In the last activities you used AI to answer questions and do tasks that have a right answer. Creative work is different: there is no key. Two people can look at the same picture and disagree about it. But you can still compare tools fairly, if you do two things first. Write down what you are asking for, and write down how you will tell whether you got it.
The brief
A brief is a short, specific description of what you want. This one has slots: a theme, a mood, up to two things it must include, one thing to avoid, and a form (a scene, a picture, a loop, a page). The form is what changes from station to station. The rest stays the same, which is what makes the comparison fair.
A station is one kind of AI tool you try your brief on. You'll use at least three, and your teacher decides which ones are available.
Four kinds of station
Station
The brief becomes
Watch for
✍️ Words
Write a scene of about 60 words about [theme].
Word counts, and facts or details that are invented.
🖼️ Image
Make one picture of [theme].
How many of something there are, text inside the picture, small objects.
🎧 Sound
Make a 20-second music loop that fits [theme].
Length, which instruments you get, and whether the loop repeats cleanly.
🧩 Build
Make a one-screen poster page for [theme].
Whether it runs at all, and whether the text is readable.
Brief lab · every concrete slot is something you can check later
Station
Theme
Mood (a taste)
Must include (up to two)
Avoid (one)
Round 1 prompt: paste the brief exactly as it is
Make one picture of a night market in the rain. Mood: cozy but a little mysterious.
Yes/no criteria this brief gives you (1)
Has no letters or words inside the picture
"cozy but a little mysterious" is a taste. You can rate it 1 to 5, but two people won't agree on a yes or no. This brief has nothing concrete to check except the form, so afterward all you could say is whether you liked it.
Checkable or taste?
Look at what the lab did. "Includes a stray cat" is a criterion: a stranger could look at the result and answer yes or no. "Cozy but a little mysterious" is a taste: it matters, and you should say how you felt about it, but you can't score it yes or no, because you and your partner may feel differently. Keep the two apart. Score the criteria, and rate the taste from 1 to 5 on its own line.
🔑Write the criteria before you run anything. If you write them after you've seen the result, you'll unconsciously write what the result already does. This is the same rule as the Productivity Study: decide what good looks like first.
How does a generator know what to make?
When you ask for a tiger, the tool has no tiger photo to look up. Image generators are trained on very large collections of images paired with text descriptions, and they learn which patterns of color, shape and texture go with which words. A request is answered by combining those patterns, which is why a tool can draw a pink tiger dancing even though no such photo exists. Text and music tools are trained the same general way, on large collections of writing and audio. Who made those collections, and whether they agreed, is a real question, and the next activity (2.2.2) is about it.
⚠️Ground rules for every station. Use only tools your teacher approved. Many tools have their own age requirements, which is one reason a tool can work at home and not at school. Don't upload photos of people, and don't type anyone's personal details. Describe what you want in your own words: don't name a living artist or a copyrighted character in a prompt, because whose style something is gets its own activity.
Principles of AI · 2.2.1Skill⚠ Watch Out
Change One Thing
round 1 = the brief · every round after changes one thing
Your first result will rarely be your best, so you iterate: look at what came back, change the prompt, and try again. Round 1 is always the brief, pasted exactly as written.After that, each round changes one thing, and you write down what it was.
Why one? Suppose you change three things and the result improves. You don't know which change helped. Maybe one helped, one did nothing, and one was a coin flip. If you change two things and the result gets worse, you can't tell which change broke it. A log with one change per round tells you something you can use on the next brief.
Below is an invented six-round log. For each round, decide what the log lets you say about the change. Look at which criteria were met, not only how many. In round 4 the score stays at 3 of 4, but one criterion was gained and another was lost.
Log lab · an invented image-station log, six rounds
Round 1 prompt (the brief, pasted as is):
Make one picture of a night market in the rain. Mood: cozy but a little mysterious. Include a red paper lantern and a stray cat. Avoid people.
Round 11 / 4
No changes: this is the brief itself.
What came back (invented): A wet street at night with glowing stalls and a red lantern over one of them. Three vendors stand behind the counters. A sign in the corner reads “NITE MARKRT.” No cat.
✓ red lantern✗ stray cat✗ no people✗ no text in picture
Failure tags: Left out (stray cat) · Broke a limit (no people) · Garbled (no text in picture)
Round 22 / 4
What I changed (1): Moved “a stray cat” to the start of the prompt
What came back (invented): A cat sits under a stall in the rain. The lantern is still there. Three vendors are still behind the counters and the sign still has garbled letters.
✓ red lantern✓ stray cat✗ no people✗ no text in picture
Failure tags: Broke a limit (no people) · Garbled (no text in picture)
Gained: stray cat.
What can the log tell you about this round's change?
Round 33 / 4
What I changed (1): Added “an empty street, no people”
What came back (invented): An empty street. The lantern and the cat are both there. The sign in the corner still reads “NITE MARKRT.”
✓ red lantern✓ stray cat✓ no people✗ no text in picture
Failure tags: Garbled (no text in picture)
Gained: no people.
What can the log tell you about this round's change?
Round 43 / 4
What I changed (2): Added “no signs or text anywhere” + Swapped the mood words for “warm lantern light, fog”
What came back (invented): No sign and no text anywhere. The street glows warm with fog. The lantern is there and there are no people, but the cat has disappeared.
✓ red lantern✗ stray cat✓ no people✓ no text in picture
Failure tags: Drift (stray cat)
Gained: no text in picture. Lost: stray cat. The score didn't change, but the pattern did.
What can the log tell you about this round's change?
Round 53 / 4
What I changed (1): Added “high detail, best quality”
What came back (invented): Sharper and more detailed, but nothing that was missing came back. Still no cat.
✓ red lantern✗ stray cat✓ no people✓ no text in picture
Failure tags: Left out (stray cat)
Same criteria as last round.
What can the log tell you about this round's change?
Round 64 / 4
What I changed (1): Rewrote the cat line as “a stray cat sitting in the foreground, facing the viewer”
What came back (invented): The cat sits in the foreground, facing the viewer. Lantern, no people, no text. All four.
✓ red lantern✓ stray cat✓ no people✓ no text in picture
Gained: stray cat.
What can the log tell you about this round's change?
Answer all five rounds to see the summary (0 of 0 right so far).
Why it fails: five tags
When a criterion isn't met, name why. The tag points to the kind of change that is likely to help.
Tag
What happened
A good next change
Left out
A must-include is missing.
Move it to the start, or say it more directly.
Broke a limit
Something you asked it to avoid is there.
State the limit plainly: “no people,” not “keep it quiet.”
Garbled
Letters, counts, or shapes came out wrong.
Take the demand out and add it by hand afterward, or ask for it separately.
Drift
Something that was right got lost or changed.
Restate what must stay, and undo the last change.
Generic
It fits the brief but could be anyone's.
Add one specific, unusual detail.
Generic results are common because, when your prompt leaves something open, a tool tends to fill the gap with the most common choice. "A lantern" gets you the usual lantern. "A red paper lantern with a torn corner" gets you something closer to yours. This is a tendency, not a rule, so check it in your own log.
💡When do you stop? When every criterion is met, or after five rounds at one station, whichever comes first. If you reach round 5 and still miss a criterion, that is a result too. Write it down and move on. Keep a copy of your best round so far, so a bad change can't cost you a good result.
Principles of AI · 2.2.1ConceptExample
What Your Log Can Show
brief-fit and taste · what you decided · what you can claim
After three stations you have three logs. Put the best round from each into one table. Compare the stations on the things you recorded: how many criteria the best round met (brief-fit), how many rounds it took to get there, and how much you liked it (taste).
An invented comparison
Station
Rounds used
Brief-fit
Taste (1–5)
What it tells you
Words
3
4 / 4
3
Does what the brief asked, but it didn't excite you. Try more specific details.
Image
6
4 / 4
5
Matches the brief and you like it.
Sound
5
2 / 4
4
Liked, but it isn't what the brief asked for. Decide which matters for this job.
Neither number is the right one. Brief-fit tells you how well the tool followed the instructions. Taste tells you whether you'd keep it. A tool that follows instructions closely and produces something dull is a different tool from one that surprises you and ignores half your brief. Which one you want depends on the job.
Whose work is it?
Yours
The brief. The criteria. Which result to keep. Which one thing to change next. When to stop.
The tool's
The pixels, words, notes or code that came back, and the details it filled in that you didn't ask for.
Your log is the proof of your side. When you share or hand in something made with a tool, add one line to the AI-use statement from 2.1.3:
AI-use statement
Tool: [which tool]
Used for: [what, in one line]
Not used for: [what you did yourself]
What I checked: [and how]
Policy that applies: [source and the words]
My decisions: [the brief, the criteria, what I kept, what I changed]
"Does it understand?"
This question comes up the first time a tool nails your brief, and again the first time it ignores half of it. It's a good question, but a log can't answer it directly. Your log shows what came out: the lantern was there, the cat wasn't. It does not show what happened inside the tool, or why. You can ask a sharper question that a log can answer: did it do what a person who understood the brief would do? Count where it did and where it didn't.
Sort the claims below by how far your log can take them.
Claim lab · what can a log support?
An invented student, Ravi, ran three stations and wants to say these things. Sort each one.
In round 1 the image tool left out the cat and added three vendors.
My sound station never scored above 2 of 4 in five rounds.
Moving the cat to the start of the prompt is what fixed it.
Image tools always ignore small details.
The AI understood the mood but didn't understand the cat.
The tool was trained on stolen art.
The words station reached 4 of 4 in three rounds.
The AI wants to be creative.
Sort all eight to see how you did.
🔑A strong report says what happened in the rounds, marks what needs more runs, and leaves out claims about what's inside the AI. A short, honest claim is stronger than a big one you can't support.