How to use this board
This page is your lab notebook — it doesn't talk to a model. Open any chat model (ChatGPT, Claude, Gemini…) in another window, run the prompts there, and log what happened here. Everything you type saves in this browser automatically.
- Pick one small topic — a story, an apology note, a lesson intro. You'll reuse it in every row.
- A — Baseline. Write a plain prompt on your topic, run it, and note what the model produced.
- B — Change one thing. Same prompt with exactly one deliberate change: the genre, the audience, the tone, or a constraint. Run it in a fresh chat so answer A can't influence it. Log what changed.
- C — Respond to something. This one stays in the chat: pick something in the output that surprised you — a wrong fact, an odd assumption, a detail you never asked for — and write a follow-up prompt that pushes on it. You're now testing how the model behaves with context and a challenge.
- D — Push further (optional). Exaggerate the change from B, or stack two changes, until the output breaks or gets strange. Finding the edge is the point.
- When you're done, write your evidence-based claim below, then hit Copy my notes (top of the page). It copies everything you've typed here as formatted notes — paste them into Zoom chat, the session thread, or your own planning doc. Print / Save as PDF makes a clean handout instead.
Worked example — one A → B pair
A (baseline): "Write a four-sentence story about a family dinner." B (one change — genre): "Write a four-sentence story about a family dinner, as a nature-documentary voiceover."
What changed in the output: same dinner, new voice — B added a hushed narrator, "the adolescent male," and suspense nobody asked for. What I learned: one genre word swapped the whole vocabulary; the model carries a genre's defaults with it.