Practice: Crafting Your First Prompts

Theory is valuable, but practice is where understanding solidifies. Now that you've learned the fundamental concepts of prompts, clear instructions, and providing examples, it's time to interact directly with a Large Language Model. This section provides hands-on exercises to help you craft your first prompts and observe the results.

For these exercises, you'll need access to an LLM. You can use one of the web interfaces or basic API methods that will be discussed in more detail in Chapter 5 ("Using Pre-trained LLMs"). Many free and paid services offer simple chat-like interfaces perfect for getting started. Don't worry about finding the "perfect" LLM right now; the goal is to practice the process of prompting.

Remember, LLMs can sometimes produce unexpected or slightly varied outputs even with the same prompt. Focus on the general structure and intent of the responses rather than precise word-for-word replication.

Getting Started: Your First Interaction

Let's start with the most basic type of interaction: asking a direct question or giving a simple command.

Exercise 1: Simple Question

1. Open the interface to your chosen LLM.
2. In the input area, type the following prompt:

   What is the main function of a CPU in a computer?
   

3. Submit the prompt and observe the response.

Expected Outcome: The LLM should provide a concise explanation of a CPU's role, likely mentioning executing instructions or performing calculations. Notice how a direct question often yields a direct answer.

Exercise 2: Simple Command

1. Clear the previous interaction or start a new one.
2. Type the following prompt:

   List three primary colors.
   

3. Submit the prompt and examine the output.

Expected Outcome: The model should output a list containing red, yellow, and blue. This demonstrates following a straightforward instruction. Did it format it as a numbered list, bullet points, or just comma-separated text? The format might vary unless specified.

Providing Clearer Instructions

As discussed earlier, clarity is important. Let's try refining an instruction.

Exercise 3: Specifying Format

1. Use the instruction from Exercise 2, but modify it to request a specific format.
2. Type the following prompt:

   List the three primary colors as a numbered list.
   

3. Submit and compare the output to Exercise 2.

Expected Outcome: This time, the LLM is more likely to present the colors using numbered points (e.g., 1. Red, 2. Yellow, 3. Blue). This shows how adding specific constraints influences the structure of the response.

Basic Generation and Length Control

LLMs excel at generating text. Let's try a simple creative task and attempt to control the output length.

Exercise 4: Short Sentence Generation

1. Start a new interaction.
2. Type the following prompt:

   Write one sentence describing a rainy day.
   

3. Submit the prompt.

Expected Outcome: The model should generate a single sentence related to rain, perhaps mentioning the sound, the look, or the feeling.

Exercise 5: Expanding the Generation

1. Now, let's ask for a bit more detail.
2. Type the following prompt:

   Write three sentences describing a rainy day.
   

3. Submit and compare the output length to Exercise 4.

Expected Outcome: The model should generate a short paragraph, approximately three sentences long, about a rainy day. While LLMs don't always adhere strictly to exact sentence counts, explicitly requesting a number often guides the output length effectively.

Introduction to Few-Shot Prompting

Providing examples can significantly guide the model, especially for specific formats or tasks it might not immediately grasp.

Exercise 6: Simple Analogy (Few-Shot)

1. Imagine you want the LLM to complete analogies. You can provide an example first.
2. Type the following prompt, including the example:

   Complete the analogy:
   Dog is to bark as cat is to meow.
   Tree is to leaf as flower is to petal.
   Sun is to day as moon is to 
   

3. Submit the prompt.

Expected Outcome: By seeing the pattern (Object : Related Part/Concept), the LLM is guided to complete the final analogy correctly, likely outputting "night". This demonstrates how a simple example (or two) sets the context for the desired task.

Reflection and Experimentation

Go through these exercises, but don't stop there. Try modifying the prompts slightly:

• Change the topic (e.g., ask about planets instead of colors).
• Rephrase the instructions (e.g., "Tell me about..." vs. "Explain...").
• Ask for different output formats (e.g., "Use bullet points," "Write it as a paragraph").
• Try slightly more complex few-shot examples.

Observe how these changes affect the LLM's responses. Note when the model follows your instructions well and when it seems to misunderstand or ignore parts of the prompt. This experimentation is fundamental to developing effective prompting skills. You are learning how to communicate your intent to the model through the text you provide. Keep practicing!