TALAMANA · THE AI LITERACY MAP FOR ARCHITECTURE AND DESIGN · Judgment · AGE 18—20 · POSITIONAL · HELD
Curation is diagnosis
The batch shows the gap between what you meant and what the machine heard.
The dilemma
Twelve images are on the table. You can treat them as a shop (which one do I want?) or as a report (what do these twelve say about the words I typed?). The first is quicker and feels like progress. The second is slower and tells you something about your own brief.
The choices
Shop: pick the closest one and move on. Or sort the twelve into families, name what each family has in common, and read each family as a sentence about the brief: "the brief said open, and the model heard no walls."
The consequence
Shopping leaves the brief exactly as vague as it was, so the next batch drifts the same way, and so does the one after that. Reading the batch as a report shows you the gap between what you meant and what the machine heard, and the next brief closes that gap. The first loop goes round in circles. The second one gets somewhere.
The case
The brief asks for "a light, open verandah house for Dharwad." Nine of twelve come back with glass walls and white floors. The student is about to pick the least glassy one. The tutor asks a different question: what did the model hear? It heard light as brightness, not as a light building. It heard open as transparent, not as shaded and ventilated right through. The next brief says "deep verandahs, filler-slab roof, shade first," and the batch changes character completely. The model did not change. The reading of the batch did.
Our position
The middle phase between generating and refining, which the Lab calls curation, is where the real learning happens, and it is the phase most people skip. Its job is to read the batch for what the brief actually said. Everything else in the loop depends on that reading.
Why we hold it
The Lab's workshop exercise is built around this phase, and the failure pattern is easy to spot in a log: three near-identical briefs, three near-identical sets, and a selection made on feel. The pattern the exercise wants is the opposite: a brief that changed because the previous batch exposed an assumption. The log shows which of the two happened, and the difference is obvious.
The strongest objection
A trained eye shops well. An experienced designer looking at twelve images is already reading them. The diagnosis is happening silently, inside the choosing. Making the step explicit may help a beginner and only slow down an expert. A routine that helps novices is not a law for everyone.
What would make us revise it
Evidence that experienced practitioners who choose fast write second briefs as sharp as those who diagnose on paper. If their logs show the brief improving the same way without the written step, the step becomes training wheels: kept for the first year, dropped after that.
Try it
Sort a twelve-image batch into families. For each family write one line: what this family reveals about the brief. Then, using only the outputs, write the brief the model actually heard. Put it next to the brief you wrote. The gap is your next revision.
Take it to crit
Ask the student what the batch taught them about their own brief, not which image they like. If the answer is a preference, ask again.
How it works
CONTACT SHEET taught you to see the set before any frame. This card takes the next step: reading the set as a report on the words you wrote. A model answers the words you gave it, read through the habits in its training. So the batch records two things at once: your words, and the model's reading of them. Sorting the batch into families separates the two. Where every family agrees, the model read you clearly. Where the families split, your brief was ambiguous. Where every family does something you never asked for, you have found the model's habit, and now you can write against it.
What this idea builds on
What this idea opens up
Sources
- saurashtra spine — Dreaming·Curation·Grounded trio
- Dreaming Engines and Grounded Generators
- Your Prompt Is Your Brief
Open this idea on the map · The complete map · Logika · RBDS AI Lab, India · revised every edition.