Skip to Main Content
The University of Manitoba campuses are located on original lands of Anishinaabeg, Cree, Ojibwe-Cree, Dakota, and Dene peoples, and on the National Homeland of the Red River Métis. More

Using Generative AI for Library Research

A resource for the library workshop: Empowering Scholars: Using Generative AI for Library Research

Definitions

BLACK BOX:  a system or model whose internal workings are hidden from the user or operator, making it challenging to understand how it arrives at its decisions or predictions.

DEEP LEARNING: A subset of machine learning that uses neural networks with multiple layers to model and solve complex problems.

GENERATIVE AI:  A branch of artificial intelligence that focuses on creating new data, such as text, images, or music, using machine learning techniques.

HALLUCINATIONS: Occur when an AI system generates information that, while it may seem plausible, is actually inaccurate or nonsensical

LARGE LANGUAGE MODELS: Sophisticated AI systems, powered by deep learning and trained on extensive text data, capable of comprehending and generating human-like text for various applications.

NEURAL NETWORKS: Computational models inspired by the human brain, used in AI to process and analyze data.

PREDICTIVE AI: subset of artificial intelligence that uses historical data and machine learning techniques to make informed forecasts and predictions about future events or outcomes, helping businesses and systems make proactive decisions.

Citing Generative AI Tools

While advice may vary, citing the use of generative AI tools in your research output is important, and a way to acknowledge the sources of generated content that helped guide your process. The very nature of genAI tools makes it difficult to provide complete transparency or reproducibility so it is encouraged that you create and save any chat transcripts. 

Visit the below website for specific instruction on how to cite genAI tools:

Hal Loewen

Profile Photo
Hal Loewen
he/him
Contact:
(204) 599-9648 (voice)

Room 237, NJM Library
Website

Acting Head, Health Sciences Division

Profile Photo
Mê-Linh Lê
she / her / hers
Contact:
Rm 223
NJM Health Sciences Library
Bannatyne Campus
204-228-6775
Website

Overview

Curious about the buzz around ChatGPT and AI's role in research? Eager to harness generative AI's power in a responsible manner that builds on your existing research skills? The resources and workshop on this page are intended to introduce you to the world of genAI, while equipping you with the vital know-how to seamlessly integrate generative AI into your current research toolkit.

 

Generative AI Tools

The following generative AI tools can be used to assist you in your research process. However, we urge you to use genAI tools thoughtfully and with caution, ensuring that you understand the ethical implications and potential consequences of the content they generate. Students should always consult with their instructors or advisors on the use of genAI in any assignments or course work. Refer to Artificial Intelligence at the University of Manitoba for more information and advice on use of genAI in academic settings.

Prompting*

Keys to building good prompts:

  1. Clarity: Be as clear and specific as possible. If you're too vague, the model may not understand exactly what you're looking for.
  2. Open-Ended vs. Direct: Depending on your needs, decide if you want an open-ended answer or a direct one. For instance, "Tell me about photosynthesis" is open-ended, while "What is photosynthesis?" is more direct.
  3. Specify the Format: If you have a specific format in mind (e.g., a list, a paragraph, or step-by-step instructions), mention it in your prompt.
  4. Anticipate Ambiguity: If there are multiple ways to interpret your question, try to anticipate and reduce ambiguity. For instance, instead of asking "How long is it?", ask "How long is the Mississippi River?"
  5. Provide Context: Sometimes, a bit of background or context helps the model generate better responses. For instance, instead of asking "What are its benefits?", you might say "What are the benefits of intermittent fasting?"
  6. Iterate and Refine: Don’t hesitate to adjust your prompt and ask again if the first response isn’t quite what you were looking for.
  7. Limit Bias: To get a more neutral and objective answer, avoid using leading or loaded questions.
  8. Engage in a Dialogue: Instead of relying on a single prompt, you can ask follow-up questions or ask for clarifications. This conversational approach can help in guiding the model to the desired answer.
  9. Check Facts and Cite Sources: If you're looking for factual information, you might include a request for sources or further reading recommendations in your prompt.
  10. Limit Length When Necessary: While it's sometimes useful to provide detailed context, overly long prompts can be counterproductive. Remember, there's a maximum token limit (the model’s internal unit of text, roughly equivalent to words), so very long prompts might truncate or limit the length of the response.
  11. Experiment: ChatGPT is flexible, and sometimes even slight changes in phrasing or question structure can produce different outcomes. Don't hesitate to try different prompt formulations to see what works best.

No model is perfect, and there will always be times when the response might not meet your expectations. Using the above strategies can, however, greatly improve the chances of getting the desired outcome.

*OpenAI. (2023). ChatGPT4 (Oct 17 version) [Large language model]. https://chat.openai.com/chat