Artificial Intelligence Basics: A Beginner’s Guide to AI

4. Guidance on the Effective Use of Prompt Engineering with LLMs

4.1 What Is Prompt Engineering?

Defining Prompt Engineering and Its Importance

Prompt engineering is the practice of strategically designing and refining the input (prompts) given to a language model to elicit the most accurate, relevant, and useful responses. Since language models like ChatGPT generate outputs based on the prompts they receive, the way these prompts are crafted significantly impacts the quality of the generated content.

Importance of Prompt Engineering:

• Enhanced Performance: Well-designed prompts can guide the model to produce more precise and contextually appropriate responses.
• Efficiency: Effective prompts reduce the need for follow-up questions or clarifications, saving time.
• Control Over Output: By specifying style, tone, and format, users can tailor the responses to meet specific needs.
• Mitigating Errors: Clear prompts help minimize misunderstandings and reduce the likelihood of irrelevant or incorrect answers.

The Structure of a Good Prompt

A good prompt typically includes the following components:

1. Clear Instructions: Begin with a direct command or question that specifies the task. For example, “Explain,” “Describe,” “List,” “Compare,” etc.
2. Contextual Information: Provide necessary background details to frame the response. This helps the model understand the subject matter and any nuances.
3. Desired Format: Indicate how you want the information presented, such as bullet points, tables, step-by-step instructions, or essays.
4. Constraints and Modifiers: Include any limitations or stylistic preferences, such as word count limits, tone (formal or informal), or specific vocabulary to use or avoid.

Example of a Well-Structured Prompt:

• “In a 200-word formal essay, explain the causes and effects of climate change, focusing on human activities and their impact on global temperatures. Use scientific terminology and provide at least three examples.”

---

4.2 Practical Prompt Engineering Tips

Understanding Contextual Keywords

Contextual keywords are specific terms or phrases within a prompt that provide clarity and focus. They help the language model grasp the precise intent and scope of the request.

Tips for Using Contextual Keywords:

• Be Specific: Use precise terms related to your topic to narrow down the response.
  • Instead of: “Tell me about animals.”
  • Use: “Describe the habitat and diet of polar bears in the Arctic region.”
• Use Action Verbs: Begin prompts with verbs that define the expected action.
  • Examples: “Analyze,” “Summarize,” “Compare,” “Contrast,” “Define.”
• Include Relevant Details: Mention any specific aspects you want the model to cover.
  • Example: “List the economic impacts of renewable energy adoption in Europe.”

Using Constraints and Modifiers to Get Better Results

Constraints and modifiers refine the output by setting boundaries and specifying preferences.

Types of Constraints and Modifiers:

1. Length Constraints:
   • Example: “Summarize the following text in 100 words or fewer.”
2. Tone and Style Modifiers:
   • Example: “Explain the theory in simple terms suitable for a high school student.”
3. Format Specifications:
   • Example: “Provide the information in a table format.”
4. Content Restrictions:
   • Example: “Discuss the benefits of the policy without mentioning any drawbacks.”

Tips for Implementing Constraints and Modifiers:

• Combine Multiple Constraints: For greater specificity, use several modifiers together.
  • Example: “In a friendly tone, write a 150-word email inviting colleagues to the annual meeting.”
• Specify the Audience: Indicate who the response is intended for to adjust complexity.
  • Example: “Describe blockchain technology for readers with no technical background.”
• Direct the Focus: Clearly state what to emphasize or exclude.
  • Example: “List five advantages of electric vehicles, focusing on environmental benefits.”

---

4.3 Real-Life Examples of Prompt Engineering for LLMs

Case Studies with ChatGPT, Claude, and Other Models

Case Study 1: Enhancing Customer Support Responses with ChatGPT

• Ineffective Prompt:
  “Our product isn’t working. What should we do?”
• Effective Prompt:
  “As a customer support agent, draft a polite and helpful email responding to a customer who reports that their model X500 vacuum cleaner stops working after five minutes of use. Offer troubleshooting steps and assure them of our commitment to resolve the issue.”
• Outcome:
  The model generates a professional email that addresses the customer’s concern, provides specific troubleshooting steps, and reinforces customer service values.

Case Study 2: Generating Marketing Copy with Claude

• Ineffective Prompt:
  “Write about our new app.”
• Effective Prompt:
  “Create an engaging and upbeat product description for our new fitness tracking app, FitLife. Highlight features like personalized workout plans, progress tracking, and social sharing. Keep it under 150 words.”
• Outcome:
  The model produces a concise and compelling description that can be used in marketing materials.

Case Study 3: Academic Research Assistance with GPT-4

• Ineffective Prompt:
  “Tell me about climate change.”
• Effective Prompt:
  “Provide a detailed overview of the primary factors contributing to climate change, including greenhouse gas emissions, deforestation, and industrial activities. Cite reputable sources and present the information in an academic tone suitable for a graduate-level research paper.”
• Outcome:
  The model delivers an in-depth analysis with appropriately formal language, ready to be incorporated into academic work.

Case Study 4: Programming Help with LLMs

• Ineffective Prompt:
  “Fix my code.”
• Effective Prompt:
  “I have a Python function that’s supposed to calculate the factorial of a number, but it’s returning incorrect results. Here’s the code:

def factorial(n):
    if n == 0:
        return 0
    else:
        return n * factorial(n - 1)

Identify the error and provide the corrected code.”

• Outcome:
  The model pinpoints that the base case should return 1 instead of 0 and provides the corrected function.

Case Study 5: Creative Writing with LLMs

• Ineffective Prompt:
  “Write a story.”
• Effective Prompt:
  “Write a 500-word suspenseful short story about a detective in Victorian London who discovers a hidden society beneath the city. Use rich descriptions and incorporate elements of mystery and the supernatural.”
• Outcome:
  The model crafts a vivid and engaging story that meets the specified criteria.

---

By applying these prompt engineering techniques, users can significantly improve the quality of interactions with language models. Clear, specific, and well-structured prompts enable the models to understand the task better and deliver more accurate and relevant responses.