Skip to main content

Prompt Engineering

What is a Prompt?

From a user's perspective, when interacting with an AI, a prompt is the message or instruction you provide to the AI to execute a task or answer a question. Essentially, it’s the command or request that initiates the interaction.

For example, when you type a question or ask the AI to act, like "Explain what a prompt is", that is the prompt. The AI then processes this request and responds based on its information.

Here are some examples of prompts you might give to an AI:

  • "Translate this text into English."
  • "Write a 100-word summary on artificial intelligence."
  • "Create a title for this issue on GitHub."
  • "Explain the concept of Clean Code."

The quality and clarity of the prompt directly influence the AI's response. The clearer and more specific the prompt, the better the AI's response will be.

Prompts can vary in format and complexity, including direct questions, specific instructions, or detailed descriptions, depending on the desired outcome.

1. What is Prompt Engineering?

It is creating and refining prompts to obtain more accurate and relevant responses from language models like AI. It involves carefully choosing the prompt's words, structure, and context to guide the AI model (such as GPT) in generating the desired output.

For example:

If you want the AI to generate a text summary, a simple prompt like "Summarize this text" may yield a generic response. However, by applying prompt engineering, you can be more specific: "Summarize the following text in 100 words, highlighting key points about sustainability." This helps the AI better understand your expectations, resulting in a response more aligned with your goal.

1.1 Objective

Prompt Engineering aims to enhance interactions with AI models, ensuring that responses meet expectations.

1.2 Importance

  • Improves response quality.
  • Reduces ambiguity in interactions.
  • Increases efficiency when using AI models.

2. How Does Prompt Engineering Work?

Prompt Engineering involves formulating questions or commands that guide the AI model in generating specific responses. This may include:

  • Defining context.
  • Specifying the response format.
  • Limiting the scope of the answer.

2.1 Example of a Simple Prompt

"Explain the concept of supervised learning in machine learning."

2.2 Example of a Detailed Prompt

"Explain the concept of supervised learning in machine learning, including examples of popular algorithms and their practical applications."

2.3 What is the Ideal Length of a Prompt?

  • Very short prompts: Lack enough context.

  • Overly long prompts: It may introduce unnecessary confusion. Longer writing isn't always better.

3. Best practices in Prompt Engineering

3.1 Be clear and specific

  • Use clear, objective language.
  • Avoid ambiguities.
  • Specify what you expect as a response.

Example:

"List three benefits of using containers in software development."

3.2 Style: Provide context

  • Give the model enough information to understand the scenario.
  • Contextualize the question to avoid generic responses.
  • Specify the language in which you want the response.

Example:

"In an agile development environment, what are the main benefits of using containers?"
"Help me explain the concept of AI with fewer technical terms for my MBA project."
"Explain this to me as if it were a topic in a children’s educational program, teaching elementary students."
"I am a software engineer using large language models for summarization. Summarize the following text in fewer than 250 words."

3.3 Formatting: Define the response format

  • Specify the desired format, like lists, paragraphs, or examples.

Example:

"Explain the concept of microservices and provide three examples of its application in tech companies."
"Return as JSON."

3.4 Use examples

  • Include examples in the prompt to guide the model toward more relevant answers.

Example:

"Explain the concept of DevOps and how it relates to CI/CD pipeline automation. For example, how can Jenkins be used in this context?"

3.5 Iterate and Refine

  • Test different versions of the prompt to see which yields the best response.
  • Adjust the prompt as needed to improve accuracy.

3.6 Restrict

  • Limit options to make answers more precise.

Example:

Respond using only academic articles.
Don’t use sources older than 2020.
If you don’t know the answer, just say “I don’t know.”

4. Common Mistakes in Prompt Engineering

4.1 Overly Generic Prompts

  • Avoid broad questions that may lead to vague responses.

Don't do it:

"What is AI?"

Do it:

"What is AI and how is it applied in image recognition?"

4.2 Ambiguous Prompts

  • Avoid questions that could be interpreted in multiple ways.

Don't do it:

"Explain the use of networks."

Do it:

"Explain the use of neural networks in machine learning."

5. Techniques for Prompt Engineering

The following techniques help improve the precision and relevance of responses generated by language models.

5.1 Zero-shot

  • Definition: The model responds without related examples.

  • Use: Useful for broad, general tasks.

  • Example:

    • Prompt: "Explain what a relational database is."
    • Expected Response: The model responds based on general knowledge without providing examples.

5.2 Few-shot

  • Definition: Includes brief examples in the prompt to guide the response.

  • Use: Improves accuracy by providing specific context.

  • Example:

    • Prompt: "Write a Python function that calculates the average of a list of numbers. Input example: [10, 20, 30]. Expected output: 20."
    • Expected Response: The model follows the provided pattern to generate the response.

5.3 Chain-of-Thoughts (CoT)

  • Definition: The model is guided to think step-by-step, explaining its reasoning.

  • Use: It is useful for tasks that require detailed explanation and logic.

  • Example:

    • Prompt: "Explain step-by-step how to implement JWT authentication in a Flask application."
    • Expected Response: The model details each process step, such as library installation, route configuration, token generation, etc.

5.4 Your Technique Here

  • Definition: A custom technique you can create to meet a specific need.

  • Example: Role-based Prompting (Creating prompts based on roles or perspectives).

    • Prompt: "You are a virtual tour guide. Describe the Eiffel Tower to a group of tourists, including its history and fun facts."
    • Expected Response: The model responds as a tour guide, providing a more engaging and relevant description.

6. Use Cases for Prompt Engineering

6.1 Guiding Virtual Assistants

  • Developing prompts to help virtual assistants deliver useful and contextually relevant responses.

6.2 Content Creation

  • Utilizing prompts to produce articles, summaries, or product descriptions.

6.3 Automating Tasks

  • Designing prompts to automate repetitive tasks, such as generating reports or conducting data analysis.

Other Best Practices for Prompting

1. Chat History

It's essential to remember that your chat history can impact the answers you receive during a chat session. For example, if you're asking questions about a Java application and then suddenly switch to asking about Python applications, the answers you receive may need to be more helpful and accurate.

  • If you want to change the subject of the conversation, it's best to start a new chat rather than simply clearing the chat history. Clearing the chat history removes the messages from the page, but creating a new chat resets the conversation history and can result in better answers.

  • If you want to change the subject, click on a new chat to start fresh.

2. Chat way

Chat in phases. Ask questions, wait for the answer, and depending on that, ask for more details. See below:

  • How do I implement this?
  • Bring more information on this.
  • Change this to that.
  • Get this class and do this.

Enhancing Prompts with StackSpot AI

Knowledge Source

Creating a Knowledge Source is crucial as it helps provide context to StackSpot AI. If you have already created one and are asking questions that don't yield the desired results, check if your Knowledge Source has enough information and context.

See below some bits of advice:

1. When creating a knowledge source, it's best only to include the information that you need and separate it by subject. StackSpot AI selects and ranks documents based on their relevance to your topic, so if you have a lot of unrelated information in one source, you may not get accurate results.

2. Instead of grouping different code snippets, frameworks, and programming languages in one Knowledge Source, it's better to create separate ones for each context. It allows StackSpot AI to find more relevant and in-depth documents for each subject, resulting in a wider variety of high-quality documents.

3. See examples:

  • Knowledge Source with Kafka Integration snippets.
  • Knowledge Source with best practices for functional programming.

4. Consider organizing Knowledge Sources to improve search results.

5. You can add specific Knowledge Sources to your Workspace's context.