There are many generative AI tools available and a lot to consider if you are going to use one in your teaching, research or work. Before using a generative AI tool be sure your instructor, colleagues or team are comfortable with the intended use. While McMaster does not endorse any one generative AI tool, this page gives an overview of some common generative AI tools, as well as considerations around privacy when using generative AI tools.
In addition to privacy, with any generative AI tool you use you will want to consider common limitations and risks with these tools, including
- “Hallucinations” or confabulations where a generative AI tool creates false information
- Biases inherent in the training data that may be replicated in the generated outputs
- Copyright questions about the use of data in training AI models and
- Environmental impacts of generative AI tools
You can learn more about these limitations and risks in the McMaster Guidebook on Generative AI in Teaching and Learning or through this asynchronous module.
Though these risks are present, generative AI tools have astonishing capabilities in creating text, images, audio, video and code. Tools analyze data, interact with the internet and images, respond to voice prompts in text or audio and much more.
Some ways you might consider using a generative AI tool include: explaining a concept, creating examples, analyzing a data set, summarizing, providing feedback, generating an outline.
Common Generative AI Tools
With thousands of AI tools available it can be hard to know what tool to use and how to get started. Here we offer a general overview of two popular multi-modal tools – tools that can produce images and text. Check back as we continue to update this page.
OpenAI’s ChatGPT has two versions: a free version using the foundational model GPT 3.5 or a paid version using foundational model GPT 4. Both ChatGPT 3.5 and ChatGPT 4 are powerful tools that can produce text responses.
ChatGPT is a conversational chatbot that responds to a user’s prompt (or request) and can engage in continued dialogue to refine and improve responses.
The paid version, ChatGPT 4, also allows users to access plug-ins. Plug-ins add specific capabilities – like computation, news retrieval, trip planning and more. There are more than 160 available plugins for ChatGPT4.
Beyond the plugins, ChatGPT 4 can also interact with the internet through Microsoft Bing search, create images using Dalle3, interact with images, complete data analysis and work in voice-text, and text-voice.
When using ChatGPT you can use “data controls” in your account settings to turn off the collection of your chat history.
You can see an overview of how to use ChatGPT in this video.
Microsoft’s Copilot uses both the foundational model GPT 3.5 and GPT 4 and is free to access.
Copilot interacts with the internet, can create images using Dalle3, completes data analysis and can interact with an open webpage. Like ChatGPT, Copilot is a conversational chatbot that responds to prompts and can further refine responses based on feedback.
McMaster has an enterprise wide license for Copilot for faculty and students that means neither Microsoft or McMaster are collecting your prompt data.
More information on how to use Microsoft Copilot can be found at this link.
Scite.ai is designed for citation analysis. It integrates machine learning and human input to provide a broader perspective on how publications are received and discussed within scholarly literature. It also includes a new feature called scite Assistant (in beta). Scite Assistant helps to answer questions with research backed information by using generative AI (Artificial Intelligence) to query the scite.ai citation index.
McMaster Libraries provides access to scite for current McMaster University students, faculty, and staff.
- Scite.ai allows users to upload a manuscript and see if references have been retracted or heavily contrasted, find missing citations, or explore how others reference the same studies.
- Scite.ai facilitates more granular tracking of citation information by categorizing citations based on their contextual relevance to a cited work, classifying them as “supporting”, “mentioning”, or “contrasting”.
- Scite.ai facilitates more granular tracking of citation information by categorizing citations based on their contextual relevance to a cited work, classifying them as “supporting”, “mentioning”, or “contrasting” (known as Smart Citations). The Smart Citations API is integrated into the Lean Library browser extension and provides retraction information.
- Scite Assistant helps to combat the “hallucination” phenomenon wherein some generative AI tools produce citations that are nonexistent. It searches against full-text research articles and over 1.2 billion citation statements in scite.ai.
- Scite Assistant allows users to specify a response length for their query: short (100 – 200 words), medium (200 – 500 words), or long (500+ words).
- Scite Assistant can answer questions about a specific paper when the DOI (digital object identifier) is included in the search query.
- Scite Assistant provides users with the ability to enter a quote or use natural language in a search query to identify a specific source. This feature can be used to help verify information or fact check claims made on social media and elsewhere.
- Scite.ai. includes limited grey literature, (e.g., no conference proceedings or dissertations), but does include books or book chapters when they have a DOI and preprints.
- Scite.ai’s reliance on DOIs means that articles without them, such as some older publications, might not be covered.
- Scite.ai’s accuracy in classifying citations can be low, particularly in distinguishing between the categories of “supporting” and “mentioning”.
- Scite Assistant may be limited in its ability to capture nuanced meanings, such as linguistic and cultural variations.
- Scite Assistant may cite a real reference in a way that does not reflect what the reference says.
- Scite Assistant tends to rely on a particular article in its response to a query, which could limit the scope of highlighted literature.
- Scite Assistant may provide inconsistent results to a query which raises concerns around reproducibility.
Poe is not one generative AI tool, but many. The site aggregates common generative AI tools – including ChatGPT, Anthropic’s Claude, Midjourney, Meta’s Llama and Google’s PaLM – and lets users select the tool you want to use and interact with it. Anthropic’s Claude can be useful as it works well with pdf files and can engage with very long texts.
Information Box Group
When considering using a generative AI tool you will need to think about the privacy of the data you are working with and the way the tool you are using interacts with personal information. Review this webpage to learn more about privacy considerations and generative AI tools at McMaster.
If after reviewing these considerations and available tools you have decided to use a generative tool, you will want to learn more about how to ‘prompt’ the generative AI tool effectively. This short article introduces ideas of prompting in the post-secondary context along with sharing some concrete examples.