Tips and Best Practices for Creating Effective Prompts for Large Language Models (LLMs)

The idea of a "prompt" is not new. Prompts are used in many areas: in the arts (to inspire a writer or encourage spontaneous speech), in science (to start an experiment), in criminal investigations (to follow or provide clues), and in computer programming (to begin a task based on a specific context). In all these cases, a prompt is a request meant to produce a particular response.

Interacting with large language models (LLMs) is no different. You need to create the right prompt to get the best response. This has led to a new field called prompt engineering, which is the practice of structuring text in a way that a generative AI model can understand and respond to.

What is Prompt Engineering?

Prompt engineering is a practice that involves creating clear, concise, and creative prompts to instruct a large language model (LLM) to perform a task. In simple terms, prompt engineering is the art of communicating with an AI model to get the right results.

In this article, we will look at the best techniques for creating well-structured prompts and the types of prompts that guide an LLM to give the desired response.

Let's discuss the following tips to understand the Prompts better:

1. Be Clear and Detailed with Your Prompts

The clarity and detail in your prompt play a key role in shaping the accuracy and relevance of the model’s response. Being specific helps narrow the model’s focus, ensuring that it generates information that directly addresses your query. This is especially helpful when dealing with complicated topics or when you need in-depth answers.

Example: Consider the prompt "Tell me about climate change" versus "Explain the impact of climate change on coastal cities, focusing on rising sea levels and extreme weather events in the last 20 years."

The first prompt is vague and could lead to a broad, general answer. The second prompt, however, is much more specific, directing the model to focus on a particular aspect of climate change (coastal cities), and providing clear guidelines about the type of information required (rising sea levels and extreme weather), as well as the time frame (the last 20 years). This clarity helps the model deliver a more accurate and detailed response tailored to your needs.

2. Incorporate Context When Necessary

Contextual cues are key to helping the model understand your prompt correctly, especially when the topic can be interpreted in different ways or involves recent changes. By providing context, you guide the model to use the most relevant information, improving the quality of its response.

Example: Consider the prompt "Explain the impact of technology on society" versus "Explain the impact of technology on society in the last 10 years, focusing on social media and its effects on mental health."

The second prompt gives more context by specifying a time frame and particular areas of interest. This helps the model provide a more focused and relevant answer that reflects the specific aspects of technology's impact you're concerned with.

3. Use keywords to Show What You Want

Using specific keywords in your prompt can help the model understand what you're looking for, whether it's a factual answer, a creative piece, or a technical explanation. This makes it easier for the model to give you the right type of response, improving the relevance and quality of the answer.

Example: If you want a factual response, you might use words like "explain" or "describe." For instance: "Explain how photosynthesis works."

If you're looking for a creative piece, you might use words like "write a story" or "imagine." For example: "Write a short story about a dog who learns to talk."

These keywords help the model know whether you're asking for facts or creativity, ensuring a better response.

4. Use Simple and Clear Language

Clear and well-organised prompts are important for getting the best responses from LLMs. A well-structured prompt helps the model understand the order and importance of the information you're asking for, leading to better, more organised answers. Try to avoid confusing or complex language that could make the model lose focus or lead to unclear responses.

Example: Instead of saying, "Can you tell me the impacts of global warming in a lot of places, like the sea and the air?"

A clearer prompt would be: "Explain the effects of global warming on sea levels and air quality."This clear structure helps the model give a more focused and relevant answer.

5. User Repetition

LLMs pay more attention to instructions when they are repeated. Repeating instructions tell the model what is most important and what should be focused on. By repeating key points, you help the model understand what is expected and ensure it follows those instructions carefully.

Example: If you want the model to concentrate on a particular aspect of a historical event, you might say:

"Describe the causes of World War II. Focus on the role of the Treaty of Versailles. Describe the causes of World War II, especially the impact of the Treaty of Versailles."

By repeating the emphasis on the Treaty of Versailles, you guide the model to give more attention to that specific factor in its response.

6. Define Output Structure

When creating prompts for models or APIs, it’s important to specify how you want the output to be structured. This ensures the response fits well with other systems or applications that need to use it.

Example: If you want the model to provide a list of books in a specific format, you might say:

"Provide a list of books in CSV format, including the title, author, and year of publication."

This instruction helps the model generate the information in CSV format, making it easy to use in spreadsheets or other data-processing tools.

Conclusion

Creating effective prompts for large language models requires clarity, context, specificity, and a structured approach. By following best practices like being clear and detailed, using keywords, and defining output structures, you can ensure that the model delivers relevant and accurate responses tailored to your needs. The key is to communicate with the model in a way that guides it toward generating the desired outcome, making prompt engineering an essential skill for maximising the potential of AI interactions.