Student Guide to Generative AI

Search Term Specifics

If I wanted to search for information on Google about organic foods, I may type in "pros of organic foods". However, that will give me biased results that only focus on the positive aspects of organic foods. A better search query would be "positive and negatives of organic foods". 

And if I need information specifically about the process of organic farming, my search for "positive and negatives of organic foods" would not be specific enough. I should search for "positive and negatives of organic farming".

The same idea of specific search terms is true of generative AI! 

The way we phrase requests that we give generative AI are important for a few reasons:

• The training data of LLM tools is VAST. We must give context to our request so the AI can "reach" into that specific knowledge area for a relevant response. 
• Choosing biased wording for our requests will create biased responses like the organic farming example. 
• Just like asking a colleague to assist with a task, the more examples and information we can provide to them would give them the tools to provide a better finished product. 

Effective Chat Prompts

There are ways to craft a chat prompt, depending on your information need, which has been shown to be more effective than simply inputting a question or vague request. Using AI tools and practicing with various prompts can help you improve the responses you receive. 

• Try the PROMPT Framework in the box below.
• Try a few different prompt patterns in the box below. These can be used to create prompt templates that can be reused in different scenarios.
• The "7 ChatGPT Prompt Frameworks" link below provides more options for prompt framework acronyms.

Watch the video below for examples of creating effective prompts.

Since large language models (LLMs) which power generative AI tools are trained to predict patterns in text, it can be helpful to think about patterns that emerge in our prompt requests. 

The list below is a selection of prompt pattern types which have been developed to help people create templates that they can use to more efficiently use generative AI:

• The Persona Approach Pattern: "Act as Persona X, Perform task Y"
  • Example: "Act as an expert in marketing strategy. Brainstorm ideas for a marketing campaign for a ________ product."
• The Recipe Pattern: "I would like to achieve X. I know that I need to perform steps A,B,C. Provide a complete sequence of steps for me. Fill in any missing steps."
  • Example: "I need to create a daily study routine for college. I know that I need to perform steps like selecting a place to study and deciding what tasks to do each day. Provide a complete sequence of steps for me. Fill in any missing steps."
• The Cognitive Verifier Pattern: "When you are asked a question, follow these rules: Generate a number of additional questions about X. Combine the answers to the individual questions to produce Y"
  • Example: "When you are asked to create a research strategy, generative additional questions about my keywords choices, database selections, and Boolean combinations. Combine the answers to the individual questions to produce Y"
• The Alternative Approaches Pattern: "If there are alternative ways to accomplish a task X that I give you, list the best alternate approaches" (You will need to replace "X" with an appropriate task.)
  • Example: "For every prompt I give you, if there are alternative ways to word a prompt that I give you, list the best alternate wordings. Compare/contrast the pros and cons of each wording."

Personas and examples adapted from Vanderbuilt University. There are many more pattern types on their guide.