The Workshop
How the 90 minutes works.
The run of show, the shape of each challenge, and the four prompting moves that do most of the work.
Why we’re here The premise of the next 90 minutes is simple: answers are cheap, judgment is expensive. Any chatbot will hand you a paragraph, a figure, or five hypotheses on demand. What it cannot hand you is the taste to know which answer is good, the skepticism to catch the one that is quietly wrong, and the framing that turns a vague wish into a useful result. That judgment is what matters most — and it is exactly the thing a working pain scientist already has. You will leave with a small, reusable toolkit of prompting moves, hands-on practice applying them to real academic tasks, and a clearer sense of where these tools earn their keep and where they fall short. Newcomers are in the right room; experienced users will still find an edge.
The 90-minute arc
| Time | Block | What happens |
|---|---|---|
| 0:00–0:10 | Warm-up discussion (10 min) | “What do you actually use AI for at work?” Round-robin at tables (not whole-room), each table posts 1 use + 1 frustration to a shared slide. Doubles as the silent group-formation + tool-assignment step. |
| 0:10–0:20 | Tor: ideas & possibilities (10 min) | The judgment-vs-answers framing; the 4 prompting moves (decompose, role, adversarial, ask-it-to-improve-your-prompt); a 60-sec live teaser of an advanced agent (Claude Code) doing something a chat app can’t — watch, don’t replicate. End by assigning each table its tool. |
| 0:20–0:55 | Challenge 1 (35 min) | Set challenge: The next step — use AI to understand a complex recent paper, then design the follow-up experiment. Run on the per-challenge timeline below. |
| 0:55–1:30 | Challenge 2 (35 min) | Set challenge: This is not a drill — use AI to understand a live federal rule change and draft a substantive public comment. Same timeline; debrief on what made comments impactful, then close on the ethics moment plus one CLI “what’s possible next” pointer. |
How each challenge works
The shape of each challenge Every 35-minute challenge follows the same three-act structure: we frame a problem, you build against it in teams, and then we compare results out loud. The instructor never hands you the method — only the goal and a deliberately weak baseline to improve on. Nobody is competing; the point is to learn from how different tables and tools approached the same task.
- 0:00–0:05 · Challenge introduction (5 min)
- 0:05–0:20 · Work in your group (15 min)
- 0:20–0:22 · Post your best prompt (2 min)
- 0:22–0:32 · Share & debrief (10 min)
- 0:32–0:35 · Reset (3 min)
Prompting & strategy principles
Four moves do most of the work. Hand each group a card with one to “specialize” in, then share across the room in debrief.
1 · Decompose the goal AI executes; you set direction. Before prompting, name the high-level goal, then the sub-tasks and angles that get you there. “Make a figure” is a wish; “show the group × time interaction, with CI error bars, a colorblind-safe palette, and a one-line takeaway caption” is a plan. ask: what would ‘good’ require?
2 · Identity / role prompting Give the agent a persona. “You are a data-viz editor for Nature who hates chartjunk” or “You are an expert open-source developer.” A role front-loads taste, standards, and vocabulary you’d otherwise have to specify line by line. role first, task second
3 · Adversarial self-evaluation Make the model its own harshest reviewer. “Score your confidence 0–100 and list the three weakest claims.” “Find the jargon a 10th grader would miss.” “Where might this hallucinate — flag anything you can’t cite.” This turns a single answer into a checkable draft. trust, then verify
4 · Ask it to improve your prompt The cheapest upgrade: “Rewrite my prompt to get a better result, and tell me what you changed and why.” Great equalizer for novices — the model teaches prompting in real time. meta-prompting
The throughline Answers are cheap; judgment is expensive. Every move above moves effort from producing an answer to auditing one. That’s the skill we’re actually building.
When you are ready, meet your tools in the AI Toolkit, or browse the full Challenge menu.