Tips for Using Generative AI

Generative AI programs like ChatGPT, DALL-E, Bing Chat, and others are powerful tools that offer instructors and researchers a variety of new opportunities.

To harness the power and potential of Artificial Intelligence, you must also consider how and when the capabilities of these tools best meet your needs. Just like you wouldn’t use a calculator to proofread an essay, generative AI tools can’t fix every problem. But when used strategically, these tools can be powerful. On this page, we provide some strategies for addressing the following questions:

What are some of the strengths and limitations of generative AI?
When and how should I use generative AI?

Perhaps the most important step in using generative AI is to first learn how to appropriately and effectively use these tools, which includes developing skills for effectively writing prompts—also known as prompt engineering. This page offers some initial guidance for using generative AI tools. We also provide additional resources for developing your competencies, including our page on prompt patterns and a free, self-paced course on Prompt Engineering taught by Vanderbilt University Professor of Computer Science Jules White.

What are the strengths and weaknesses of generative AI?

To understand the strengths and weaknesses of generative AI, it is worth keeping in mind how AI works. Generative AI is essentially a very advanced pattern recognition software. Programs like ChatGPT analyze large sets of data and produce unique outputs that mimic the patterns identified in their data. Because of this, generative AI programs are great at recognizing patterns, but they may be less skilled at producing factually-sound output. Additionally, the quality of the output a generative AI program generates depends heavily on the quality of your prompt. When writing prompts, consider how you might capitalize on generative AI’s strengths while minimizing its weaknesses.

The lists below, while not exhaustive, highlights some key strengths and weaknesses of generative AI programs.

Strengths of generative AI

Proofreading

Generative AI programs are skilled at pattern recognition, which makes them excellent proofreading tools. This can be especially helpful if you are writing in a second language or are taking final steps to prepare a piece of writing for publication.
Brainstorming

Generative AI programs can generate questions to help you consider certain ideas more deeply. They may also generate arguments for or against a topic if you are trying to view an issue from multiple perspectives.
Generating code

Coding is one example of the varied formats of output you can get from generative AI. Even if you are inexperienced in coding or web design, generative AI tools are skilled at mimicking the patterns and rules of common coding languages. By using a well-crafted prompt and taking the time to verify the quality of the code, you can quickly write or troubleshoot code in languages such as Python, R and HTML.
Generating visuals

Create images for presentations, handouts, or web pages where you need to add imagery to something to improve its design. Using tools like DALL-E, Stable Diffusion, Midjourney, or many others, you can give the AI tool a text prompt describing what image you want, and then it will generate that image for you.
Mimicking and explaining genre

Because many generative AI programs are skilled at recognizing patterns, this makes them particularly adept at mimicking certain genres, especially those that are more formulaic. They may also be able to describe certain elements that should be present in a given genre of writing.
Mimicking tone

Given the appropriate prompt, generative AI tools can mimic the tone of either a specific author, artist, or a specific body of work. This can be useful when you need to consider questions of audience in a given piece of writing.

Weaknesses of generative AI

Factual inaccuracies (i.e. hallucinations)

Generative AI tools have been known to experience hallucinations, meaning that they provide information that is either entirely or partially fabricated. These tools provide text that they know to be grammatically and semantically correct, but they have no actual understanding of the language they’re using. It’s important to remember that generative AI tools are designed to identify and recreate patterns. These tools have no concept of what is true or accurate.
Inherent bias

Training data used to develop generative AI tools is often primarily based on English text from Western sources. This means that this data inherently demonstrates racial and cultural biases. This issue is also seen in image generation tools, which have been shown to demonstrate biases in race, culture and gender.
Privacy and liability
- Some generative AI programs may not meet the minimum requirements to protect sensitive data or information protected by privacy laws such as HIPPA or FERPA. Additionally, many organizations, including the National Institute of Health, have strict regulations about inputting other people’s research into generative AI software.
- Sharing your own unpublished research or data into generative AI programs could potentially result in your research and data being used to train generative AI programs. Before using generative AI programs, review all relevant policies and regulations and understand how your input will be used.
Citation and authorship

By nature, generative AI programs may summarize or paraphrase a concept or idea without properly attributing the ideas to their sources.
Inequity of access

Currently, many generative AI programs have both free and paid versions. If you are in a context where you may require or permit the use of generative AI, such as in the classroom, consider how you will navigate questions of access.
Linguistic and cultural limitations

Generative AI programs are programmed on large volumes of data. However, these data are not comprehensive and may underrepresent certain languages and cultures. Moreover, these programs may not be adept at distinguishing between reliable, high-quality content and unreliable content.
Unintended consequences of use

Given the newness of generative AI tools, there are a variety of ways that generative AI use may have consequences that users may not anticipate. For example, some scholars have identified dilemmas related to intellectual property, environmental impacts, and the future of work. To make an informed decision related to how you will use generative AI, you might decide to research and consider the impacts of generative AI use from many angles. Resources such as the Vanderbilt Policy Accelerator can serve as a useful starting point for exploring topics that might be of interest.

When and how should I use generative AI?

To reap the full benefits of generative AI, it is worth considering both why you want to use it and how you will use it.

As a starting point, this flow chart¹ can help you determine if generative AI is an appropriate option for you.

Flow chart describing when it may be safe to use ChatGPT with the following possible use cases: 1. If it does not matter if the out is true, it is safe to use ChatGPT. 2.If it matters if the output is true, you have the expertise to verify that the output is accurate, and you are willing to take full responsibility (legal, moral, etc.) for missed inaccuracies, it is possible to use ChatGPT (but be sure to verify each output word and sentence for accuracy and common sense). 3. If it matters if the output is true, you have the expertise to verify that the output is accurate, but you are not willing to take full responsibility for missed inaccuracies, it is unsafe to use ChatGPT. 4. If it matters if the output is true and you do not have the expertise to verify that the output is accurate, it is unsafe to use ChatGPT.

As you begin a project or task that incorporates generative AI, consider building a strategy based on the following questions:

What is my purpose and goal in using generative AI? Do you want to make a certain process more efficient? Fill a knowledge gap? Generate new ideas?
Who is my audience? What are their attitudes towards the use of generative AI? What expectations do they have about the use of generative AI and citation practices? How can I use these tools to meet the needs and/or expectations of my audience?
What kind of information will I be sharing with generative AI? Am I okay with this information potentially being shared or circulated? Am I sharing information that is subject to certain privacy restrictions such as FERPA or HIPPA?
What specific information or product do I want to get from generative AI? Do I want to generate brainstorming questions? Do I want it to generate a polished version of a piece of writing?
How will I prompt AI in a way that generates the output I want? How might generative AI misunderstand my request? What kinds of biases might the AI generate, and how can I adjust my prompt to mitigate these biases?
How will I revise and review the output that AI generates? What kinds of errors or limitations do I want to pay close attention to? How will I distinguish between helpful and unhelpful output?

Example 1: Creating Tailored Assignments and Readings

Suppose you are teaching a course where you require your students to respond to several relevant hypothetical situations for an upcoming lesson. You have thirty-five students and are hoping to use generative AI to create unique scenarios for each student. You might use these questions to generate the following answers:

What is my purpose and goal in using generative AI?

I want to create a different and unique written situation for each student I have in class, but I don’t have the time to think of and manually type out 35 different options.
Who is my audience?

Students in my course. I want students to each have a unique experience so they can formulate their own approach to each situation and then share their decisions with the class later. In order to avoid repetition and to provide many unique experiences, I want different scenarios for each student.
What kind of information will I be sharing with generative AI?

I will be sharing basic public information about the course I’m teaching along with the subject matter that this assignment is about.
What specific information or product do I want to get from generative AI?

I want to end up with different examples of hypothetical situations that match the topic I am teaching. I want each example to be detailed enough to get students started, but open-ended enough for them to draw their own conclusions and plan their own approach to addressing this situation.
How will I prompt AI in a way that generates the output I want?

I know I want hypothetical scenarios that match the topic that we are discussing in class, and I do have certain requirements that should and should not be included. I will draft a prompt that asks for plausible scenarios of my topic and the content I do and do not want included. I also know that generative AI tools can only provide so much text output at one time, so I will start with my basic prompt, then ask it to produce multiple examples again and again until I have my total of 35 scenarios.
Review Prompt Patterns
How will I revise and review the output that AI generates?

After I try my first prompt, I will evaluate the output to see if it matches what I want. If it does, then I will prompt the generative AI to create more examples by saying “Please generate 5 more.” If it does not match what I want, I will ask it to adjust its output to include or remove information as needed.

By following these steps, you decide that you can use generative AI to create these scenarios. You formulate a prompt to list the information you want the tool to include in the scenarios as well as what information you do not want. You are aware of limitations on the amount of text generative AI tools can produce at one time, so you have a plan on how to create many scenarios through multiple prompts.

Example 2: Using generative AI to create visuals for a presentation

Suppose you are preparing a conference presentation, and you want to include visuals that accompany some of your slides. You wonder if generative AI is a good option for this task. You might use these questions to generate the following answers:

What is my purpose and goal in using generative AI?

I want to include relevant images in my presentation. I often struggle to find images that are relevant to my research, so I often lack visuals or have to take photos or create custom graphics myself.
Who is my audience?

I am presenting to colleagues in my field. In my experience, my colleagues are fairly open to the use of generative AI, although there are some folks in my field who have concerns about its use. I know that my colleagues expect me to cite my use of generative AI, and I will also include a verbal disclosure and explanation for my colleagues who are more skeptical about the use of these tools. I typically use MLA style, so I will follow their citation practices for generative AI.
What kind of information will I be sharing with generative AI?

I am presenting research that I haven’t published yet, so I don’t want to share my argument with generative AI. Instead of sharing excerpts from my conference presentation to create a visual, I will write a prompt that explains the type of image that I want.
What specific information or product do I want to get from generative AI?

I’m hoping to get some unique, clip-art style images. I know that some of the visuals I need for my presentation will be fairly easy to find free images or images in the public domain with a quick Google search, so I will primarily use generative AI to create unique images that I cannot easily find online or in place of images that may require licensing to use.
How will I prompt AI in a way that generates the output I want?

I know that I specifically want clip-art style images, so I will be sure to include that information in my prompt. I want to include images of professionals from different academic disciplines. I will be specific in my prompting, since I know that generative AI programs may overrepresent certain demographics.
Review Prompt Patterns
How will I revise and review the output that AI generates?

Since I am asking for images of professionals, I know there is a chance for biased outputs, especially in relation to representations of gender, race, and ethnicity. When reviewing the outputs, I plan to pay close attention to these possible biases.

By following these steps, you decide to use generative AI to produce images that you cannot easily find online. You also decide not to disclose the main takeaways of your research in your prompts, to use detailed, specific prompts, and to be wary of potential biased outputs related to representations of gender, race, and ethnicity. You also have a plan in place for disclosing your generative AI use during your presentation.

Example 3: Navigating the ethics of incorporating generative AI in grading

Suppose you are teaching a course where you require your students to write a ten-page essay. You have thirty-five students and are wondering if generative AI could help you grade more efficiently. You might use these questions to generate the following answers:

What is my purpose and goal in using generative AI?

I want to spend less time on grading while still giving my students meaningful, action-oriented feedback. I tend to spend a lot of time revising my feedback because I am afraid it is too direct and may make students feel intimidated.
Who is my audience?

My students don’t seem to have strong feelings about the use of generative AI, but I would like to model practical and responsible uses of these tools. As such, when I give students their feedback, I want to disclose how and why I used generative AI.
What kind of information will I be sharing with generative AI?

My student’s essays are protected by FERPA, and I don’t have access to a generative AI program that guarantees that this privacy is respected. My prompts, however, aren’t FERPA protected. In addition to privacy concerns, I’m also hesitant to share student writing with generative AI because it is ill-suited for evaluating the factual information in student essays.
What specific information or product do I want to get from generative AI?

I would like AI to help me generate a rubric based on my essay assignment. I also know I can be blunt, so I want generative AI to help me take an evaluative statement like “your tone is too informal” and make it more growth oriented.
How will I prompt AI in a way that generates the output I want?

I’m not sure how generative AI will do if I prompt it to write a rubric based on my prompt. I’ll try that first, but if I don’t get helpful output, I may make my prompt more specific by adding information about the categories I want the rubric to include and the type and number of descriptions I want for each rubric item. I don’t think that there are major opportunities for biases when creating the rubric, although if I ask it to help me generate and revise my comments, there are chances that some of the adjectives it uses may have negative connotations for certain students.
Review Prompt Patterns
How will I revise and review the output that AI generates?

I know I want to review each rubric item to make sure it correlates with the assignment and that the generative AI hasn’t added additional criteria that may not be in my assignment. I also know I have some concerns about negative connotations my students could have in the language of certain adjectives or phrases that the generative AI may suggest, so I want to be sure to keep an eye out for that as I revise.

By following these steps, you realize that you can’t ask the software to grade your students for you, but you can help maximize your time by using generative AI to create a rubric for your essays. You also decide that you will use generative AI to help you brainstorm more growth-oriented language for student comments. As you work with generative AI, you know that you want to pay attention to potential factual errors or fabrications that may emerge.

Additional Resources

Flowchart created by AI and Data Policy Lawyer Aleksandr Tiulkanov
Additionally, the following works were consulted in the development of this webpage. For additional perspectives on these topics, we encourage you to review the following sources.