Generative AI-Centered Assignments for UIC Faculty

Course Context

This is a 400 level combined grad/undergrad course on secure web application development. As a bit of additional context – “Copilot” is a LLM-based tool which provides “autocomplete” suggestions as grayed out text while writing code, and the user can hit tab to accept the suggestion. While this is similar to the autocomplete that exists in various other writing tools, an important difference is that in many cases, if Copilot believes it can “guess” what the author is trying to do, it can provide over a dozen lines of code as a single “suggestion” to be accepted. This makes it a very powerful tool, but also a very dangerous one that causes beginners to end up over their head without realizing it.

Learning Objectives

• Assess the reliability of generative AI tools
• Determine how and when to consult reliable sources to verify AI-generated code

Rationale for Incorporating Generative AI

Since generative AI has become ubiquitous in the computer programming industry, understanding how to use it and troubleshoot it responsibly is vital to preparing students for a future in computer science.

While this is not exactly a particularly deep or challenging question, my intention here is to simultaneously highlight (a) the reliability of these tools and (b) the importance of knowing how and when to consult with reliable sources, which is a key component of successful use of generative AI.

Assignment

[See image above].

This is a moment from the livecode of Homework 1. Copilot was suggesting the text in bold as a way to use Drizzle to accomplish the goal mentioned in the comment above that line. Suppose you took that suggestion from your friendly LLM assistant, ran it, and it immediately crashed. What resources should you consult to make progress on this problem? Be specific – answers like “Google” or “the stack trace” or “tell ChatGPT/Copilot that it crashed and ask it to fix the problem” are too vague. If you watched the video and remembered what I actually did in this situation, that’s an appropriate and correct answer.

For those interested – the intended correct answer (which most students got – it was the highest scored question on this exam), was to consult the documentation for the library (software tool) I was using, as it contains “ground truth” about how to actually use a given library.

The example assignment above asks students to demonstrate understanding of how to troubleshoot LLM errors in the context of AI-assisted coding. To successfully respond to this question, students need to show that they understand the notion of “ground truth” and know where and how to ascertain it in this particular instance.

This represents a simple but effective question-type adaptable to almost any disciplinary context, and easily attachable to quizzes, tests, or post-project reflections.

Course Context

Medical students in the preclinical years require training and guidance from teachers to develop the skills necessary for mastering Self-directed learning (SDL).

SDL empowers learners to diagnose their learning needs, formulate goals, identify resources, apply learning strategies, and evaluate their learning outcomes. SDL is a crucial component of the medical student curriculum. Research by Ambrose Bridges et al. (2010) highlights several important strategies to promote SDL, including providing multiple opportunities for assessment, monitoring progress, and encouraging students to analyze the effectiveness of their efforts.

To achieve these goals, we have developed assignments and assessment tools that foster learning, monitor progress longitudinally, and provide feedback on SDL skills. These assignments have been utilized for several years.

Learning Objectives

• Evaluate the reliability and limitations of AI-generated outputs
• Reflect on the differences between personal and AI-generated evaluations

Rationale for Incorporating Generative AI

This year, we modified the assignment by adding questions that encourage students to reflect on the strengths and pitfalls of using Large Language Models (LLMs) in critical appraisal of medical literature. The aim was to help students learn how to effectively utilize LLMs. Students were asked to compare their appraisal of the strengths and weaknesses of an article with that of the LLM, identify any discrepancies or incorrect responses in the LLM’s output, and reflect on how the information from the article and the LLM’s response applied to their patient’s problem.

Assignment

We aim for you to be able to address patient questions by integrating their concerns with your clinical expertise and best available evidence.  Large Language Modules (Co-Pilot/ChatGPT) will be used by both patients and clinicians, and we want you to understand its capabilities and limitations.  We would like you to use the free version of a LLM for the following question.

Run the following prompt in the LLM along with a copy of your article: 

“As a physician, critically appraise the article below, identify two strengths and weaknesses not discussed by the author and comment on the internal validity of the study”

Copy and paste the LLM Response below:

Share how your appraisal of the strengths and weaknesses of the article was superior to that of the LLM.  Comment on any discrepancies/incorrect responses you noticed in LLM’s response:

Present your answer to the PICO question as you would present to your preceptor.  Explain how the information in the article and LLM response applies to your patient’s problem:

Course Context

4th year medical students will critically evaluate pharmacogenomic recommendations provided by a large language model (e.g., ChatGPT) and compare them with evidence-based guidelines from the CPIC (Clinical Pharmacogenetics Implementation Consortium) database or equivalent for a given patient case.

This activity aims to develop students’ skills in utilizing digital tools, interpreting evidence-based resources, and identifying potential benefits and limitations in the AI and databases.

Learning Objectives

• Compare AI-generated recommendations with traditional resources
• Evaluate the accuracy and relevance of AI-generated recommendations

Rationale for Incorporating Generative AI

As generative AI is increasingly adopted across the healthcare system, it’s important to foster future medical professionals’ development of clinical judgment in the context of AI-assisted medicine.

Assignment

Background:

A 45-year-old patient is interested in exploring genetic testing options to determine if it could help improve his mental health care. Patient was diagnosed with major depressive disorder two years ago and has been on citalopram 20 mg daily for the past year. He has tried sertraline and mirtazapine in the past without much success.  While the medication citalopram has provided some relief, he reports persistent mild depressive symptoms, including low energy, difficulty concentrating, and occasional feelings of sadness. His health care team suggested nortriptyline as another option to optimize his antidepressant regimen.  Patient recently learned about direct-to-consumer (DTC) genetic testing through a friend who used 23andMe to identify potential health-related genetic traits. The patient is curious if similar testing could provide insights into his mental health care, especially in selecting “better” antidepressant medications and dosages for him.  He has no other significant medical history.  Family history includes a father with a history of depression and generalized anxiety, who is treated with mirtazapine.

Instructions:

It’s time to demonstrate your mastery of pharmacogenomics and evidence-based knowledge. For this assignment, you will use a large language model (LLM), such as the free version of ChatGPT, along with evidence-based resources like CPIC, PharmGKB, or FDA databases to gather the information needed to address this patient’s question.

In 500 words or less (you are welcome to use table to summarized contents), please include the followings:

• The exact search query words/phases used in your search strategy.
• The information provided by each resource used.
• Compare and contrast the results in the recommendation.
  • Discuss the details and the explanations/recommendation provided by the LLM versus CPIC.
  • Reflect on whether the LLM identified contents similar to CPIC or other pharmacogenomics resources in its response.
  • Most importantly, based on the recommendations from these resources, which one provided you the most efficient information to address this patient’s question.  Or do you’ve enough information to write a prescription or lab test order

Your submission will be evaluated according to the following rubric: Clarity and Conciseness, Accuracy, Relevance

Course Context

First-Year Writing Program courses guide students into academic inquiry and direct them in completing several writing projects across public and academic genres, including a documented research paper.

Learning Objectives

• Analyze how AI-generated writing suggestions impact the clarity, tone, and meaning of one’s own writing
• Make informed revision choices to maintain personal intent in generative AI-assisted writing

Rationale for Incorporating Generative AI

Automatic writing feedback programs like Grammarly have become widespread with campus and corporate adoption and individual use. The automated suggestions to correct spelling and punctuation may help the user clean up their lower-order writing concerns. The suggestions to grammar correction and word choice that impact clarity, concision, tone, and rhetorical effect are higher-order writing concerns that require more deliberation from the user. In many cases, the user may automatically accept suggested changes to grammar and wording which in effect alter the intended meaning and rhetorical impact of their writing to varying degrees.

Assignment

We aim for you to gain firsthand insight into the specific corrections that Grammarly suggests, to evaluate the extent to which the suggestions are helpful, and which types of suggestions tend to be more accurate and appropriate for which writing situations. This activity should give you critical insight into the usefulness of Grammarly as a tool for revising writing, without entrusting Grammarly to edit the entirety of our writing and retaining our writerly autonomy.

• With Grammarly enabled in your PC or Macbook browser, create a new Word Doc and copy and paste a single paragraph of an academic essay you wrote for one of your UIC courses.

• Check the Grammarly Alerts menu on the righthand side of the screen. “Accept” every suggestion that Grammarly makes in the passage.

• Compare your original paragraph with the corrected paragraph and write down any differences you notice to Clarity, Engagement, and Delivery between the two.

• Click the “Set Goals” option in the Grammarly menu and adjust the Audience, Tone, Formality, or Intent. Run the Grammarly diagnostic again on the original passage and accept each of the Goal suggestions.

• Compare the original document with the Goal suggestions and write down any differences you notice to Clarity, Engagement, and Delivery between the two.

• If you’re not satisfied with the newly corrected Grammarly versions, run through the original paragraph one more time and accept only the Grammarly changes that you deem necessary, and reject the suggestions that would otherwise change your original writing.
• Finally, compare this final draft with the other three drafts and discuss which draft is most appropriate and effective for your target audience and purpose.

These three examples ask students to complete a task and compare their results with an LLM’s in an effort to thicken their working familiarity with LLMs’ affordances and limitations in specific practical contexts.

This is an enormously adaptable framework for assignments aimed at tuning students to the particulars of almost any kind of critical, technical, and intellectual task in conjunction with LLMs. This assignment type’s emphasis on comparison and critical reflection foregrounds the task itself (its demands, criteria, and challenges), placing the LLM in a flexible support role.

Course Context

This assignment is part of the final project for a general education course on folk and fairy tales. The course emphasizes the difference between oral storytelling and modern literature captured, in a nutshell, in the following:

• Storyteller vs. Author
• Improvisation vs. Writing (Performance to Text)
• Action vs. Character / mind-Reading
• Thematic Shift: wealth/romance/power to existential crisis and social mobility/values
• Existential Shift: fate and destiny vs. psychological analysis
• Audience: Multigenerational

Learning Objectives

• Reflect critically on the difference between oral tales and modern literature from the perspective of adaptations by guiding them through a series of prompts to give an LLM.

Rationale for Incorporation Generative AI

Generative AI is utilized in this assignment to provide students an accelerated glimpse of the complex process of adaptation. Without generative AI, the same assignment would require students to perform the work of adaptation themselves, a cognitive load perhaps too large to place on students individually. Generative AI in this case allows each student to have a unique instance of adaptation to reflect on.

Assignment

(documents are supplied by the student: one “fairy_tale_summary,” one “sample01_modernist_literature,” one “sample02_modernist_literature”)

Phase 1: students work without generative AI

(1) Analyze the fairy_tale_summary and break it down into plot events. Place these in tags. (e.g. “”)

(2) Then, analyze the samples of modernist literature for style, literary features, and narrative devices. Place these in tags. (e.g. “, ,” etc.)

(3) Next, review the suggestions for adaptation supplied by the instructor.

Phase 2: students work with generative AI

Prompt 1:

You are an AI assistant with superb skills in creative writing. You also have expert knowledge in folk and fairy tales as well as in the devices and the techniques of modernist literature. Your task is to help college students to create a modern adaptation of a classic fairy tale.

>>The LLM responds, acknowledging their role.

Prompt 2:

Outline a modern adaptation of a classic fairy tale, using the following draft and employing the following style. Make sure that your story preserves the plot structure of the classic fairy tale. Include the protagonist’s thoughts and feelings in each stage of the story. Here are the draft and the style guide:
(Student supplies the LLM with the summary of their fairy tale and the style guide they’ve constructed with tags.)

>>The LLM produces an outline in bullet points

(The student now places those bullet points in tags, resulting in an outline that represents the careful step-by-step construction of the preceding steps)

Prompt 3:

Your task is to write a creative modern adaptation of a classic fairy tale. Please review the outline and make sure you maintain narrative coherence and stylistic consistency between the parts. Here is the outline:

(Student provides or refers to the outline previously generated by the LLM)

>>The LLM composes an adaptation as requested

In this example, instead of having a student complete a task and compare their results to an LLM’s, the student is guided through a detailed process showing how to get an LLM to complete a complex task. In the course of following the process, the student is simultaneously shown the inner workings of the task.

In this kind of assignment, the LLM is heavily “supervised” by the student, and acts primarily as a tool confirming the conceptual insights and framework being articulated to the student.

The use of AI writing tools (including, but not limited to, ChatGPT, Bard, or Sudowrite) is NOT permitted in this course. Students who use these tools for class assignments undermine the goals and learning objectives for this course, reducing the effectiveness of instruction. The instructor may submit student writing to an AI writing detector (e.g., GPTZero) at any point throughout the term. Any confirmed use of AI writing tools will be treated as cheating. Students should reference UIC’s Student Disciplinary Policy for more information.

The recent advances in AI technology are already transforming the ways humans communicate. In order to prepare students for an AI-infused world, the use of AI writing tools in this class is permitted in some ways. Students are encouraged to use AI writing tools (such as ChatGPT, Bard, or Sudowrite) to generate ideas for their writing and course work in this class; however, it is expected that all AI-generated content be reviewed, edited, and verified for accuracy before submission. Please note that you need to cite the specific AI writing tool as a source if you present any significant amount (i.e., more than one sentence) of minimally edited AI-generated text as your own. Please review the APA or MLA guidelines for citing generative AI writing tools.

The recent advances in AI technology are already transforming the ways humans communicate. In order to prepare students for AI-assisted work, the use of AI writing tools is permitted in this course with no restrictions.

In this course, we will be using technology like graphing calculators, Microsoft Excel, and other statistical software in this course to help you understand and apply statistical concepts. However, the most important goal here is for you to develop your own analytical thinking, critical reasoning, and statistical skills. That is why generative AI tools are not permitted for any part of this course.

Generative AI tools, such as ChatGPT, are tools that use machine learning to create entirely new content, like text, images, charts, or computer code. Using them in this course can make it harder for you to learn effectively and for your instructor to guide your learning.

All your work for this course, including assignments and assessments, must be completed without the use of generative AI. If you are unsure about using a particular technology tool or platform, please ask your instructor before you use it for an assignment, quiz, or test.

Following this policy will help you learn the most in this course and build skills that will benefit you down the road.

Knowing how to use generative AI is essential for today’s marketers. It lets you quickly create personalized content, come up with fresh ideas, and explore unique customer insights. In this course, you will use generative AI tools to boost your marketing and consumer analysis skills.

You will be using generative AI platforms for different assignments and tasks throughout the course. This will give you hands-on experience with how these tools can help you in your marketing work. Assignments have been designed to be completed with the free version of any of the most popular generative AI tools, including OpenAI’s ChatGPT, Antropic’s Claude, Google Gemini, or Microsoft’s Copilot. For additional guidance on getting started with generative AI, please review the University of Illinois System’s Generative AI Guidance for Students. You can also review these prompting guidelines from MIT’s Sloan Technology Services Teaching and Learning Technologies.

You are encouraged to use generative AI to its fullest potential in this course. You can use it however you like or as we direct in specific assignments. The key is to guide the AI to get what you need while demonstrating your own critical thinking. Make sure to review and tweak anything the AI gives you before you turn it in.

Using generative AI responsibly and ethically is important in this course. By the end, you will know how to effectively and thoughtfully use these powerful tools in your future marketing jobs.

Please follow our recommendations but know there is flexibility as we learn how best to use this new tool together. Transparency in your process and honest use are valued above all else. AI is a step in the process, but not an end. AI is an extension of thinking, not a replacement for thought. Completed work must be your work, meaning your idea and final output. Not only should this be a repeatable process, but it should make you a more valuable designer. The risk of submitting AI work as your own is the tendency for AI to rely on existing work from the web, thus introducing potential issues of plagiarism.

You are permitted to use generative AI in the best way you see fit to accomplish your work in this course. You will be responsible for all work submitted.