Generative AI

What AI is (and what it isn't)

Generative AI

When most people talk about AI, they are thinking of generative AI - machine models that are designed to create new content, such as text, images, videos, audio, presentations, or other data forms. When a users enters a prompt into the generative AI interface, the AI uses algorithms to calculate what the user wants to see based on the prior inputs the model was trained on before being released. For example, a image-generating AI model that has only ever been trained on photographs of real life will never be able to create a cartoon-style image, because it has no frame of reference for what a cartoon is. Generative AI models can only work withing the data sphere they were trained on. 

The most common type of generative AI models are Large Language Models (LLMs) that "talk" to the users in response to typed or verbalized prompts. LLMs are very good at sounding human in their responses, because they have been trained on hundreds of thousands of examples on how humans communicate. They have also been coded to respond like a customer service representative - being polite, offering additional solutions or alternatives, and mimicking the tone of the user. The LLM is merely replicating what it has seen before and matching the tone of the prompt. 

Because generative AI models can only work within the data sphere they were trained on, they are particularly prone to bias. There have been many cases where a company or organization deployed an AI tool and later had to discontinue it because the AI learned to be biased, like the Amazon AI hiring tool that taught itself to favor men, or the social media chatbot that became a Holocaust-denying racist. When the AI model's training data contains bias (whether implicit or explicit), the AI model recognizes that pattern and repeats it. 

14 Real AI Bias Examples

 

Machine Learning

In machine learning, an AI tool is trained on large amounts of data in order to recognize patterns. From there, the AI can sort data, calculate statistics, make predictions and recommendations, or take actions (known as agentic AI). This type of AI is not intended to create anything new, only analyze and report. Theoretically, using AI to sort data eliminates unconscious human biases, although the AI itself can be biased, as shown above. Machine learning can also have unintended consequences as the AI learns what to prioritize and what to ignore. For example, a warehouse robot that uses machine learning to move products quickly may end up breaking products if that means the product gets to its destination faster. A human is able to reason that throwing a glass vase to get it to the other side of the room faster is a bad idea, but a machine needs to be told that. 

Natural Language Processing is a subset of machine learning in which an AI tool is able to understand human text or speech, and determine sentiments. This technology underpins an AI tool's ability to categorize written survey responses, function as a chatbot, and understand LLM prompts. 

 

---

Prompt Engineering

All generative AI tools rely on prompts to tell them what to do - the content AI generates for you can only be as good as the prompt you give it. The process of creating and refining prompts to get the best results is referred to as prompt engineering. Like research in general, creating prompts for AI is an iterative process. Adapting your prompts involves experimenting with different formats, phrases, tones, and perspectives. This can also include rewriting your prompt to be more concise or specific based on the results you got from your initial prompt.

To get the best results from an LLM, feed it S.C.R.A.P.S!

The SCRAPS framework was developed by Jessica Cowden for use at Saint Elizabeth University. It is openly licensed under CC BY-NC 4.0.

If you plan on feeding some of the same SCRAPS to an LLM multiple times, try out Gemini Gems (free with your SEU email). Gems are programmed to respond with the same role/audience/purpose/content every time you start a new chat. There are some pre-built ones, like a learning coach, writing editor, and brainstormer, but you can also create your own. 

 

---

General Considerations

• NEVER put personal or sensitive information into an AI software. LLM's often use your chat history to continue training the model, so you never know where your info will end up. 
• You can play around with genAI, ask it to generate ideas for you, and ask it for help, but remember that you are ultimately responsible for checking its output. If you choose to use the information an LLM has given you, you are responsible for verifying that the information is correct and accurately represents what you want it to.
• Always check your syllabus or ask your professor before using AI to help with your assignments. Each professor will have different policies and may allow different uses. 
• If you are using genAI for your papers or research, you must cite it. See our Citation & Style Guide for info on how to do so.
• To get the best answers, use neutral language. AI models always want to agree with you, so be aware of the tone you use in your prompts. If you ask an LLM "Should I quit my dumb boring job and follow my dream of becoming a glamorous billionaire influencer?", it will tell you to follow your dream. If you ask "Should I quit my job and become an influencer?", it will tell you to stick with the stable job. 
• ChatGPT and other LLM's have gotten much better at reducing hallucinations since their start, but they are not fully error-free. Some models are able to actively search the internet for answers, while others are not. Ask the AI to give you links to its sources (or search the web yourself) to verify information is correct and avoid citing sources that don't exist. 
• Avoid getting attached to your LLM of choice. Different models have different strengths and weaknesses, so test multiple models to see which one works best for the task at hand.
• AI models do not have sentience, opinions, or brand new ideas. If you address the AI as if it were sentient, the model will pick up on that and adjust its output to match how you perceive it. 

 

---

AI Tools

---

Additional Resources

CLEAR framework for prompt engineering by Leo Lo at the University of New Mexico

Prompt Engineering Guide
Provides information on prompting techniques, different AI agents and tools, ways to use AI, risks of AI, guides to genAI, and research on AI.

AI Tools and Resources Guide by the University of South Florida Library

Artificial Intelligence (Generative) Resources by Georgetown University

Artificial Intelligence Now: ChatGPT + AI Literacy Toolbox by Florida International University

 

Updated 11/17/25