Teaching and Learning with Generative AI

As Generative AI tools continue to evolve, it is important to provide the Capilano University community with guidelines for their use in responsible and ethical ways. As faculty are tasked with determining if and how these tools are to be used in the classroom, the guidelines also provide language and teaching and learning suggestions that support faculty members’ decisions to include or exclude the tools in their own classrooms and curricula.  

About Generative AI 

Generative Artificial Intelligence (GenAI) or Large Language Models (LLMs) are tools that use deep learning models to produce new content such as text, visual content1069 (images/video), and audio. They do this by analyzing large data sets, learning patterns and relationships, and generating new content that aligns with these patterns. It’s important to remember that Generative AI does not understand concepts and meanings as humans do, but instead imitates human behaviour based on probability.  

Generative AI is math, not magic. Generative AI works by learning patterns from massive amounts of data and then creating new content that follows those patterns. At its core, these systems mimic neural networks that process information through layers of interconnected nodes. During training, the AI analyzes examples (like text or images) and adjusts internal parameters to capture statistical relationships between elements. For large language models specifically, the system learns to predict what word should come next given previous words, developing a deep statistical understanding of language. When generating new content, the AI starts with some input (like a prompt) and uses probability calculations to determine the most appropriate next elements to add, creating a sequence that statistically resembles what it learned during training, without simply copying its training examples. 

It’s important to understand that LLMs do not reason in the way that we understand reasoning.  It cannot create or imagine beyond its training data. When we assign these verbs to LLMs, we are anthropomorphizing them, that is attributing human-like qualities to non-human entities.