Generative AI / Lesson 4

Prompting Techniques

Master the art of communicating with large language models

Introduction

Prompting is the primary interface for interacting with large language models. The way you structure your prompts can dramatically affect the quality, accuracy, and relevance of the model's responses. This lesson explores fundamental prompting techniques that form the foundation of effective LLM interaction.

Zero-Shot vs Few-Shot Prompting

Zero-Shot Prompting

Asking the model to perform a task without providing examples.

Classify the sentiment:
"This product exceeded my expectations!"

Sentiment:

Few-Shot Prompting

Providing examples to guide the model's behavior.

Classify the sentiment:
"Great service!" → Positive
"Terrible experience" → Negative
"This product exceeded my expectations!" →

Core Prompting Strategies

1. Instruction Following

Clear, explicit instructions help models understand the exact task:

❌ Vague:

"Summarize this article"

✓ Specific:

"Summarize this article in 3 bullet points, focusing on the main findings"

2. Role-Based Prompting

Assigning a role or persona can improve response quality:

You are an experienced data scientist. Explain how neural 
networks work to a business executive with no technical background.

3. Format Specification

Explicitly define the output format you need:

List 5 machine learning algorithms with the following format:
- Algorithm: [name]
  Type: [supervised/unsupervised]
  Use case: [primary application]

4. Constraint Setting

Add constraints to guide the model's output:

  • Length constraints: "in 50 words or less"
  • Style constraints: "using simple language"
  • Content constraints: "without using technical jargon"
  • Format constraints: "as a numbered list"

Advanced Techniques

Temperature Control

Adjust randomness in responses. Lower temperature (0.1-0.5) for factual tasks, higher (0.7-1.0) for creative tasks.

System Messages

Set overall behavior and constraints that persist throughout the conversation.

Prompt Chaining

Break complex tasks into steps, using outputs from one prompt as inputs to the next.

Self-Consistency

Generate multiple responses and select the most consistent or frequent answer.

Common Pitfalls

⚠️

Ambiguous Instructions

Vague prompts lead to unpredictable outputs. Be specific about what you want.

⚠️

Overloading Context

Too much information can confuse the model. Keep prompts focused and relevant.

⚠️

Assuming Knowledge

Don't assume the model knows specific context. Provide necessary background.

Best Practices

  1. 1. Start Simple: Begin with basic prompts and iteratively refine
  2. 2. Be Explicit: State exactly what you want, including format and constraints
  3. 3. Provide Context: Give relevant background information when needed
  4. 4. Use Examples: Few-shot prompting often improves accuracy
  5. 5. Test Variations: Try different phrasings to find what works best
  6. 6. Consider Edge Cases: Test prompts with various inputs

Next Steps

Ready to explore advanced reasoning techniques:

→ Chain-of-Thought Reasoning
← Language Modeling & Next-Token Prediction