Generative AI FAQ

What is new about generative AI is the sophistication shown in the reasoning and computation behind the replies to your messages. Generative AI can perform surprising tasks that would be difficult to replicate with any other computing technology. The buzz isn’t because these tools can write simple poems and answer quiz questions—it is the deeper computing capabilities and accessibility of these capabilities that are the sources of the excitement. Generative AI can craft a variety of sophisticated outputs including:

• Reasoning about what would need to change in a world without odd numbers
• Analyzing and critiquing the assumptions in a document
• Completing step-by-step tasks

For example, generative AI can perform tasks like turning data from a given source into a set of visualizations, inserting each visualization into a slide in a PowerPoint presentation and then replying to your message with a link to the PowerPoint presentation it created based on your instructions.

One of the best ways to learn about generative AI and its capabilities is to see it in action. The videos below demonstrate some examples of using generative AI to accomplish different tasks. The first video demonstrates ChatGPT, a chat-based generative AI model, in action.

This second video demonstrates Microsoft Bing’s Image Creator, which is an image-based generative AI tool.

To access generative AI tools, you typically need to visit the website of a given tool. Many of these tools will then prompt you to create a login. As these tools become more sophisticated and widespread, it is increasingly common to find them embedded into popular software tools and services, such as search engines and social media platforms like Google or Meta (Instagram and Facebook’s parent company). Most of these search engines and social media platforms currently require you to opt-in to using their generative AI tools and capabilities.

For an example of how to access generative AI tools, check out the video below excerpted from Vanderbilt University Professor of Computer Science Jules White’s free, self-paced prompt engineering course. In this video, Dr. White walks through how to access ChatGPT and write a basic prompt.

While every generative AI model is unique, there are some basic similarities in the way that these tools are developed and trained. Current generative AI models are essentially sophisticated pattern recognition and imitation tools. Generative AI tools are typically trained on large amounts of sample data (think hundreds of gigabytes), which are often collected through web-scraping tools. Through various deep learning techniques, these tools are trained to recognize patterns in their datasets and use those patterns to create unique outputs. These outputs are often intended to realistically mimic elements of human communication like writing, music, art, or speech. As these models are trained, they typically receive human feedback on the plausibility and quality of their outputs. This training process is iterative, and these tools are often continuously updated and refined based on human feedback.

Another key element of these models is that their outputs include some randomness. Each time you put a prompt in, the model will generate new and different ideas. This means if two people use the exact same prompt, the outputs will not necessarily be the same.

As a user, you can interact more strategically with these tools by keeping in mind the elements of pattern recognition, randomness and variation in output. Generative AI tools generate content, not facts, and there is no distinction between the two inside of most current tools. As such, when using these tools, you will always want to check the output for errors.

For more information on how these models work, check out the video below from Vanderbilt’s Data Science Institute, which offers a close look at how these generative AI models were trained.

ChatGPT, which is currently one of the most well-known generative AI programs, is an example of a Large Language Model. Large Language Models are a subset of generative AI tools that target language patterns. Essentially, these large language models work by taking your input and trying to generate the next word or phrase. Then they will take that generated word and add it to you what you originally input, and and so on. For example, if you provide the prompt “Mary had a little,” the output will likely include “lamb.” Likewise, if you provide the prompt “Roses are red,” the output will likely include “violets are blue.”2

For additional information on how Large Language Models work, check out this video below, which is excerpted from Vanderbilt University Professor of Computer Science Jules Whites’ free, self-paced Prompt Engineering course.

Although it appears that you can write a prompt about anything and get a result, some results will be more useful than others. Generative AI tools are exactly what they sound like: tools. Certain uses of the tool simply don’t make sense. You probably wouldn’t scramble eggs with a hammer. Similarly, there are uses of generative AI that don’t make sense. For example, using generative AI tools like a search engine or as an absolute source of truth is typically not the most effective use of these tool.

Knowing that these tools make errors can help you select what types of problems you attempt to solve with them. You may want to use generative AI to tackle problems where either:

1. a partially correct solution is valuable.
2. checking if the solution is correct isn’t time consuming or expensive.

For example, having generative AI help you brainstorm ideas is a reasonable use case, particularly for domains that combine multiple diverse topics and don’t have a single right answer, such as combining cuisines from Ethiopia and Uzbekistan that are vegetarian. Another example would be generating solutions to a crossword puzzle. It is easy to check if the solutions are correct by simply seeing if the generated words fit into the constraints of the puzzle. In contrast, having these tools generate output that you can’t easily check for correctness isn’t usually a good idea, such as having them produce a translation of “Computer Science” into Babylonian Cuneiform.

For more guidance on when to use generative AI, check out our Tips on Using Generative AI page.

Just like most tools, generative AI tools require training to use them effectively. Their deep capabilities are likely to cause dramatic changes in the world over the next few years and will generate changes in how faculty, staff and students approach education. As such, it is important to think critically about how to employ these tools.

For example, what you write and how you write it directly impacts the quality of the output. Many of the same techniques for effectively communicating with humans are also important to use when communicating with generative AI. These techniques include:

• Writing clearly
• Providing step-by-step instructions

One of the best ways to get the most of your use of generative AI is to learn techniques for writing prompts, which is part of the discipline of “prompt engineering.” Generative AI was taught to identify and respond to patterns in human language. By understanding patterns that it responds to and how it responds, you can more effectively structure the writing in your prompts and solve more complex problems with these tools. Often, you can get significant improvements in the quality of the output by changing how you word and structure your requests. For a comprehensive introduction to prompt engineering, check out Vanderbilt University Professor of Computer Science Jules White’s free, self-paced prompt engineering course.

Additionally, this website provides several resources to help you build your comfort and literacy with generative AI tools, including:

• Tips for using generative AI
• Prompt engineering
• Resources for faculty
• Resources for students
• Resources for staff

Generative AI tools can augment and amplify human creativity and reasoning. They can form an “exoskeleton for the mind” that helps you find new ways of solving problems, solve larger problems than before or provide a new medium for artistic expression. Approaching these tools as “Augmented Intelligence,” rather than as an artificial substitute for human reasoning, will help you use them effectively and responsibly. When you use the tools, make sure that you use them in a way that augments and amplifies your own unique human spark and doesn’t diminish it.

Due to the diverse capabilities of generative AI tools, there are a plethora of ways to adapt the capabilities of these tools to different situations. Across Vanderbilt, researchers across disciplines are developing unique strategies for using these tools within the context of their research, teaching, and learning. For example, for those who have work that requires coding knowledge, generative AI can serve as a powerful tool for troubleshooting code. For those who work in writing-heavy disciplines, generative AI provides new avenues conceptualizing and theorizing rhetorical patterns and genre conventions. For more resources for exploring how researchers from different academic disciplines are considering the implications and uses of generative AI, check out the following resources:

• Generative AI News and Features
• The Robert Penn Warren Center for the Humanities’ Generative AI and the Humanities Seminar
• Vanderbilt Arts and Science Artificial Intelligence Grand Challenge Initiative
• The Data Science Institute’s Workshop series
• The Data Science Institute’s generative AI showcase
• “I Secretly Let ChatGPT Take My Final Exam” by Vanderbilt computer sicence professor Dan Arena