Skip to main content

How to Write Better Prompts for LLMs

Writing effective prompts is crucial for getting the best results from Large Language Models (LLMs). This guide will help you understand how to craft better prompts to improve the quality of the generated outputs

Understanding Prompts​

A prompt is the input you provide to an LLM to generate a response. It typically includes instructions or questions that guide the model on what kind of text to produce. A well-crafted prompt can significantly enhance the relevance, coherence, and quality of the generated content.

Key Principles for Writing Effective Prompts​

1. Improve Clarity​

Ensure your prompt is clear and easy to understand. Avoid vague language and be specific about what you want. The clearer your instructions, the better the model can understand and respond.

Example:

  • Poor: Tell me about history.
  • Better: Provide a brief overview of the American Civil War, including the main causes and outcomes.

2. Add Context​

Include relevant context to help the model understand your request. The more context you provide, the better the response will be, as the model can generate more accurate and relevant information.

Example:

  • Poor: Explain the benefits.
  • Better: Explain the benefits of regular exercise for cardiovascular health, focusing on how it improves heart function and reduces the risk of heart disease.

3. Increase Specificity​

Be specific about what you want the model to do. Mention the format, length, or style if necessary. Detailed instructions help the model generate content that meets your exact needs.

Example:

  • Poor: Write a story.
  • Better: Write a 200-word short story about a young girl discovering a hidden talent in painting

4. Maintain Conciseness​

Keep the prompt concise while including all necessary information. Long prompts can be confusing and may reduce the model's effectiveness. Focus on the essential details to avoid overwhelming the model.

Example:

  • Poor: Can you tell me something about the history and the significant events that took place during the early days of the United States of America, focusing particularly on the Revolutionary War?
  • Better: Summarize the key events of the American Revolutionary War.

5. Provide Clear Instructions​

Provide clear instructions on the desired tone, style, or perspective if needed. This helps the model generate content that matches your expectations.

  • Poor: Describe a sunset.
  • Better: Describe a sunset from the perspective of an artist using vivid and poetic language.

Advanced Techniques​

1. Use Examples​

Providing examples of the desired output can help guide the model more effectively. Examples serve as a reference for the model, making it easier to understand and follow your instructions.

Example: Translate the following English sentence to Spanish: 'Hello, how are you?' Example: 'Hello' in Spanish is 'Hola'.

2. Iterative Refinement​

Start with a general prompt and iteratively refine it based on the generated outputs to improve the results. Refining prompts over multiple iterations helps in achieving the best possible response.

Example:

  • Initial Prompt: Describe a futuristic city.
  • Refined Prompt: Describe a futuristic city in the year 2200 with advanced technology and sustainable living practices.

3. Break Down Complex Requests​

Break down complex requests into multi-part prompts to guide the model step-by-step. This approach helps in managing the complexity and ensures that each part of the request is addressed thoroughly.

Example: First, describe the setting of a medieval castle. Then, introduce the main character who is a knight on a quest.

Fixing and Iteratively Improving Existing Prompts​

If you have existing prompts that aren't yielding the desired results, you can use the following steps to improve them:

1. Identify Issues​

Review the responses to identify what is not working. Are they too vague, off-topic, or missing details? Understanding the problem is the first step toward fixing it.

2. Simplify and Clarify​

Simplify the prompt and remove any ambiguous language. Ensure the instructions are clear and concise. A straightforward prompt is easier for the model to understand and follow.

Example:

  • Original: Write something about space.
  • Improved: Write a 150-word paragraph about the potential for human colonization of Mars.

3. Add Context​

Provide additional context to help the model understand the request better. More context can lead to more accurate and relevant responses.

Example:

  • Original: "Describe the benefits."
  • Improved: "Describe the benefits of a plant-based diet for heart health."

4. Be Specific​

Specify the format, style, or length if needed. Clear specifications help the model produce outputs that align with your expectations.

Example:

  • Original: "Explain photosynthesis."
  • Improved: "Explain the process of photosynthesis in plants in 3-4 sentences suitable for a 10-year-old."

5. Test and Refine​

Test the refined prompt and review the output. Make further adjustments as needed to get the desired result. Continuous refinement ensures the prompt evolves to produce the best possible response.

Example:

  • Initial Prompt: "Describe a futuristic city."
  • Refined Prompt: "Describe a futuristic city in the year 2200 with advanced technology and sustainable living practices."
  • Further Refined Prompt: "Describe a futuristic city in the year 2200, focusing on its advanced technology, green spaces, and sustainable living practices in 100 words."

Practical Examples​

1. Simple Instruction​

Prompt: "List the main ingredients in a traditional Italian pizza."

Output: "The main ingredients in a traditional Italian pizza are flour, water, salt, yeast, olive oil, tomato sauce, and mozzarella cheese."

2. Detailed Description​

Prompt: "Describe a day in the life of an astronaut on the International Space Station."

Output: "A day in the life of an astronaut on the International Space Station begins with a wake-up call from Mission Control. After a healthy breakfast, the astronaut conducts scientific experiments, maintains the spacecraft, exercises for two hours, and communicates with family back on Earth..."

3. Creative Writing​

Prompt: "Write a short poem about the changing seasons."

Output: "The leaves turn gold, then fall away / As winter's chill comes into play / Spring awakens, flowers bloom / Summer dances in the warm sun's room."

Writing better prompts is an essential skill for maximizing the effectiveness of LLMs. By being clear, specific, and providing sufficient context, you can guide the model to produce high-quality, relevant outputs. Experiment with different techniques and refine your prompts iteratively to achieve the best results.