Skip to Main Content

History 101-7 Genghis Khan: History and Myth

What is Generative AI

Below are terms and definitions for concepts and processes often associated with Generative Artificial Intelligence.

Artificial Intelligence (AI) a subfield of computer science, is the simulation of human intelligence processes or tasks by machines or computer systems.

Algorithm is a sequence of rules given to an AI machine to perform a task or solve a problem.

Generative Artificial Intelligence (Generative AI or GAI) is a type of artificial intelligence used to create content, including text, images, video, audio and code. A generative AI system is trained using large amounts of data so that it can find patterns for generating new content based on a prompt or question. 

Hallucination refers to false, inaccurate or imaginary output or responses generated by the AI tool or LLM.

Large Language Models (LLMs)​ are AI systems or models trained on large quantities of text in order to interpret prompts and generate human-like text-based outputs. GAI tools like OpenAI's ChatGPT, Google Gemini and Microsoft Copilot are some well-known examples.

Training Data is the data used to train the model, in other words, to enable it to find patterns and create new content. It is sometimes referred to as the "corpus" or information universe that a GAI tool or model relies on to generate new output or content in response to your prompt. 

More about training data

Training data for various tools may come from text and other kinds of data across the open internet, from licensed or closed sets of information, or from information created by the entity, corporation, etc. that developed the tool. 

Garbage in, garbage out

Biased, false or inaccurate text and other data can generate false claims as well as perpetuate those harmful biases. However, AI tools are not always fully transparent about what is included within their training data. Some information can be found under "About" pages or FAQs. 

In simpler terms, GAI...

  • "...is a word completion tool... predicting what the next word in a sentence should be so it can write a paragraph for you, what an image should look like based on a prompt."   --Ethan Mollick

    • Ghassemi, M., Birhane, A., Bilal, M., Kankaria, S., Malone, C., Mollick, E., & Tustumi, F. (2023). ChatGPT one year on: Who is using it, how and why? Nature, 624(7990), 39–41. https://doi.org/10.1038/d41586-023-03798-6

  • predicts text, rather than producing verified content.  It may hallucinate, or make things up.

  • Is not an encyclopedia of knowledge nor a search engine that looks for information or sources. 

  • Is not a discrete work by an author - your prompt can generate different information each time

  • Is not unbiased or neutral.

GAI can, however: 

  • Assist with writing tasks, including brainstorming, composing an outline, grammar checks, summarizing and generating research questions and keywords for your research purposes

  • Provide generic writing or knowledge that may lack specific details or accuracy

  • Write code, analyze data

Negatives & Limitations

Inaccuracy: misinformation, fabricated information, bias
Ethical considerations: privacy, intellectual property and copyright, environmental concerns and labor
Insentience: doesn't know, discern, think or feel
Potential effects on learning: demotivation, impairs thinking, exposure to superficial or inaccurate information
Free vs. Fee: usage limits

Bias and Harm

  • Dominated by some perspectives or others in terms of race, gender, ethnicity, ability
  • "Black Box" problem.  The model's algorithms and training data are not visible or publicly known.

Types of GAI Tools

Types of GAI Tools

Some of the more common types of Generative AI tools or LLMs, based on the types of data they train on.

Chat / Conversational:  OpenAI's Chatgpt, Google Gemini, Microsoft Copilot

Images and Graphics: Adobe Firefly, Canva, Stable Diffusion, Midjourney

Video / Audio: Synthesia, PlayHT, Soundraw

Research / Literature Review: Elicit, Consensus, Research Rabbit, Perplexity

Summary of Features (not comprehensive!)

Additional updates from Generative AI Product Tracker by Ithaka S+R

Consider the Tool and your Purpose
Consider the AI tool itself.  Is information shared about what is in its training data or the corpus of content it draws from?  Is it the best tool in terms of:
  • your purpose (personal, research, writing..)
  • currency (is it connected to live internet and if not, how recent is its training data?)
  • subject coverage (how comprehensive is the coverage in a specific academic discipline compared with library research databases or other research tools?)
  • how accurately does it perform compared with other AI tools, library research databases, search engines.
  • Additional information at Evaluating Generative AI Tools from SUNY-Albany Libraries

Evaluating AI-Generated Content

●Can claims be verified?
●Any info missing or anyone missing from conversation?
●Inherent biases?
Consider cross-checking against credible or scholarly sources (not against another LLM)
 
Credible sources discoverable through NUsearch / article databases, Google Scholar, or internet search engines
●If web pages, URL may provide some insight: edu., gov., org. 
●Authored by experts
●Blogposts, News, Govt web pages, trade organizations, Britannica, etc.
●Check against links to sources provided in output as responses that cite actual sources may still contain inaccuracies. 
Below is an example from Google Gemini.  Even when citing a Wikipedia article, its generated content still included incorrect name and dates. 
 

Wikipedia entry that is referenced above:

Verified by the source below:

Library of Congress, Washington, D.C. 20540 USA. “About This Collection | Margaret Bayard Smith Papers | Digital Collections | Library of Congress.” Collection. Accessed Feb. 3, 2025. https://www.loc.gov/collections/margaret-bayard-smith-papers/about-this-collection/.

About Scholarly Sources

Scholarly Sources
●Authored by subject experts, scholars
●Include bibliographies, references
●Editorial process for quality, accuracy
●Books, chapters, encyclopedias from academic presses
●Peer-reviewed academic journals

 
Scholarly Research Databases (Find these and many others at NU Library Databases A to Z or ASK a librarian for help!)
Edited, compiled and published by academic publishers or scholarly societies and associations.

Considerations

●Your instructor may not approve of GAI
●Students need to check the syllabus or ask the instructor
●Always declare your use of GAI to your instructor
●Be aware of Northwestern rules against cheating and plagiarism

Northwestern University Academic Integrity: A Basic Guide

Writing effective prompts: Various methods

The CLEAR method (created by Leo S. Lo):

C: Concise: Brevity and clarity (eliminate unnecessary words)
L: Logical: Structured and coherent
E: Explicit: Clear specifications
A: Adaptive: Flexibility and customization
R: Reflective:  Continuous evaluation and improvement

Lo, Leo S. “The CLEAR Path: A Framework for Enhancing Information Literacy through Prompt Engineering.” The Journal of Academic Librarianship 49, no. 4 (July 1, 2023): 102720. https://doi.org/10.1016/j.acalib.2023.102720.   Article online

TRACI Method

Task, Role, Audience, Content, Intent

{Structured} Prompt. “Download the Free User’s Guide to the TRACI Prompt Framework for ChatGPT.” Accessed October 11, 2024. https://structuredprompt.com/free-traci-users-guide-white-paper/.

Six Components

Task, Context, Exemplars, Persona, Format, Tone

Edwards, A. (2024, August 29). The Perfect Prompt: 6 Essential Components for Creating Effective AI Prompts. DocsBot AI. https://docsbot.ai/article/the-perfect-prompt-6-essential-components-for-creating-effective-ai-prompts