Introduction

This guide is intended to provide a brief overview of, and resources about, Generative Artificial Intelligence and Large Language Models. It is not a list of best practices but can be used as a starting point for conversations about these tools. 

What is Generative Artificial Intelligence? 

Prior to November 2022, you'd be forgiven for thinking of artificial intelligence (AI) as something in sci-fi movies or maybe the voice assistant on your phone. But then suddenly it was everywhere, or so it seemed. So, what is it? 

Artificial intelligence is the ability of computers or computer-controlled robots to perform tasks commonly associated with intelligence. Basically, if a human did it, it would be classified as intelligent. Think playing chess or analysis, spotting patterns and making predictions based on huge amounts of data. 

Advancements in AI have led to generative AI, where systems using machine learning have been trained using large amounts of data to perform, mimic or imitate tasks. This technology essentially tries to reproduce human behaviour and create new content based on what has been done before, much like predictive text on your phone. OpenAI released ChatGPT in November, one of the first major platforms to utilise this technology. 

The release of ChatGPT in particular led to speculation that it could be the end of the essay, high school English and even some jobs. However, the impact hasn't been as bleak as first thought. Many Australian and international universities haven't adopted Turnitin's AI-generated plagiarism detection tool; it's thought ChatGPT could lead to a reduction in contract cheating and even help make education better by changing the way we approach assessment and helping develop inclusive and accessible lessons. 

What is ChatGPT?

ChatGPT is a generative AI chatbot that has been developed by OpenAI. It is an example of a Large Language Model (LLM) which includes Bard and Gopher from DeepMind (Google), and BLOOM from BigScience. It is a variation of OpenAI's GPT-3 (Generative Pretrained Transformer-3) and has been fine-tuned specifically for conversation response generation and trained on a large dataset of text, allowing it to generate responses.

ChatGPT was trained on 175 billion parameters. These parameters include internet sources such as digitised books from The Gutenberg Project and Bibliotik as well as approximately 61GB of Reddit and Wikipedia pages.

On Tuesday 14th March 2023 OpenAI officially launched GPT-4. However, unlike its predecessor, it is currently not freely available for everyone to use. To access the GPT-4, you need a paid subscription to ChatGPT+. GPT-4 has several enhancements including increased creativity, more advanced reasoning, the ability to use visual input commands and the capacity to handle significantly larger amounts of text. Additionally, GPT-4 performs better than ChatGPT on a range of standardised tests. For example, GPT-4 scored among the top 10% in a simulated US bar exam, whereas GPT-3.5 was in the bottom 10%. 

GPT-4 still has many of the same limitations as ChatGPT, namely, it still makes reasoning errors and 'hallucinates' information. However, OpenAI does suggest that GPT-4 is more reliable than ChatGPT making GPT-4 a new and improved version of ChatGPT, not a leap forward in AI technology. 

How ChatGPT works 

The answer to how ChatGPT works is in its name, specifically GPT, or Generative Pre-trained Transformer. 

• Generative reflects its ability to produce text that resembles natural human language text
• Pretrained means that the model has been trained on a specific dataset; this includes a large amount of conversational data 
• Transformer refers to the machine-learning architecture 

This means that ChatGPT works by predicting answers to questions, based and all of the data it has been fed -- similar to studying and cramming before a test a high school math test. ChatGPT didn't come out of nowhere though. The technology that runs ChatGPT has been around for much longer, just with much less publicity. While the model for ChatGPT is not new, its functionality is based on OpenAI's GPT line of AI models, the interface, layout and user experience, and this is what has led to its rapid adoption. Upon opening the app, the interface has strikingly similar to other messaging and chat software which allows users to feel familiar and safe using. 
 

Thank you to Harper College Library for allowing us to reuse parts of their guide Teaching and ChatGPT. 

If you have any feedback or comments about the guide, or something you think should be added, let us know! We'd love to hear from you. 

Please email us and we will respond when able.