The University of Vermont supports responsible use of Generative AI (GAI) tools.   

Integrating GAI technology into our work offers numerous benefits that enhance productivity, efficiency, and innovation. AI tools can automate repetitive tasks, generate ideas, summarize information, compose information for different audiences, and build understanding.  GAI can also support data analysis, providing insights that improve decision-making, student experienceand program development.  All of these allowstaff to focus on strategic and creative responsibilities that require human judgment and expertise. Staying current on the ways that GAI can leverage our work is our responsibility.  Doing sofosters a culture of innovation, positioning UVM  to remain competitive and responsive to evolving industry and educational trends.    

As of February 2024, UVM is pursuing its own closed GAI on the Lama GAI platform which will become the standard tool once adopted.  This will provide the greatest protection for UVM’s intellectual property. 

There is a wide variety of tools available for specific purposes. Selecting a tool should be guided by the output desired.  The blog One Useful Thing by professor Ethan Mollick is just one of many sources of guidance on tool selection. 

Choose a tool with the ability to disable data sharing for training purposes.   While this setting is helpful for protecting data, this does not guarantee privacy for the following reasons. 

• Your prompts are often stored temporarily
• ‘data sharing’ definitions differ among companies.
• Legally companies need to honor requests only for residence of certain countries and certain states that have passed privacy laws.  Vermont is not particularly strong in this area as of February 2025.  

Avoid entering NPPD (non-protected public data) about an individual. Instead, generalize any queries. The following examples illustrate safe and unsafe prompts.

Examples of Safe and Unsafe Prompting 

Example 1: Generating Quiz Questions 

• Unsafe Prompt: “Create 10 multiple-choice questions based on this passage [paste passage here] from Chapter 3 of Professor Smith’s unpublished textbook for UVM’s Environmental Science program.”
• Safe Prompt: “Generate 10 multiple-choice questions on this passage [paste passage here] on introductory environmental science topics, focusing on ecosystem dynamics.”  

Example 2: Drafting Email Templates 

• Unsafe Prompt: “Write an email to student Alex Johnson about their grade in UVM’s PSYC 101 course, referencing their essay on Freud.”
• Safe Prompt: “Draft a general email to a student about scheduling a meeting to discuss their grades and assignments in an introductory psychology course at UVM.”  

Example 3: Creating Educational Materials 

• Unsafe Prompt: “Generate slides summarizing research conducted by Professor Jones in UVM’s Department of Biology on CRISPR gene editing.”
• Safe Prompt: “Create a presentation outline explaining the basic principles of CRISPR gene editing and its potential applications in biology.” 

Avoid using personal accounts for work-related tasks to prevent data breaches. If you sign up for an GAI tool, do not use your NetID password. 

Clearly disclose the use of GAI in creating documents, reports, or any outputs. Depending on the context, this may involve attribution or a statement of collaboration with GAI tools. 

See this comprehensive guide to citations in MLA, APA, and Chicago style from University of Maryland.

As many tools acknowledge, GAI results may contain hallucinations: inaccuracies, misrepresentations, or outright falsehoods  Read and edit all output. Treat GAI outputs as drafts or suggestions, not final solutions.. Take steps to verify the accuracy of output by looking at the source material—review the references given by the GAI.  Ask your GAI for references as needed—and verify that those references exist.  Does the source material support the output? Fact-check against established sources and published material.   

Be mindful of biases in GAI tools that may affect the outputs. Avoid using GAI tools for decisions or developments that may have significant ethical implications without oversight. 

Data centers are currently 3% of global energy demand (EPRI 2024, Goldman Sachs 2024). UVM’s first principle of Planetary Health references stewardship of the earth.  According to the Electric Power Research Institute, and GAI query consumes 10 times the electricity of a google search.   For this reason, use a search engine when you are retrieving information.  Use GAI when its helpful (such as for generative work) rather than as the default. 

Stay informed about advancements in GAI technologies and their potential risks and benefits. Participate in university-led training programs on GAI ethics, privacy, and secure software development practices.

There are new tools and applications of GAI every day, if not every hour.  The list below provides a few ideas as a starting point and is not intended to be comprehensive.  For current discussion on GAI tools at UVM, join the [teams channel devoted to this topic]

1. Marketing (See Bettyjo Bouchey’s presentation for more detail). AI can be looked at as an “agency partner” to support project needs outside of the current in-house staff bandwidth or expertise
2. Content development
   1. Written Content, including differentiating content by audience.
   2. Image generation  
   3. Video generation
3. Website UI
   1. Chatbots and GAI assistants
   2. AI powered site Search
4. Coding and scripting
   1. Developing scripts.  Some tools are also able to run those scripts within the tool
5. Instructional content
   1. Generating quiz questions—although consider that students can also use GAI to answer these same questions.  Authentic assessment can mitigate these limitations.
   2. Adjusting reading level of content
   3. Virtual tutors and chatbots
   4. Summarizing online discussion
   5. Assess possible bias in instructional materials
6. Correspondence
   1. Re-phrasing email content
   2. clarifying ideas
   3. changing the tone of a message, etc. 

Do not use GAI tools for any prompts that include FERPA or HIPAA-protected data or PII. Staff are responsible for verifying the accuracy, reliability, and appropriateness of GAI-generated content before using or publishing it. AI outputs must not include unverified claims, misinformation, or copyrighted material.