Basic Prompting Techniques

There are a few basic prompting techniques that help us design our prompts more effectively, ensuring the LLM understands our intent accurately.

1. Zero-Shot Prompting

In this technique, we do not provide any examples to the LLM (zero). We rely solely on the model's pre-trained knowledge.

When to use it:

• For simple, straightforward tasks.
• When you want to test the model's inherent knowledge or domain expertise without biasing it.

Example:

🐍Python
Copy
1
2
3
4
5
def add(x,y):
	answer = x + y
	return answer

prompt: Explain what this code does.

2. One-Shot Prompting

In this technique, we provide exactly one example to the LLM.

When to use it:

• When we need a specific output structure (like JSON or a specific list format).
• When the task is slightly ambiguous, and one example clarifies the desired style or logic.

Example:

📦JSON
Copy
1
2
3
4
5
6
7
8
9
10
11
12
13
14
Prompt:
Task: Find weather for a city and output in JSON.

Example 1 (The Shot):
Input: "Weather in London"
Output: {
  "city": "London",
  "weather": "Cloudy",
  "temp": 15
}

Main Request:
Input: "Weather in New York"
Output:

3. Few-Shot Prompting

In this technique, we provide more than one example to the LLM (usually 2-5).

When to use it:

• For complex tasks where a single example isn't enough to capture the pattern.
• When dealing with text that can have more then one meaning.

Example:

📝Markdown
Copy
1
2
3
4
5
6
7
8
9
10
11
12
prompt = """
Task: Extract food items and their prices from the text into a JSON format.

Text: "I bought a burger for $5 and fries for $2."
Output: [{"item": "burger", "price": 5}, {"item": "fries", "price": 2}]

Text: "The salad cost $8, and the soda was $1.50."
Output: [{"item": "salad", "price": 8}, {"item": "soda", "price": 1.5}]

Text: "We ordered a large pizza for $15 and two cokes for $3."
Output:
"""

4. Role Prompting

In this technique, you ask the LLM to act like a specific persona, character, or profession.

When to use it:

• Adjusting Complexity for the Audience: Providing different explanations for different types of audiences.

Example:

📝Markdown
Copy
1
2
3
4
5
6
7
8
9
10
11
# Vibki for developers who know technical concepts
prompt = """
Role: You are a Tech Lead. Your main goal is to mentor and explain difficult
concepts to developers.
"""
# Vibki For Kids 
prompt = """
Role: You are a school teacher. Your main goal is to explain difficult
concepts to kids.
"""

• Tone of Voice: When we want the LLM to speak in a specific style, for example: professional, humorous, etc.

Example:

📝Markdown
Copy
1
2
3
prompt = """
Role: You are a copywriter. Your writing style is humorous. Write a post for LinkedIn.
"""

• Professional Point of View: When we want the LLM to think like a specific professional.

Example:

📝Markdown
Copy
1
2
3
prompt = """
Role: You are a Headhunter with 10 years of experience. Help me write my CV.
"""

• Training and Simulation: When we want to practice a scenario.

Example:

📝Markdown
Copy
1
2
3
prompt = """
Role: You are a Backend Team Leader at Google. Let's simulate a technical interview.
"""

In the next posts, we are going to explore more advanced ways of prompting techniques.