Overview
U-M Maizey is a tool that enables U-M faculty, staff, and students to use custom datasets to enhance their GenAI experience, helping them extract insights, discover patterns, and gain deeper knowledge from the data. As a new technology, we look forward to working with you to improve it.
Understanding how to use U-M Maizey proficiently will provide you with the most insight into your data. Refer to the tips below for guidance on how to use U-M Maizey.
U-M Maizey
- Getting Started with U-M Maizey (for Dropbox, Google Drive, public website)
- How to Use U-M Maizey (video - 1 min)
U-M Canvas Maizey Integration
See the U-M Canvas Maizey Integration page for more information about setting up a Maizey project in your Canvas course, including upcoming in-person workshops.
- Getting Started with Maizey Canvas Connector (step-by-step instructions)
- How to Use U-M Maizey Canvas Conector (video - 2 min)
Understand Prompts
Understanding how to write prompts and queries effectively will help you get the best from U-M Maizey. Refer to the following to learn more about prompts:
Source Data
Learn more about supported file types in U-M Maizey.
You can add a maximum of 50 data sources to a single Maizey project. When adding a source using the Data from a public website option, you can list multiple URLs and it still only counts as one data source.
U-M Maizey works best with unstructured, natural-language text (words, phrases, sentences). The more documents indexed, the better the U-M Maizey data set becomes.
Best Practices with Source Data in U-M Maizey:
- Use several small documents in your datasets rather than one large document.
- Structured content such as spreadsheets may produce inconsistent results.
- Consider splitting your data into a collection of contextual text files and indexing that way; again, natural language sentence structures work best.
- Content comprised of non-sentence text, such as abbreviations and condensed phrases (as you may find in resumes and similar documents), may not index well.
- Remove special formatting, graphics, etc., to improve results.
- Documents with embedded structured data such as JSON should not be used.
Creating a Project
Note the following when creating a project in U-M Maizey:
- You can edit, delete, or add data new sources after creating a project.
- After creating a project, you can update the MCommunity Groups, Shortcode, Project Name, or Project Path.
- Tips for creating a project in U-M Maizey:
- Test the URL before sharing it with others in the MCommunity group.
- MCommunity groups can have multiple owners; therefore, if you are an owner in the MCommunity group, you are also the owner of any project created with that MCommunity group.
- Projects that you own will be listed under "Shared Projects."
- You have edit rights for all projects that you "own."
- Your project will not be accessible to the viewer group until it is published.
- Indexing Data for your project:
- There is a delay as data is being indexed.
- To determine if indexing is complete, check the Task Activity list.
- If your dataset is updated, click Reindex on the data source to refresh the data in the project
User Interface
What to know about the user interface:
- The name of your project is not related to the "Title" of your project's AI chat tool. The title of the project cannot be changed. The title of the AI chat tool can be changed by editing the project.
System Prompt Augmentation
System Prompt Augmentation allows you to define and fine-tune how the system responds to user prompts. View System Prompt guidance and examples.
Temperature
Temperature allows you to adjust the sensitivity and output-randomness of the model. A medium temperature is recommended for most use cases.
The higher the temperature is set, the more "creative" the response to a query (i.e., you're giving it permission to "think outside the box"). However, a high temperature might result in answers that are too unusual. Higher temperatures are generally better suited for use cases where creativity is valued (e.g., generating song lyrics, poetry, etc.)
The lower the temperature is set, the more "conservative" of a response the AI model might provide to a query (i.e., you want it to stick to safe answers with a high probability of being accurate). However, a low temperature might result in the AI model missing some correct answers because it is not specific enough.
Data Chunks
Think of chunks as individual pieces of a puzzle. Each chunk contains a small, specific piece of information from a larger database. Just as each puzzle piece contains part of the image, each chunk contains a part of the response to your prompt. Maizey combines these chunks, like assembling a puzzle, to give you a complete response.
Maizey's default setting is four chunks, which works well for most use cases. Increasing the number of chunks allows Maizey to access more indexed data, which may enhance the quality and variety of query responses.
Review
U-M Maizey responds to queries based on your custom data set alone. As with any AI technology, it is important to fact-check to ensure the information is accurately presented.