Skip to Main Content
Skip to main content

Academic Integrity Tutorial: How Generative AI Works

Capabilities and Limitations

Generative AI, utilizing a process called deep learning, can generate content such as images, music and text, based on patterns learned from vast amounts of training data.1 Yet, despite these capabilities, GenAI has notable limitations that necessitate human oversight and critical evaluation of its outputs. These limitations include perpetuating the biases present in its training data, producing convincing but factually incorrect outputs, and ignoring context, restricting its ability to provide meaningful analysis or original thought.

Large Language Models

In this module, we'll focus specifically on Large Language Modules or LLMs like ChatGPT. LLMs have been trained on vast amounts of text data and are very capable of generating text that resembles human-created content, making them tempting to use for various university writing tasks, which, in turn, can lead to numerous ethical dilemmas.

Interactive Video

The following interactive video describes Large Language Models, how they work and the limitations to be aware of. Listen attentively and respond to the embedded questions as you watch. You cannot skip ahead, but you can watch it at 1.5 times the speed and rewind in ten-second intervals , if needed.

Source: What are Large Language Models (LLMs)? by Google for Developers on YouTube

Test Your Understanding

Which of the following limitations are true of large language models (LLMs)?
LLMs inherit biases present in the training data, which can result in prejudiced or politically incorrect outputs.: 82 votes (3.96%)
LLMs generate responses based on language patterns rather than factual verification making it crucial to fact-check their outputs.: 112 votes (5.41%)
LLMs lack the ability to reason and cannot provide explanations for their responses beyond pattern recognition.: 29 votes (1.4%)
LLMs might generate inappropriate, nonsensical, or irrelevant responses even with well-structured prompts.: 14 votes (0.68%)
LLMs require regular updates to stay relevant as language and information evolve over time.: 13 votes (0.63%)
All of the above: 1801 votes (86.96%)
None of the above: 20 votes (0.97%)
Total Votes: 2071

1. Martineau, Kim. “What Is Generative Ai?” IBM Research Blog, April 20, 2023.