AI and Prompt Generation
What is Prompt Generation?
Prompt generation is the process of creating instructions or descriptions to guide a generative AI to produce content and refine its output to best fit what you want.
What is Generative AI?
Generative artificial intelligence (generative AI) is a machine learning system that can create new content such as text, images, music, audio, and videos, in response to a prompt or series of prompts. ChatGPT is an example of generative AI.
Some Types of Prompts Used in Generative AI
Brainstorming
Information Seeking
Problem-Solving
Creating Content
Other
General Tips for Prompting with AI
Consider: Is this an appropriate task to carry out using AI?
Running Conversation: You can continue prompting continuously through the conversation.
Custom Instructions: Use Custom Instructions whenever they are available to help minimize the amount of information in your prompt.
New Chats: Use a new chat if you are working on a different topic
Ask Me Questions: If you’re unsure of what information to provide, ask it to to ask you questions to “better fulfill your request”
Experiment with Word Choice: Consider what certain words may mean in the data set (ex. Game vs activity, or ESL vs EAL).
Reorder Ideas in Prompts Based on Priority : Reorder your prompt (ideas placed earlier will be given larger priority.
CORE Principles of Prompting
Context - Providing the necessary context
Adopt a Persona: What persona would you like the GenAI to adopt?
Audience: Who is the target audience? What does the GenAI need to know about the audience?
Background Information:
What background information might the GenAI need to know to provide relevant and informed responses?
What vocabulary should the GenAI know that may not match what is in its data set?
Is there any reference text/document that is necessary to provide to the GenAI?
What limitations should the GenAI be aware of?
Output - Providing explicit and clear instructions.
Desired Outcome: What is the overall goal of your prompt?
Details, Format, and Structure:
What keywords or ideas should be included in the output?
How should the output be formatted? Are there specific formatting elements needed in your output?
Is there an approximate maximum number of words?
Specific Subtasks: If needed, can you break down the steps needed to achieve the overall goal?
Reflect - Reflect on what information is provided
Data and Privacy: What should I consider from a data security and privacy lens?
Does the prompt include confidential or personally identifiable information? (It should not).
How can you anonymize information or take out information that may not be necessary for the output?
Quality: Consider the quality of the data that you are providing.
Bias/Perspective: Does the prompt provide the opportunity for multiple perspectives to be considered or is it one-sided?
Responsibility: Could the results of your prompt harm someone if they are inaccurate?
Evaluate - Refining the Output
Completeness: Does the output fully address all aspects of the prompt? Is there missing information that was expected based on the prompt?
Add additional information or requests into your prompt to tailor further. Consider providing an example if you have one.
Clarity: How clear and understandable is the response for your audience?
Consider adding a restriction of specific tone/style, specific format (ex bullet points), age/reading grade level
Bias: Does the output exhibit any bias? Are perspectives represented fairly without discrimination?
Consider asking it for various perspectives or asking it for pros and cons.
Accuracy: How accurate is the output? Are there any factual errors or misconceptions?
Fact-check it with your own knowledge, a friend, Google Search, etc. (Note: ChatGPT cannot fact check its own work)
Efficiency: How many prompts did you need to reach your desired output?
Reflect on how you could have added all the information together in the first prompt.
Responsibility: Are you able and willing to take responsibility for any errors or inaccuracies in your prompt?