teaches you that providing examples (shots) sets a pattern for the AI to follow.

Q: I have 10 socks. I lose 2. How many do I have?
A: 8

Q: Roger has 5 tennis balls. He buys 2 more cans of tennis balls. 
Each can has 3 balls. How many tennis balls does he have now?
A:

Result: Because you gave an example of a math Q&A, the AI understands it needs to perform a calculation. It is now more likely to say 11.

Technique 3: Chain-of-Thought (The Expert Mode)

This is the most popular technique in modern prompt engineering. You explicitly tell the AI to break it down.

Q: Roger has 5 tennis balls... (rest of question)

A: Let's think step by step.
1. Roger starts with 5 balls.
2. 2 cans * 3 balls per can = 6 new balls.
3. 5 + 6 = 11.
The answer is 11.

Result: Perfect accuracy. The phrase "Let's think step by step" acts like a magic spell that improves reasoning.

Under the Hood: File Organization

How is this knowledge organized in the repository? These techniques are stored in the pages/techniques folder.

Unlike the Introduction, which was general, these files are highly specific.

pages/
└── techniques/
    ├── zeroshot.md         # The "No Example" method
    ├── fewshot.md          # The "Example-Based" method
    ├── cot.md              # Chain-of-Thought
    ├── rag.md              # Retrieval Augmented Generation
    └── react.md            # Reason + Act

When you click "Techniques" in the navigation bar of the website, the system is fetching these specific Markdown files.

Sequence Diagram: Learning a Technique

Here is the flow when a user wants to learn how to make an AI "Think":

Implementation Details

Let's look at how the Few-Shot guide is written inside pages/techniques/fewshot.md.

The guide uses a standard format to explain the technique: Definition -> Example -> Tips.

File Content: fewshot.md

# Few-Shot Prompting

**Definition:**
Providing a set of examples (shots) to the model to guide its generation.

**Standard Prompt:**
<Example Input 1> -> <Example Output 1>
<Example Input 2> -> <Example Output 2>
<Target Input> ->

The guide explains that the "shots" act as training data inside the prompt window.

Advanced Technique: RAG (Retrieval Augmented Generation)

The file rag.md describes a more complex concept.

• Analogy: If Zero-Shot is taking a test from memory, RAG is taking an "Open Book" test.
• How it works: You don't just ask the AI a question. You attach the "textbook" (your specific data) to the prompt so the AI can look up the answer.

Conceptual Code for RAG:

# Simplified RAG Prompt Structure
query = "What is the company vacation policy?"
context = "Policy Document: Employees get 20 days off..." # Retrieved from database

prompt = f"""
Context: {context}
Question: {query}
Answer using only the context above:
"""

The guide explains that by adding the Context variable (the retrieved info), you prevent the AI from making things up.

Summary

In this chapter, we opened the toolbox of Content Structure - Techniques.

• We learned: That specific prompting styles change how the AI thinks.
• The Toolbox:
• Zero-Shot: Just asking.
• Few-Shot: Teaching by example.
• Chain-of-Thought: Forcing step-by-step logic.
• The Files: These guides are located in pages/techniques/.

Now that we know how to prompt technically, let's look at what we can build with these skills.

Next Chapter: Content Structure - Applications

Generated by Code IQ