Basic Prompting Techniques

In this lesson, we'll dive into the fundamental techniques that make your prompts effective. Think of these as your building blocks for clear communication with Large Language Models (LLMs).

We'll cover how to give clear instructions, use special characters called delimiters, and control the style and length of the LLM's response.

The most important rule in prompt engineering is to be crystal clear. LLMs are powerful, but they follow your instructions literally. Ambiguity can lead to unexpected or incorrect outputs.

• Be direct: State exactly what you want.
• Avoid jargon: Use simple language unless a specific term is required.
• Assume nothing: Don't expect the LLM to guess your intent.

Let's look at an example. A vague instruction might get you a generic response. A precise instruction guides the LLM to exactly what you need.

Vague Prompt: Write about dogs.
Precise Prompt: Write a 3-sentence fun fact about Golden Retrievers, focusing on their friendly nature.

Notice how the second prompt specifies the topic, length, and focus.

When writing instructions, consider these points to make them effective:

• Task: What should the LLM do (e.g., summarize, explain, generate)?
• Subject: What is the task about?
• Constraints: Any limitations (e.g., length, tone, format)?
• Goal: What is the desired outcome?

The more detail you provide, the better the LLM can meet your expectations.

Sometimes your prompt needs different sections: instructions, input text, examples, etc. Delimiters are special characters or tags that help the LLM understand where one section ends and another begins.

They prevent the LLM from getting confused and misinterpreting your request, especially when you provide a lot of information.

Here are some commonly used delimiters:

• Triple backticks: ```
• Triple quotes: """
• XML-style tags: <example>, </example>
• Hashtags: ###

Choose one that clearly separates your content, often matching the format of your input data.

Let's see how triple backticks can isolate the text you want the LLM to process from your instructions:

Summarize the following text in one sentence:
```The sun was setting, casting long shadows across the quiet meadow. A gentle breeze rustled through the tall grass, and the distant sound of crickets filled the air.```

The LLM now clearly knows which part is the instruction and which is the text to summarize.

You can also instruct the LLM on how its output should be presented. This includes things like:

• Tone: Formal, casual, humorous, professional.
• Format: Bullet points, numbered list, paragraph, JSON.
• Style: Blog post, email, poem, news article.

Being explicit about style helps achieve the desired aesthetic and purpose for your generated content.

For many tasks, the length of the response is important. You can specify this directly in your prompt:

• "Summarize in three sentences."
• "Provide a response no longer than 50 words."
• "List 5 key points."

This helps keep responses concise and focused, or comprehensive when needed, preventing overly long or short outputs.

It's time for a quick check on what you've learned about basic prompting techniques.

Great job! You've learned the essentials of basic prompt engineering:

• Clear Instructions: Be specific and direct in your requests.
• Delimiters: Use characters like ``` to clearly separate sections of your prompt.
• Output Control: Specify desired length and style for the LLM's response.

These techniques are foundational for effective communication with LLMs. In the next lesson, we'll explore understanding LLM capabilities, which will help you set realistic expectations for your prompts.