Topic outline

• Optimizing Your Prompts for Generative AI

  Optimizing Your Prompts for Generative AI

  Content Type
  Exercise

  Duration
  40m

  Level
  Introductory

  Learn the essentials of AI prompt engineering and see how to effectively craft requests. Then, see how prompt engineering techniques fit with an AI application and try them yourself.

  By leveraging prompting techniques, you can optimize the output of generative AI tools, helping you achieve your goals. While many techniques are available, you will try the most common in this exercise:

  • Zero-shot learning: Providing a basic prompt to the AI tool without training or examples.
  • Task-oriented prompting: Designing prompts for specific task completion.
  • Few-shot learning: Sharing examples with the AI tool to guide output.
  • Chain of thought: Asking questions to generate intermediate reasoning steps.
  • Instructional prompting: Providing direct instructions for task execution.
  • Chain prompting: Using consecutive prompts that build on each other.

  Watch the video and try the exercise below!
• Start the Exercise

  Start the Exercise

  In this exercise, you will explore six prompting strategies described in the video and listed above. Each technique offers a unique approach to interacting with generative AI tools, and you will see how to craft queries in different ways to achieve the desired output. Specifically, you will leverage ChatGPT to plan a vacation to Boston, MA in the United States.

  Note:

  At the beginning of each task, continue in the same thread unless directed to start a new chat.

  • Sign up for a generative AI tool
    

    To begin, make sure you have access to a generative AI tool. You do not need a paid version; the most common tools have a free option. This exercise was written with ChatGPT in mind, but you can also try it with Gemini, Anthropic, or other tools.

  • Start with zero-shot learning
    

    Zero-shot learning involves prompting the AI tool to perform a task without prior specific training or examples. Here, you will develop an initial itinerary for a vacation in Boston.

    1. Enter an initial prompt: Create a basic 5-day itinerary for a vacation in Boston.
    2. Evaluate the response. Assess the itinerary, considering whether it includes activities that match your interests.
    3. Identify areas for improvement. Note any gaps or impractical suggestions and consider how the itinerary could be better balanced or tailored to specific interests. You will better define the desired output using the next technique, task-oriented prompting.
  • Try task-oriented prompting
    

    Task-oriented prompting is useful when you would like the AI tool to help you achieve a particular task—in this case, creating an itinerary with a specific output. Add detail to your previous request, and specify what the output should contain (a best practice for using generative AI).

    1. Refine the prompt by defining the desired outcome. Specify a preference for three types of activities that would interest you, and ask for a table format of Day:Area | Time | Activity.
       Sample prompt: Create a basic 5-day itinerary for a vacation in Boston with a focus on history, art, and outdoor activities. Please put this in a table format Day:Area | Time | Activity.
       Note that some AI tools format tables better than others. If you do not get the output you expected, simply re-run the request to see if it produces what you expect.
    2. Read the response, considering whether it is practical. Is the plan realistic, accounting for opening hours and travel time?
    3. Ask the AI tool to check itself.
       1. Copy or download the response, then open a new chat window and ask: Is this itinerary realistic, accounting for opening hours and travel time? Paste in the response.
          
          Asking the AI tool to check itself is a great strategy that you can use with any of these techniques to catch errors and hallucinations in the output. AI tools often overlook issues in the same chat because as they track conversations, they may retain incorrect details from that conversation. It can often be easier to spot potential issues in a new chat.
       2. Choose one identified concern, convey this information in the original chat, and ask for an updated recommendation given the feedback.
    4. In the original chat, provide dietary needs and see how the recommendation changes. For example: I’d like to focus on gluten-free meal options. Please update the itinerary based on this and prioritize accuracy, suggesting real restaurants that have a high likelihood of gluten-free options.
       Be aware that the AI tool may still hallucinate, so you should check its work. The chances of hallucination are greatly decreased if you specify that accuracy is important.
  • Use few-shot learning
    

    With few-shot learning, you are offering examples to guide the AI tool. Here, you will provide examples of activities that interest you; these will help to customize your Boston vacation plan.

    1. Provide examples of activities that you would enjoy, and add these to the refined prompt. For example:
       Create a basic 5-day itinerary for a vacation in Boston with a focus on history, art, and outdoor activities. Please put this in a table format: Day:Area | Time | Activity. For example, I'm interested in checking out the Isabella Stuart Gardner Museum and the Beacon Hill Bookstore.
    2. Review the response. Does it include activities you would be interested in? If not, add more examples such as seeing a baseball game, visiting an outdoor museum, or visiting well-known universities.
  • Explore the chain-of-thought technique

    Dive into the reasoning behind the AI tool's responses to better understand options and considerations so you can develop your plan thoughtfully. Identify potential challenges in the vacation plan, such as weather or peak tourist seasons, and ask how to address them.

    1. Still in the original chat, initiate with a probing question such as: What are potential challenges when visiting Boston in April, and how can they be mitigated?
    2. Analyze the reasoning. Review the response to ensure it covers issues you would expect such as weather and crowding during peak seasons.
    3. Deepen the discussion, highlighting one of the challenges and asking for specific alternatives or additional tips in case of unforeseen circumstances: What suggestions do you have for a plan B for my itinerary given this challenge?
  • Use instructional prompting

    Instructional prompting is useful when you have specific instructions to follow, or when you would like the AI tool to create a set of clear instructions. Here, you will guide the AI tool to list steps for preparing for the vacation, from booking flights to packing.

    1. Create the instructional prompt: What are the step-by-step actions needed to prepare for a week-long vacation in Boston starting from one month before departure?
    2. Review the response to see if the suggestions are helpful and meet your needs and expectations. What's useful, and what isn't? Does the response include the right level of detail?
       Refine for clarity, adjusting the steps to be more detailed or simplified, depending on your travel experience.
    3. Refine your prompt by asking for clarity and order, ensuring instructions are logical and easy to follow. Also, provide additional context on the user (in this case, you, the traveler): What are the step-by-step actions needed to prepare for a week-long vacation in Boston starting from one month before departure? I am a new traveler, so please include as much detail as possible. This is VERY important. Please only outline what new things you'd add to the step-by-step actions.
  • Try chain prompting

    Use chain prompting—interacting in a back-and-forth manner with the AI tool—to refine each day of the Boston itinerary based on your initial suggestions and preferences. Ensure each prompt builds on the previous response and stays focused for a coherent and functional itinerary.

    1. Start with the initial outline of the trip: Review the table of the Boston itinerary and identify any days that seem too packed or too light.
    2. Build on the response, for example: For days that are too packed, suggest how activities could be spread out, or provide alternative options.
    3. Use follow-up questions to adjust dining options, add evening activities, or include rest periods. Stay focused on the output of the previous response to dive deeper. For example, if an alternative option is suggested that you’re interested in, highlight it and ask the AI tool to elaborate on that option: Please tell me more about this.

    
  • Assess and further your understanding

    Pull these skills together and use the AI tool to check your understanding, recommend further areas to explore, and grow your skills.

    1. Open a new chat and have the AI tool define these six prompting techniques that you’ve tried: Please define several AI prompting techniques: zero-shot learning, task-oriented prompting, few-shot learning, chain of thought, instructional prompting, chain prompting.
    2. Check your knowledge of prompting techniques with interactive quizzes that result in a summary of what you might explore next: One at a time, please ask me 6 questions to assess my understanding of AI prompting techniques and prompting best practices. Wait for me to respond between each question before asking the next question. At the end, please give me a summary of my understanding, provide recommendations for skills I could improve, and suggestions for how to continue learning. Please prioritize accuracy.
    3. Further focus the questions on the topics in this exercise: Ask me two more questions on the topic of zero-shot learning, task-oriented prompting, few-shot learning, chain prompting, instructional prompting, or chain of thought. Wait for my response to the first question before asking the second question. Summarize any additional recommendations to enhance my skills.
• What's Next?

  What's Next?
  • Learn about responsible use of Generative AI (video, 15m).
  • Explore the basics of Generative AI (learning path, 2h 45m).
    

  • Share Your Feedback!