What is Generative AI?

Generative AI refers to artificial intelligence systems that can generate new content—such as texts, images, audio, and video—in response to prompts by a user, after being trained on an earlier set of data. Platforms like Dall-E and NightCafe produce digital art and images that appear like photos,, and tools like Synthesia allow users to generate videos from text, using AI avatars. Large Language Models (LLMs), on the other hand, generate text by predicting the next word based on patterns learned from vast amounts of data.

In November 2022, OpenAI released ChatGPT, a powerful natural language processing model that uses a large language library comprised of data scraped from the internet and other sources to write answers to questions, summarize and translate text, and write code. While ChatGPT is currently available as a free service, its popularity has caused it to reach capacity at times, making it difficult to access. In February 2023, Open AI released ChatGPT Plus, a subscription service that provides faster response times and early access to new features.

In March 2023, Open AI released GPT-4, a more powerful LLM with the ability to take in both text and images as inputs (image inputs not yet publicly available as of May 2023).

It is worth noting that other text generators have also recently emerged, some of which combine chat with internet searches, such as Bing AI and perplexity.ai. Google has released its own chat platform called Bard as well (not yet available in Canada).

As the number of tools available on the market and their capabilities is expected to change rapidly, it is important to consider the significance of generative AI rather than only focussing on specific tools, as these technologies are changing rapidly (Veletsianos, G. 2023). Still, it is useful to be aware of at least some of the existing tools and their capabilities, when designing courses and assessments. In the “Additional Resources” section of this page you can find a list of various types of generative AI tools, as of mid-2023.

What are the strengths and limitations of Generative AI tools

Platforms such as ChatGPT and other generative AI writing tools can write articles, essays, blog posts, and other texts. Many can also write code in various programming languages. However, they are not able to fact-check the text or code they generate.

The language model underlying early versions of ChatGPT is capable of passing medical exams, MBA courses, and can almost pass bar exams (Bommarito, M.J. and Katz, D.M., 2022). Open AI has reported that GPT-4, a more powerful language model, performs significantly better on various exams, including the Law School Admissions Test and the Graduate Record Exam. The free version of ChatGPT has been trained on data that only goes up to 2021, and is not connected to the internet, so it cannot answer questions about events after that. However, ChatGPT Plus subscribers have access to functionality that connects the chatbot to the internet.

ChatGPT produces text that sounds very authoritative, but it is sometimes inaccurate. It is important to remember that LLMs such as ChatGPT work by predicting the likely next word based on training data, and that may mean that the sentences that are produced make incorrect claims. ChatGPT also frequently provides seemingly plausible but entirely made-up references to scholarly and other works. Note, however, that some text generator tools, such as Bing AI and Perplexity AI, are connected to the internet and therefore can be used to write text with references to real sources.

It is expected that there will be significant improvements to all generative AI tools in the near future, including in the accuracy of information that is produced. These new systems will likely make mitigation and detection strategies more challenging as they will be multi-modal, allowing people to input images, text, and data, and then generate a mixture of outputs.