Workshop 1: Foundations & Capabilities
Workshop 1: Foundations & Capabilities
What generative AI is, which Drexel tools to use, and how to prompt for useful results
Monday, June 29, 2026 | 1:00–2:00 PM | Bring a laptop
What You’ll Leave With
By the end of this hour, you will be able to:
- Explain in plain terms what a large language model does—and what it doesn’t
- Choose the right Drexel-approved tool for a task and know the data rules
- Sign in to ChatGPT Edu, select GPT-5.5 when available, and run your first useful prompts
- Use a simple framework (PTCF) to get noticeably better outputs
- Recognize how AI fails—confident errors and bias—and the basic accountability and governance rules at Drexel
Run of Show
| Time | Segment |
|---|---|
| 0:00–0:10 | Overview: what AI can and can’t do; the Drexel tools and data rules |
| 0:10–0:25 | Ethics & Governance: how AI fails, and who’s accountable |
| 0:25–0:42 | Activity 1: First prompts in ChatGPT Edu |
| 0:42–0:57 | Activity 2: Better prompts with PTCF |
| 0:57–1:00 | Wrap-up and what to try this week |
Overview (10 min)
What generative AI is. Tools like ChatGPT and Copilot are large language models—they predict likely text based on patterns learned from huge amounts of writing. They are very good at drafting, summarizing, rewriting, explaining, and brainstorming. They do not “know” facts the way a database does, and they can state wrong things confidently (a “hallucination”). You stay in the loop: AI drafts, you verify and decide.
Which tool, and the data rule. For Drexel work, use ChatGPT Edu, Copilot Chat with Enterprise Data Protection, Microsoft 365 Copilot, Zoom AI Companion, or Adobe tools according to Drexel’s data classifications. Claude, personal ChatGPT, Gemini, Perplexity, and other consumer tools are Low Risk Data only unless separately approved by Drexel. Never paste PHI, sensitive PII, protected student records, or identifiable research data into a tool unless that tool and use case are explicitly approved. Full details are on the Tools & Access page.
Good first uses at DSPH. Drafting and tightening emails, summarizing long documents and reports, turning notes into clean prose, explaining unfamiliar terms, brainstorming options, and reformatting content. We start there.
Treat AI like a fast, eager assistant who is widely read but sometimes wrong and never offended by edits. You give clear direction, it produces a draft, and you check the work before it counts.
Ethics & Governance: How AI Fails, and Who’s Accountable (15 min)
This is the first piece of an ethics and governance thread that runs through every workshop. Today: the two ways AI gets things wrong, and the rules that decide who is responsible.
Two failure modes to expect. First, confident errors (“hallucinations”)—an LLM predicts plausible text, so it can invent facts, statistics, and citations that look right and are wrong. Second, bias—models learn from internet-scale data that carries society’s inequities, so outputs can reflect and amplify stereotypes about who belongs in which role and what communities “look like.” Both matter more in public health than in most fields, because we work with diverse populations and our outputs can affect care and resources.
Where bias comes from. It enters at several stages: training data that reflects historical inequities, populations that are over- or under-represented, and proxies that don’t mean the same thing across groups. A landmark example: an algorithm used across US health systems used past healthcare cost as a proxy for health need; because less had historically been spent on Black patients, it rated them as healthier than equally sick White patients (Obermeyer et al., 2019). The lesson isn’t “don’t use AI”—it’s that AI trained on historical data can encode inequity, so you check outputs, especially across groups.
Who is accountable. The person who acts on AI output owns the result—not the tool, not the vendor. That is the governing principle for everything in this series: AI drafts; a human verifies and decides. Where AI informs a real decision or document, note that it was used and verify what matters.
The governance frame at Drexel. Drexel sets the rules through its AI policy and the data-classification system on the AI Tools page—which tool may touch which data. Nationally, the ASPPH AI for Public Health framework and the WHO’s Ethics and Governance of AI for Health lay out the principles (equity, transparency, accountability, privacy) we’ll apply throughout. See the Resources page.
If AI helped produce it and your name is on it, you are responsible for it—accuracy, fairness, and tone. “The AI wrote it” is never the answer.
Activity 1: First Prompts in ChatGPT Edu
Activity 1: First Prompts in ChatGPT Edu
Goal: Get signed in and feel what the tool does well by running real tasks.
Instructions
- Sign in to ChatGPT Edu with your Drexel account. Use GPT-5.5 if it appears in your model picker; otherwise use the current default/latest GPT model in the Drexel workspace.
- Run each of these prompts and look at how it responds. Use public or made-up details only—no sensitive data.
Prompt A — Summarize
Summarize the following text in 5 bullet points for a busy colleague, then give me a one-sentence “bottom line.”
Between 2018 and 2022, pedestrian fatalities in the United States increased by 18 percent, reaching the highest levels since 1981. Urban areas accounted for roughly 84 percent of these deaths, with arterial roads—wide, high-speed corridors often lacking sidewalks, crosswalks, or pedestrian signals—representing a disproportionate share. Black and Hispanic pedestrians were killed at roughly twice the rate of white pedestrians when adjusted for population. Older adults (65+) also faced elevated fatality rates, particularly at unsignalized intersections. Several cities have adopted Vision Zero plans, but implementation has been uneven: many plans focus on education campaigns rather than the infrastructure redesign (road diets, curb extensions, reduced speed limits) that evidence links to the largest reductions. Federal funding through the Bipartisan Infrastructure Law includes $5 billion for Safe Streets, though local match requirements can slow uptake in under-resourced communities.
Prompt B — Draft
Draft a friendly but professional email reminding committee members that their self-study section drafts for the CEPH reaccreditation report are due Friday at noon. Mention that late submissions delay the full committee review scheduled for the following Tuesday. Keep it under 120 words.
Prompt C — Explain
Explain what a “data use agreement” is to a new research coordinator who just started at a school of public health, in plain language, in about 100 words.
Prompt D — Reformat
Turn these rough notes into a clean numbered agenda with time estimates:
curriculum cmt mtg weds — need to talk about new DrPH concentration proposal, carla is presenting enrollment projections, also CEPH competency mapping still not done, someone said we need to vote on the revised practicum hours (120→150), also brief update from assessment subcommittee maybe 5 min?, oh and the AI policy draft — alex will share it, probably 15 min discussion, don’t forget to approve last meeting minutes at the start
- Pick the output you found most useful and ask one follow-up: “Make it shorter,” “More formal,” or “Add a sentence about X.” Notice that it remembers the conversation.
Quick Reflection
- Which task did it handle best? Which needed the most fixing?
- Where did you have to add information it couldn’t have known?
Activity 2: Better Prompts with PTCF
Activity 2: Better Prompts with PTCF
Goal: See how a little structure turns a generic answer into a usable one.
The PTCF Framework
| Part | Ask yourself | Example |
|---|---|---|
| Persona | Who should it act as? | “You are an experienced grants administrator…” |
| Task | What exactly should it do? | “Draft a 150-word reminder…” |
| Context | What background does it need? | “…for faculty submitting NIH progress reports, due in two weeks.” |
| Format | How should the output look? | “Bulleted, warm tone, with a clear deadline line at the top.” |
Try It
Start with a bare prompt:
Write a project update.
Now rewrite it using all four PTCF parts:
Persona: You are a research coordinator at a school of public health. Task: Write a project status update for our NIH-funded pedestrian safety study. Context: For the PI and co-investigators’ monthly check-in. We finished geocoding 14,000 crash records and linking them to road-segment characteristics. Data cleaning is behind schedule—about 30% of records have missing speed-limit data that needs to be imputed or verified against PennDOT records. The IRB continuing review is due August 15 and the progress report to NICHD is due September 1. Format: Three short sections—Completed, In Progress, Needs Attention—under 150 words total, with specific dates for upcoming deadlines.
Compare the two outputs. Then refine once more: change the tone, the length, or the audience and rerun.
The same prompt can give slightly different answers each time. For anything you’ll reuse, save the prompt that worked. We’ll build reusable prompts and custom GPTs in Workshop 3.
Wrap-Up (5 min)
- AI is a drafting and thinking assistant; you verify and decide.
- For Drexel work: ChatGPT Edu or Copilot, and mind the data rules.
- PTCF (Persona, Task, Context, Format) is your default recipe for a good prompt.
Try this before Workshop 2: pick one task you do every week—an email, a summary, a set of notes—and do it once with ChatGPT Edu using PTCF. Bring it to the next session.
Looking Ahead
Workshop 2: Workflow Enhancement — we move from single prompts to real workflows: drafting and summarizing in Word and Outlook with Copilot, and using ChatGPT Edu to speed up the writing and analysis tasks you do most.