Thank you to the University of Maryland University Libraries for some of the content contained in this guide.
Walters, W. H., & Wilder, E. I. (2023). Fabrication and errors in the bibliographic citations generated by ChatGPT. Scientific Reports, 13(1), 14045. https://doi.org/10.1038/s41598-023-41032-5
What is actually going on when you use an AI-based tool? Check out this three-minute explanation from Hal Daumé III, Professor of Computer Science at University of Maryland, Institute Director for the Institute for Trustworthy AI in Law & Society (TRAILS).
Here are four facts about ChatGPT that will be helpful as you think about its uses.
"Generative" refers to a type of machine learning model that creates new content.
"Pre-trained" refers to a type of machine learning model that was trained on an enormous amount of data. In the case of ChatGPT, the model was trained and on hundreds of billions of words (mostly from websites) to learn patterns and relationships between words and phrases.
"Transformer" refers to part of the machine learning model that can better understand sentences because it can understand the relative importance of different words in context.
ChatGPT-3.5 and ChatGPT-4 don't have the ability to search the Internet in real time (although OpenAI and other companies are working on that!). We don't know for sure what data OpenAI used to train ChatGPT, but many AI programmers train large language models on Google's C4 dataset. Check out this Washington Post article on what websites are included in that dataset.