Skip to Main Content

Generative AI (ChatGPT) for Faculty

Introducing Generative AI (ChatGPT and others) and providing resources and tools for faculty. Based upon a guide originally created by Tacoma Community College Library and shared with a CC BY SA 4.0 license.

Welcome to Generative AI

This guide is a (currently under construction) basic introduction to large language model generative AI. This is not meant to be a definitive guide, nor is it to be considered "complete." I appreciate any feedback, comments, and requests for how this guide can be more useful to you.

Please contact me via my email, found at the bottom of this page. ~Claire Ehrlich

See the research guide that the library has created for students about Generative AI. Please feel free to share it with students. Send suggestions or feedback to Claire:

Definitions

Although the first conversation about artificial intelligence is credited to have started with Alan Turing in 1950, the term "artificial intelligence" (AI) was coined by John McCarthy in 1955 referring to the science and engineering of making intelligent machines. AI broadly refers to any human-like behavior displayed by a machine or system. AI makes it possible for machines to learn from experience, adjust to new inputs and perform human-like tasks. 
Sources and Further Reading
Machine learning is a subfield of artificial intelligence that gives computers the ability to learn without explicitly being programmed. Machine learning is behind chatbots and predictive text, language translation apps, the media your streaming services suggest to you, and how your social media feeds are presented.
Sources and Further Reading
Generative AI refers to a type of artificial intelligence that is capable of generating new content, such as images, music, or text, that is similar to or indistinguishable from content created by humans. Generative AI models are designed to learn patterns from large datasets and use that knowledge to generate new content that is similar in style and structure.* ChatGPT is just one example of generative AI, and is one of four broad-category types of AI known as "limited memory" AI. Limited memory AI is distinct from "reactive" AI, which is unable to build memory or store information for future, it can only respond to the current input.
Sources and Further Reading
Large language models (LLMs) use deep learning algorithms to process large amounts of text. They are designed to understand the structure of natural language and to pick out meanings and relationships between words. These models are capable of understanding context, identifying and extracting information from text, and making predictions about a text’s content. They are trained on extremely large datasets. Chatbots, like ChatGPT, version GPT-3.5, was trained on a 45 terabyte dataset which is equivalent to over 292 million pages of documents, or 499 billion words.
Sources and Further Reading
A chatbot is a computer program designed to simulate conversation with human users, often through text-based interfaces like messaging apps, websites, or voice assistants. Chatbots are typically powered by artificial intelligence (AI) and natural language processing (NLP) technologies, allowing them to understand and respond to user prompts in a human-like way.* ChatGPT is an example of a chatbot.
Sources and Further Reading
"Prompts" are what we input into chatbot AI, such as ChatGPT, to get it to provide a response or perform a task for us (output). Prompt engineering is the process of planning and refining our prompts to result in the outputs we are satisfied with. At a basic user level, this includes critical evaluation of the quality of the outputs we receive, in the context of the quality of prompts we input, and includes meta-reflective analysis of our own thinking process. This process has powerful teaching and learning potential. "Prompt Engineer" is a job in itself and also a desirable skill within other jobs. Demand for prompt engineering skills is going to grow as generative AI continues to advance.
Sources and Further Reading
GPT are the abbreviated initials for "generative pre-trained transformer"

ChatGPT explains GPT:

ChatGPT user prompt:
please explain generative pretrained transformer to someone who is not an AI expert.
ChatGPT output:
"A Generative Pretrained Transformer (GPT) is a type of artificial intelligence that has been trained to generate text that looks like it was written by a human. The 'pretrained' part means that it has already learned a lot about how language works, so it can understand things like grammar and sentence structure. The 'transformer' part refers to the way it processes language. It breaks down text into smaller parts, called "tokens," and then looks at how those tokens are related to each other to understand the meaning of the text. The 'generative' part means that it can use that understanding to create new text on its own. So if you give it a prompt like 'write a story about a magical castle,' it can come up with a whole story all by itself, using the knowledge it has learned about language. It's like having a really advanced robot assistant that can help you write things, but instead of just following your commands, it can actually come up with its own ideas and write them down for you."

------------------ One of the reasons that I chose to let ChatGPT describe GPT is that I can control the level of complexity through the criteria of my prompt. On many sites that include information about what GPT is and a bit about how it works ("a bit" because for ChatGPT this is proprietary information in the details), the level of complexity is too steep for many people, like me, who do not already have a background in in AI. I could also ask the chatbot to describe it to a 12 year old or a 5 year old, or more nuanced, such as "for a college freshman just starting their AI degree"! The other reason is to demonstrate the value of AI as a study aid. This can empower students to generate information about a complex concept on their own terms, as a way to scaffold their own learning and understanding. (Note, sometimes outputs are factually wrong, but for well-established concepts, such as GPT, outputs are more likely to be consistently factually reliable as long as the user knows enough about the concept to compose a valid prompt.)
Sources and Further Reading

How Generative AI Works

Knowing how Generative AI works makes it easier to understand what gAI chatbots are good at (and what they're not).

Generative Artificial Intelligence chatbots like ChatGPT or Copilot are called Large Language Models, because they are "trained" on huge amounts of data, usually a significant percentage of the information that's already been published on the internet. The models use this vast amount of data to learn to predict what is most likely to come next.

This means:

  • gAI is really good at producing information that seems correct. The outputs gAI produces are nearly always extremely plausible, at least on an initial quick read. 
    • It is largely not concerned about producing information that is actually true. This is how "hallucinated" sources come about: it combines existing experts and books or articles into reasonable-sounding but not-real books, articles, and websites. 
  • gAI is really good at producing text that sounds good. It has excellent spelling and grammar, and on a surface level it makes perfect sense.
    • Because it is predicting what is "most likely" to come next, it is unlikely to produce anything surprising or truly original. It is remixing what already exists (this remixing of existing material has also led to several copyright lawsuits).
  • gAI reflects the data it was trained on. That means that it is really good at "taking the temperature" about a topic: getting a sense of the scope of an issue and what the internet as a whole thinks about it.
    • That also means it contains all the errors, outdated information, and human biases present on the internet. Rather than being a neutral, unbiased third party, AI reflects society back to us: ChatGPT and other gAI tools have been caught perpetuating pervasive stereotypes and biases about race and gender. The tools might also follow your lead if you introduce bias into your prompt and show you what you want to see, rather than available information that might contradict your prompt. 
  • gAI is excellent at predicting plausible answers, which requires a lot of computing power.
    • gAI therefore does not devote much computing power to remembering what was already said. While the tools are rapidly getting better at this, their ability to remember what they have already said and remain consistent with it is limited.
      • gAI will generate its answer from scratch every time, so asking it the exact same question will not always get you the exact same answer. 
      • While the AI will attempt to remember what you tell it within a "conversation," whatever it learns in one conversational thread will not be carried over into another. 
      • ChatGPT in particular seems to have problems with arithmetic. For example if you ask it for an outline of a 60-minute presentation, it might give you an outline where all of the time breakdowns add up to 70 minutes. If you ask it to make a response shorter by about a third, it might take out only one sentence. 

Learn to use ChatGPT

ChatGPT is like Kleenex or Band-Aids: a specific generative AI that many people use to refer to all generative AI. It's a good one to start your AI journey with, but by no means the only one (see below).

There are plenty of tutorials out there on how to use ChatGPT! Here is just one for beginners; the YouTuber starts the viewer off with how to create an account. 

video: "ChatGPT Tutorial - A Crash Course on Chat GPT for Beginners" by Adrian Twarog. Standard YouTube license applies

Generative AI beyond ChatGPT

While ChatGPT is talked about quite a bit, it is only the most widely-known of several generative chatbots. Here is a partial list of other natural language processor AI:

Creative Commons License
This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License. Except where otherwise noted, the content in these guides by Mohawk Valley Community College Libraries is licensed under CC BY NC SA 4.0. This openly licensed content allows others to cite, share, or modify this content non-commercially, with credit to MVCC Libraries.

When reusing or adapting this content, include this statement in the new document: This content was originally created by Mohawk Valley Community College Libraries and shared with a CC BY NC SA 4.0 license.

Mohawk Valley Community College Libraries - 1101 Sherman Drive, Utica, NY, 13501- P. 315.792.5561