Researchers and research support staff

Generative AI (GenAI) is a type of artificial intelligence technology that can generate new content, such as text, images, audio, or video, based on patterns it has learned from existing data. This class of models includes Large Language Models (LLMs), among others. The most widespread use case for current GenAI applications involves users interacting with a generative model by providing it with an input “prompt”, whereby the model generates text, images, etc. The model’s abilities are acquired as a result of training on very large datasets. These datasets could be a collection of research papers, books, images, or any other form of data. The model identifies patterns, structures, and correlations within this data during its training phase.

Once the model is trained, it can generate content that mimics the structure and style of its training data and input. For instance, if a model is trained on a dataset which includes funding proposals, it can in principle generate a draft proposal based on the patterns it has learned. Since the model is probabilistic (i.e., produces the most statistically likely output), the output can change if the prompt is worded differently or even if the same prompt is used multiple times. While GenAI can produce impressive results, it does not "understand" the content it produces or is trained on in a way humans do. It is more accurate to consider such models as reacting to user input based on recurrent patterns in data. Therefore, a model does not recognize if its output is factually (in)correct, misleading, manipulative, offensive or in other ways legally or ethically objectionable. Therefore, while it can be a powerful tool, any GenAI-generated content should always be reviewed and edited by a human.