Using Generative AI

Many generative AI tools are available, including the examples given on the Types of Generative AI page. Another option is to run an LLM locally.

Generative AI tools are trained on masses of data, and they create models based on these data. When you give an AI tool a prompt, it uses the data model to generate a response.

Some generative AI tools (such as ChatGPT or DALL·E) require a prompt in the form of text. Other tools may have a series of prompts for users to select in order to generate content, or they may require a file to be uploaded.

Developing effective text-based prompts

Use the CLEAR Framework to remind you how to generate good prompts in text-based tools such as ChatGPT.

The CLEAR Framework (Category and Description columns in the above table) is shared with permission of Leo S. Lo from CLEARer Dialogues with AI: Unpacking Prompt Engineering for Librarians. Choice 360, September 9, 2023.

Hall & McKee suggest modeling prompts on the following formula:

[Engaging Context] + [Relevant Background Information] + [Clear Goals] + [Desired Response Format and Constraints] + [Specific Questions or Prompts]

Example 1

[Background:] Act as a pharmacy student [Context:] who is interested in creating a promotional campaign about flu vaccinations in a community pharmacy. [Goal:] You hope to increase the rate of flu vaccinations by 15% and it should not require a lot of money to run the campaign. [Response Format & Constraint:] Provide 5 detailed examples of campaign ideas bearing in mind that the campaign duration is one month.

Example 2

[Background:] I am an instructional designer [Context:] interested in creating a short course that shows business students how to find industry research, including competitors, TAM, SOM, and SAM. [Goal:] I am looking for a clear and concise storyboard demonstrating how I could scaffold my tutorial. [Response Format:] Provide a detailed list of business concepts paired with examples so that I can markup my online tutorial. [Constraint:] Limit your response to 5 scaffolded concepts and assume the tutorial will time out after 60 minutes. (Hall & McKee)

Benjamin Hall & Jimmy McKee (2024) An early or somewhat late ChatGPT guide for librarians, Journal of Business & Finance Librarianship, 29:1, 58-69, DOI: 10.1080/08963568.2024.2303944 

The video below describes how you can use generative AI at the start of a research project.

Prompt engineering guides like this one from Learn Prompting show different ways to use generative AI, including for academic work and studying. Some examples include:

Writing/research

• Generate ideas

• Brainstorm research topics

• Change citation styles

• Edit and proofread

• Translate text into different languages

Learning/studying

• Help explain complex concepts

• Summarize information from your class or research notes

• Practice learning languages

When reviewing the responses from generative AI tools, be mindful of some limitations: 

• AI tools are limited by their knowledge base (e.g., the free version of ChatGPT does not have access to very recent information).

• Responses are not replicable. It’s probabilistic, so you will not get the same answer twice.

• May generate false information (referred to as AI hallucination).

• May generate biased information based on their training data.

• Ungrounded AI tools do not give references/citations to sources, and any references provided may be inaccuate.

• Can be generic and lack true understanding of a subject area.

For more information, see Ethical Use and Evaluating AI Content.

While online services such as ChatGPT are more familiar, it is also possible (and increasingly easy) to run smaller AI tools on a desktop or laptop computer. This article describes how to run an LLM: 5 Easy Ways to Run an LLM Locally. The process is slower and not as advanced as the online services, but there are tasks for which the power of the online services is not necessary.

Using a small local LLM reduces the environmental impact and avoids the privacy concerns associated with the online services; the ethical considerations and the need for evaluation of results are the same. Running your own can be a good way to learn about the workings and the weaknesses of LLMs, since a small one is not as good at hiding its limitations as a large one. Help with running a local LLM is available from the Digital Scholarship Centre.