Module 2 of 18
Prompt Fundamentals
Master the basic components and structure of effective prompts
Types of Prompting
Prompting strategies generally fall into three main categories based on how examples are provided. The "shot" terminology refers to the number of examples you provide in your prompt. Each approach has distinct advantages depending on your use case.
Zero-Shot Prompting
In zero-shot prompting, you provide instructions without any examples of the task. The AI model must understand and complete the task based solely on your description. This is the simplest form of prompting and works well for straightforward tasks or when working with highly capable models.
Prompt
Output
One-Shot Prompting
One-shot prompting involves providing a single example of the input and desired output, followed by a new input that needs a response. This approach helps the AI understand the pattern you want it to follow through demonstration rather than just description.
Prompt
Output
Few-Shot Prompting
Few-shot prompting expands on one-shot by providing multiple examples of the desired behavior before asking the AI to perform the task. This approach is particularly effective for complex tasks, specific formats, or when you need the AI to understand nuanced patterns.
Prompt
Output
The number of examples ("shots") you provide significantly impacts how well the AI understands your task. More complex or specialized tasks generally benefit from more examples.
The Prompt Engineering Process
Effective prompt engineering is an iterative process. Expect to cycle through these steps multiple times to optimize your prompts. The best prompt engineers maintain a systematic approach while being willing to experiment.
- 01
Define your objective
Clearly articulate what you want to achieve. Are you generating content, extracting information, transforming text, or solving a problem? - 02
Choose your prompting strategy
Decide whether zero-shot, one-shot, or few-shot prompting is most appropriate for your task, based on its complexity and specificity. - 03
Draft your prompt
Create your initial prompt, incorporating the necessary components (task, context, examples, constraints, etc.) as needed. - 04
Test and evaluate
Run your prompt and assess whether the output meets your needs. Does it follow instructions? Is it accurate? Does it have the right format? - 05
Refine iteratively
Based on the results, adjust your prompt. You might need to add more examples, clarify instructions, or modify constraints to get better outputs.
Prompt engineering is both an art and a science. While systematic testing helps, developing an intuition for what works comes with experience across different models and tasks.
When to Use Each Approach
Zero-Shot Best For
Simple, straightforward tasks. When working with the latest, most capable models. Tasks common in the model's training data. When you need quick results with minimal prompt engineering.
One-Shot Best For
Tasks that benefit from a clear pattern demonstration. When you have limited context window space. Formats easy to understand from a single example. Balancing performance with prompt simplicity.
Few-Shot Best For
Complex or specialized tasks. Demonstrating patterns with subtle variations. When consistent formatting is critical. Domain-specific tasks requiring specialized terminology.
Activities
Activity 1
15-20 min · Beginner
Zero-Shot Challenge
Design five zero-shot prompts for different tasks and test their effectiveness.
Activity 2
25-35 min · Intermediate
One-Shot vs. Few-Shot Comparison
Take three different tasks and implement both one-shot and few-shot approaches. Compare the results and analyze the differences.
Activity 3
30-40 min · Advanced
Task-Specific Prompting
Create specialized prompts for three specific domains (e.g., legal, medical, technical) and test how different shot approaches affect domain-specific outputs.