Statement permitting AI use

Michael “Jack” Davis, Statistics and Actuarial Science

Sample wording 

Generative Artificial Intelligence (GenAI) trained using large language models (LLM) or other methods to produce text, images, music, or code, like Chat GPT, DALL-E, or GitHub Copilot, may be used for assignments in this class with proper documentation, citation, and acknowledgement. Recommendations for how to cite GenAI in student work at the University of Waterloo may be found through the Library’s Citation and Style Guides. 

Please be aware that generative AI is known to falsify references to other work and may fabricate facts and inaccurately express ideas. GenAI generates content based on the input of other human authors and may therefore contain inaccuracies or reflect biases.  

To protect the privacy and security of any data entered, students should use the University’s version of Copilot and login with their UW ID. Data entered into other systems can be added to training sets, monitored, geolocated and even reproduced as output which may share private personal information or result in intellectual property breaches.   

In addition, you should be aware that the legal/copyright status of generative AI inputs and outputs is unclear. Exercise caution when using large portions of content from AI sources, especially images. More information is available from the Copyright Advisory Committee.

You are accountable for the content and accuracy of all work you submit in this class, including any supported by generative AI. You should be able to readily demonstrate your knowledge of your submissions. To demonstrate your learning, you should keep your rough notes, including sources, research notes, brainstorming, drafting notes and prompts. You may be asked to submit these notes along with earlier drafts of your work, either through saved drafts or saved versions of a document. 

Additional clarification: 

The good news is that I’m allowing the use of GenAI for assignments this term. The bad news is that you have to listen to my spiel about GenAI first. 

This is especially pertinent in this class because Stat 341 has a large machine learning component to it, albeit as a classical, continuous space style of machine learning and not the large, modularized, reinforced neural networks of your favourite LLMs. 

80% of this course’s grade comes from exams, these grades come from memorization, summary, and on-the-fly analyses. These tasks are made a lot easier if you have spent at least a reasonable time wading through the assignments. 

Most of the learning in this class is expected to happen by struggling through the assignments, which are relatively low-stakes, at 5% of the grade each. In the past, these assignments have had grade averages of more than 90% despite a no-AI policy, which I expect will happen in this term too. So, if you’re going to use GenAI on your assignments, please do so in a way that doesn’t rob you of that learning, because it probably won’t affect your grade much one way or the other. 

Here's the one rule for using GenAI in this class: Include your prompts in your assignments, as if you were citing your sources in an essay. The prompts will not affect your grade. However, I might ask the permission of a few of you to share some very good prompt examples. 

One good argument I read in teacher training about using paper mills, a cheating service like today’s LLMs but with a living human doing the work, was that the assignments given were not worthy of the curiosity and creativity of you, the student. That’s not an entirely empty argument; assignments can feel like pointless busy work, just a set of paperwork to check off a list to get a degree. I’ve since tried to address that argument and put creativity into these assignments as much as the scope and requirements of the course allow. In past semesters, we’ve looked at NBA player statistics, live building inspection data from the GTA, and real brainwave data. This term, we’re going to look at the Mandelbrot set, Conway’s game of life, and the valuation of sports cards. We’ll also read and summarize a paper on XGBoost.  

A lot of this can effectively be skipped by feeding the assignments directly into GenAI. 

Examples 

Here are six good prompts. 

1) “Can you explain how the optim function in base R works?” 

2) “What does R mean when it says x is an atomic object? Why am I getting an error related to this?” 

Prompts 1 and 2 are essentially enhanced searches. You’ll find similar things on Google/Bing/Ecosia but maybe without the added context an LLM can give you. 

3) “Does a slope of 3 dollars per additional WAR seem reasonable for the value of common 1991 baseball cards?” 

4) “Can you give me a few talking points about gradient descent, and the challenges it faces with the Styblinski-Tang Function?” 

Prompt 3 is asking for subject context, which may be valuable if you’re not familiar with the subject of a dataset and can’t tell if your answer is close just from the number. Prompt 4 is asking for an alternative explanation for something we would have covered in class. 

5) “What are some other things I should know about the Newton-Raphson method?” 

6) “What prompts should I ask to better understand Bootstrap sampling?” 

Prompts 5 and 6 will actually help you learn in the long run, although the answers may go beyond the scope of the course. 

Here are two bad prompts. 

7) “Given this template [code template included], can you program a Conway’s game of life?” 

8) “Can you summarize the article on XGBoost titled xxxx, and answer the following questions?” 

Prompts 7 and 8 will help you finish the assignments, but they will also skip you over the learning that was supposed to happen. If you’re using prompts like these to check your work, I would recommend that you instead find a study buddy in the class and check with them. 

A few other comments about responsible AI use: 

1. There was a tentative study at the Waterloo of the south (called MIT by the locals), https://arxiv.org/abs/2506.08872, in which different groups were assigned to write essays using either LLMs, classic search, or unassisted. To quote the abstract, “Self-reported ownership of essays was the lowest in the LLM group and the highest in the Brain-only group. LLM users also struggled to accurately quote their own work.”
2. GenAI currently has substantial energy costs per use. This means using it a lot is going to have a real impact, but it also means there’s a lot of opportunity for clever people like you to make meaningful improvements. 
3. See any video by Rational Animations or by Robert Miles on AI Safety. https://www.aisafety.com/
4. It’s currently legally unclear the level of privilege applied to information you feed into an LLM. It’s certainly being used for training data, which can be retrieved almost directly using clever prompt engineering.