Generative AI Prompting - Center for Teaching and Learning

Understanding Prompts

A prompt is essentially your input to the AI—a question, statement, or command that tells the AI what kind of information or response you're seeking. Think of it as starting a conversation on a specific topic.

Image created using DALL-E 3

Components of a Basic Prompt

1. Clarity: Be clear about what you're asking. If you're too vague, the AI might not give you the kind of answer you're looking for.
2. Specificity: Include specific details about what you need. The more specific you are, the more likely you'll get a relevant and accurate response.
3. Context: Provide any necessary background information. This helps the AI understand the framework or perspective from which you're asking the question.

Example

Imagine you're studying volcanoes and you have the following task:

• Vague Prompt: "Tell me about volcanoes."
  • This is too broad, and the AI might not know what specific information you're interested in.
• Improved, Specific Prompt: "Explain how a volcano erupts, including the roles of magma and gases, for a high school earth science class."
  • This prompt is clear, provides a specific topic (volcano eruptions), mentions key elements to include (magma and gases), and identifies the audience (high school earth science class), guiding the AI to tailor its response appropriately.

Steps to Crafting a Basic Prompt

1. Identify Your Need: Determine exactly what you want to know or what kind of response you're looking for. Do you want a summary, an explanation, a list, etc.?
2. Add Details: Include relevant details that can help the AI understand and respond accurately. This might include the topic's scope, any particular aspect you're interested in, or the format you want the response in (paragraph, list, etc.).
3. Specify the Audience: If relevant, mention who the information is for (e.g., beginners, experts). This helps in tailoring the complexity of the response.

Prompting Best Practices

When approaching creating prompts for Generative AI, it’s important to be strategic to ensure the output is useful, accurate, and relevant. Here are some tips to remember:

1. Be Specific and Detailed

Generative AI works best with clear, detailed prompts. The more specific you are about what you want, the closer the output will be to your expectations. Instead of a broad prompt like "write an essay on climate change," try "write a 500-word essay discussing the impact of renewable energy sources on global warming in the last decade." 

2. Define the Structure and Style

If you have a particular structure or style in mind, include that in your prompt. For instance, if you’re asking for an essay, mention whether you want it to be argumentative, descriptive, or persuasive. For creative works, you might specify a tone or mood, like "write a short story in a whimsical tone about a robot discovering nature." You can also try using the “Act as a…” style of prompting. This makes the AI roleplay a certain type of communicator and respond accordingly. So, for instance, the prompt, “Act as a motivational coach giving an uplifting speech to someone who has just lost their job,” specifies a supportive and encouraging tone.

3. Use Iterative Prompting

Don’t expect perfection in one go. Use the output as a draft and refine your prompt based on what you receive. This iterative process can help tailor the AI's responses to better suit your needs. For example, if the essay lacks certain details, you could follow up with, "expand on the renewable energy technologies mentioned, focusing on solar and wind power."

4. Incorporate Keywords Wisely

If there are specific terms or concepts you want included, mention them in your prompt. Keywords help guide the AI’s focus and ensure the output aligns with your topic’s requirements. For instance, "include examples of photovoltaic technology and offshore wind farms" can direct the AI to cover those areas.

5. Set Parameters for Length and Complexity

Be clear about the desired length and the complexity level of the content. Indicating word count, paragraph number, or even the reading level can help generate content that fits your needs without additional editing.

6. Clarify the Audience

Understanding the intended audience can significantly impact the tone, style, and complexity of the AI's output. Specify whether the content is for beginners, experts, or a general audience. For example, "write an introduction to quantum computing for high school students" sets a different expectation than "for postgraduate physics students."

Image created using DALL-E 3

7. Check for Bias and Accuracy

Always review the AI-generated content for potential biases and factual accuracy, especially when dealing with sensitive topics. AI models can inadvertently propagate biases present in their training data, so it’s crucial to critically evaluate their output.

Image created using DALL-E 3

Avoiding Hallucinations

Minimizing hallucinations—instances where Generative AI produces false or nonsensical information—in AI-generated content involves crafting prompts that guide the AI towards accuracy, relevance, and clear understanding. Here are strategies to achieve that:

1. Be Specific and Detailed in Your Prompt:

• Precise Information: Include as much relevant detail as possible in your prompt. The more context you provide, the less likely the AI is to "hallucinate."
• Example: Instead of asking, "Tell me about historical events," specify "Describe the key events and their impacts of the French Revolution between 1789 and 1799."

2. Limit the Scope of the Prompt:

• Focused Queries: Narrow down the scope of your question to avoid broad or vague responses that can lead to inaccuracies.
• Example: Rather than prompting, "How do computers work?" you might ask, "What is the function of a CPU in a computer?"

3. Request Sources or Basis for Claims:

• Source Inclusion: Ask the AI to include sources or the basis of its information, understanding that it cannot browse the internet but can refer to its training data up to its last update.
• Example: "Based on your training data, what are the main reasons cited for the fall of the Roman Empire? Please specify the sources of your information."

4. Use Prompts that Encourage Fact-Based Outputs:

• Fact Orientation: Craft prompts that lead the AI to generate outputs based on widely accepted facts or data.
• Example: "Provide a summary of the photosynthesis process in plants, including the chemical equation involved, as understood in current biology textbooks."

5. Incorporate Verification Steps into Your Prompt:

• Self-Check Requests: Ask the AI to perform a self-check on the information it provides, if possible, although this technique has its limitations.
• Example: "Explain the theory of relativity, and please include a brief verification of the key facts you present."

6. Iterative Prompting:

• Refinement Through Feedback: Use the AI's responses to refine your prompt, asking for clarifications or corrections as needed.
• Example: If the initial response to a question about a historical event contains inaccuracies, follow up with, "You mentioned [incorrect detail]. Can you verify this or provide a more accurate description?"

7. Avoiding Leading Questions:

• Neutral Wording: Ensure that the prompts are neutrally worded and do not lead the AI towards a predetermined answer, which might not be accurate.
• Example: Instead of suggesting, "Why is quantum computing considered superior to classical computing?" ask, "Can you compare quantum computing and classical computing?"

8. Stay Updated on AI Developments:

• Continuous Learning: Keep abreast of the latest updates and improvements to AI models and their capabilities to craft better prompts.
• Example: If new versions of the AI are released with improved accuracy or new features, adjust your prompts to leverage these advancements.

By employing these strategies, prompters can significantly reduce the occurrence of hallucinations in AI-generated content, leading to more accurate, reliable, and useful outputs.