A Simple Introduction to Prompt Engineering
Download as PPTX, PDF1 like811 views
A Simple Introduction to Prompt Engineering
![Few-shot & Zero-shot Prompting
• Few-shot Prompt - Providing Example to Guide AI:
Here are some temperature conversion functions
… [SAMPLE CODE] … Now, write a function to
convert Fahrenheit to Celsius following the
same style."
• Why better?
• More structured & consistent.
• AI follows the given style.
• Consistent formatting and readability.
• Encourages AI to infer patterns from examples.](https://image.slidesharecdn.com/promptengineering-250223013943-bd974d44/85/A-Simple-Introduction-to-Prompt-Engineering-40-320.jpg)
![Writing Code
• Write a Python function to [describe functionality] .
• Generate a [language] script that does [specific task].
• Create a command-line tool that [functionality].
• Write a function to sort a list using [QuickSort, MergeSort,
etc.].
• Generate a [Flask/Django/React] application that does [features].](https://image.slidesharecdn.com/promptengineering-250223013943-bd974d44/85/A-Simple-Introduction-to-Prompt-Engineering-50-320.jpg)
![Debugging Code
• Find and fix bugs in this [language] code: [paste
code].
• Why is this function returning the wrong result?
[paste code].
• Help me debug this error message: [paste error].
• Explain what this error means and how to fix it:
[error message].](https://image.slidesharecdn.com/promptengineering-250223013943-bd974d44/85/A-Simple-Introduction-to-Prompt-Engineering-51-320.jpg)
![Optimising Code
• Optimize this function for performance: [paste code].
• Rewrite this algorithm to improve time complexity.
• Convert this loop into a more efficient list
comprehension.
• How can I reduce memory usage in this
[Python/JavaScript/etc.] function?"](https://image.slidesharecdn.com/promptengineering-250223013943-bd974d44/85/A-Simple-Introduction-to-Prompt-Engineering-52-320.jpg)
![Explaining Code
• Explain this function step by step: [paste code].
• What does this code do? [paste code].
• Break down this algorithm in simple terms.
• Describe how this function works in a way a
beginner would understand.](https://image.slidesharecdn.com/promptengineering-250223013943-bd974d44/85/A-Simple-Introduction-to-Prompt-Engineering-53-320.jpg)
![Testing Code
• Write unit tests for this function using
[unittest/PyTest/Jest].
• Generate test cases for this function: [paste
function].
• Create a set of edge cases to test this algorithm.
• How do I mock [API call/database query] in a test?"](https://image.slidesharecdn.com/promptengineering-250223013943-bd974d44/85/A-Simple-Introduction-to-Prompt-Engineering-54-320.jpg)
A Simple Introduction to Prompt Engineering
- 2. Prompt Engineering • Prompt engineering is designing and refining prompts to effectively communicate with AI models, improving quality, relevance, and accuracy of their responses.
- 4. Prompt Engineering •Prompt Engineering can … • increase the accuracy and relevance of answers. • help control the tone, style, and format of answers. • enhance the creativity and efficiency of answers.
- 5. Prompt Engineering •It is important to note that different AI models interpret prompts differently (ChatGPT, DALL·E, Copilot, etc.), so the most important thing to do is try out different queries and different approaches to generating queries, and see which ones work best.
- 6. Prompt Engineering • Ask a question • Be specific PROMP T • Review the answer • Try to verify PROBE • Add more Context • Add constraints POLISH
- 9. What is an AI Hallucination? •An AI hallucination occurs when an AI generates false, misleading, or nonsensical information that appears plausible but is not based on real data or facts. •Essentially, the AI "makes things up" because it lacks the right knowledge or is overconfident in its response.
- 10. What is an AI Hallucination? •For Example: • Who will win the 2044 Formula 1 World Championship? •Possible AI Answer: • The 2044 Formula 1 World Champion will be Max Verstappen, winning 10 races.
- 11. What is an AI Hallucination? • Why Do AI Hallucinations Happen? • Lack of Real Data: AI fills gaps with plausible but incorrect information. • Pattern Matching Without Understanding: AI generates text based on statistical probabilities, not true comprehension. • Overconfidence in Its Own Output: AI doesn’t always indicate uncertainty when unsure. • Ambiguous or Poorly Defined Prompts: If a prompt lacks context, AI guesses instead of asking for clarification. • Random Answers: Sometimes the AI randomly generates incorrect answers, even if it has the correct answer.
- 12. What is an AI Hallucination? •How to Reduce AI Hallucinations? • Be Specific in Your Prompts: Ask for explanations, examples, or step-by-step logic. • Use Step-by-Step Prompting: Force AI to explain reasoning before giving answers. • Verify with External Sources: Cross-check AI output using real-world references. • Use Web Searches for Real-Time Data: Prevent outdated or fabricated facts.
- 14. What is an Edge case? •Edge cases are extreme, uncommon, or boundary cases. •Edge cases help identify potential bugs, inefficiencies, or unexpected behaviors that might not appear in normal conditions.
- 16. What is an Edge case? • For example, checking if a string is a palindrome. •Normal case: • "racecar" True → •Edge cases: • Empty string "" Should return True? → • Single character "a" Should return True? → • Spaces and punctuation "A man, a plan, a canal, Panama!" • Non-alphabetic input "12321"
- 18. Prompt Engineering •Principles of Effective Prompting: •Clarity & Precision •Context & Constraints •Iteration & Testing •Role-based Prompts •Few-shot & Zero-shot Prompting •Chain-of-Thought Prompting
- 20. Clarity & Precision •Using clear and specific language to guide AI responses. •Let’s say you want an AI to generate a Python function to calculate the factorial of a number >>>
- 21. Clarity & Precision •Bad Prompt - Vague & Ambiguous • Write a function for factorial. •Issues • Doesn’t say which programming language to use. • Unclear if recursion or iteration is preferred. • No details about input validation.
- 22. Clarity & Precision • Good Prompt - Clear & Precise • Write a Python function called factorial that takes a non-negative integer n as input and returns its factorial. Implement it using recursion, and raise a ValueError if n is negative. • Why Better? • Specifies language (Python). • Defines function name (factorial). • Clarifies input constraints (non-negative integer). • Specifies methodology (recursion). • Includes error handling (raise ValueError).
- 24. Context & Constraints •Providing background information or setting boundaries to get relevant answers. •Let’s say you want to generate a function to sort a list of numbers >>>
- 25. Context & Constraints • Bad Prompt - Lacks Context & Constraints • Write a function to sort a list. • Issues • Which programming language? • What kind of sorting algorithm? • Should it be ascending or descending? • Should it handle duplicates or special cases (like empty lists)? • Should it be optimized for performance (e.g., large datasets)?
- 26. Context & Constraints • Good Prompt - Provides Context & Constraints • Write a Python function called custom_sort that takes a list of integers and sorts them in ascending order using the Merge Sort algorithm. The function should be efficient for large lists and should preserve duplicate values. If an empty list is given, return an empty list. • Why Better? • Context: Python, sorting integers. • Constraints: Must use Merge Sort, handle large lists, preserve duplicates, and return an empty list when given one.
- 28. Iteration & Testing •Experimenting with different phrasings and formats to achieve the best output. •Let’s say we want to generate a function that checks if a string is a palindrome (reads the same forward and backward) >>>
- 29. Iteration & Testing • Iteration 1 – A bit vague • Write a Python function to check if a string is a palindrome. • Issues • Doesn’t specify case sensitivity (should "Racecar" be considered a palindrome?). • Doesn’t mention whitespace (should "nurses run" be valid?). • No requirement for handling non-alphabetic characters (like punctuation).
- 30. Iteration & Testing •Iteration 2 – Adding Constraints • Write a Python function is_palindrome that checks if a string is a palindrome, ignoring case and whitespace. It should return True for palindromes and False otherwise. •Issues • Fixes case sensitivity and whitespace handling • Still doesn’t handle punctuation like "Madam, I'm Adam."
- 31. Iteration & Testing •Iteration 3 – Final Refinement • Write a Python function is_palindrome that checks if a string is a palindrome, ignoring case, whitespace, and punctuation. Use Python’s re module to remove non-alphanumeric characters. •Why Better? • Optimized & Robust • Now handles case, whitespace, and punctuation correctly!
- 32. Iteration & Testing •Methodology • Start with a basic prompt and analyze the AI’s output. • Iterate by adding constraints based on shortcomings. • Test edge cases to verify correctness. • Refine until the AI’s output meets all requirements.
- 34. Role-based Prompts • Asking the AI to act as a specific expert or persona. • This helps generate more accurate and more context- aware responses. By assigning a role, you guide the AI's thought process and ensure it tailors responses to the expected expertise level. • For example, role-based prompt for Code Reviewer >>>
- 35. Role-based Prompts •Prompt Without a Role - Generic Output • Explain this Python function. •Issues • The AI doesn’t know who the explanation is for (Beginner? Expert? Developer?) • Might provide a generic, surface-level response
- 36. Role-based Prompts • Role-Based Prompt - More Focused & Useful • You are a senior Python developer reviewing code for a junior colleague. Explain the following function with clear comments and suggestions for improvement. • Why better? • Provides comments and docstrings for clarity. • Offers constructive feedback (input validation, error handling). • Uses a mentor-like tone, making it useful for junior developers.
- 38. Few-shot & Zero-shot Prompting •Providing examples (few-shot) or asking the AI to generate responses with no prior examples (zero-shot). Prompt Type Explanation Few-shot Prompting The AI is given one or more examples to learn from before generating an answer. Zero-shot Prompting The AI is given no examples and must infer the pattern or solution.
- 39. Few-shot & Zero-shot Prompting •Zero-shot Prompt - No Examples Given: • Write a Python function that converts Fahrenheit to Celsius. •Issues • No request for docstring or explanation. • No request for input validation or edge case handling
- 40. Few-shot & Zero-shot Prompting • Few-shot Prompt - Providing Example to Guide AI: Here are some temperature conversion functions … [SAMPLE CODE] … Now, write a function to convert Fahrenheit to Celsius following the same style." • Why better? • More structured & consistent. • AI follows the given style. • Consistent formatting and readability. • Encourages AI to infer patterns from examples.
- 41. Few-shot & Zero-shot Prompting • When to use each: • Zero-shot: Useful for general tasks but may produce inconsistent results. • Few-shot: Helps guide the AI by showing examples, improving accuracy and structure. So, use few-shot prompting when you need consistency, style matching, or complex logic.
- 43. Chain-of-Thought Prompting •Also called: •Step-by-Step Instructions •Step-by-Step Prompting •Multi-Turn Prompting
- 44. Chain-of-Thought Prompting •Encouraging step-by-step reasoning for complex tasks. •This helps improve accuracy and ensures logical consistency, especially for mathematical or algorithmic tasks. •For example, we have a function that finds the sum of all even numbers in a list, but it’s giving incorrect results >>>
- 45. Chain-of-Thought Prompting •Normal Prompt • Find the bug in this Python function. •Issues • Sometimes, the AI will fix the error but doesn’t explain why, making it harder for learners to understand.
- 46. Chain-of-Thought Prompting •Chain-of-Thought Prompt • Analyze this Python function step by step to find any logical errors. Explain each step before fixing the bug. •Issues • Sometimes, the AI will fix the error but doesn’t explain why, making it harder for learners to understand.
- 47. Chain-of-Thought Prompting •AI Output - With Step-by-Step Reasoning •Step 1: Understand the function's goal. •Step 2: Analyze the features of the code. •Step 3: Identify the fix. •Step 4: Provide the corrected function.
- 48. Chain-of-Thought Prompting •Why better? • Clear breakdown of the problem before fixing it. • Step-by-step reasoning helps with learning and debugging. • Reduces AI hallucination by enforcing structured logical thinking.
- 49. Sample Prompts
- 50. Writing Code • Write a Python function to [describe functionality] . • Generate a [language] script that does [specific task]. • Create a command-line tool that [functionality]. • Write a function to sort a list using [QuickSort, MergeSort, etc.]. • Generate a [Flask/Django/React] application that does [features].
- 51. Debugging Code • Find and fix bugs in this [language] code: [paste code]. • Why is this function returning the wrong result? [paste code]. • Help me debug this error message: [paste error]. • Explain what this error means and how to fix it: [error message].
- 52. Optimising Code • Optimize this function for performance: [paste code]. • Rewrite this algorithm to improve time complexity. • Convert this loop into a more efficient list comprehension. • How can I reduce memory usage in this [Python/JavaScript/etc.] function?"
- 53. Explaining Code • Explain this function step by step: [paste code]. • What does this code do? [paste code]. • Break down this algorithm in simple terms. • Describe how this function works in a way a beginner would understand.
- 54. Testing Code • Write unit tests for this function using [unittest/PyTest/Jest]. • Generate test cases for this function: [paste function]. • Create a set of edge cases to test this algorithm. • How do I mock [API call/database query] in a test?"