TALAMANA · THE AI LITERACY MAP FOR ARCHITECTURE AND DESIGN · Judgment · AGE 18—21 · POSITIONAL · HELD

Was the brief the failure?

The batch is the most honest reader your brief will ever have.

The dilemma

The batch is wrong. Not ugly, wrong: the wrong building, the wrong climate, the wrong idea. Your hand is already moving towards regenerate, or towards a different tool. Before it gets there, ask what the outputs show the brief really asked for.

The choices

Blame the engine and re-roll, or switch engines. Or read the batch as a faithful answer to your words, find the word that produced the wrong building, and rewrite the brief before you generate again.

The consequence

Blame the engine and you carry the same loose brief into the next batch and the next tool, and get the same wrongness in a different style, now convinced the tools are bad. Blame the brief first and most "model failures" turn out to be briefing failures you can fix in one line. That is cheaper than any re-roll, and it teaches you something about what you actually meant.

The case

Brief: "a contemporary house for a hot, humid coastal site, with local materials." Batch: twelve white boxes with floor-to-ceiling glass, two with palm trees. The student says the model does not understand the tropics. The tutor asks what contemporary usually looks like in the photographs the model learnt from. White boxes and glass. The brief asked for the wrong thing, and asked for it clearly. Rewritten as "deep overhangs, laterite walls, a Mangalore-tile roof, cross-ventilated, verandah on the windward side," the next batch is recognisably coastal Karnataka. The model did not change.

Our position

When a batch disappoints, the first suspect is your own brief — suspect, not convict. Reading your own brief critically is the mature user's first move, and it comes before any re-roll, any model switch and any complaint about the tool.

Why we hold it

The batch is the most honest reader your brief will ever have. It takes your words literally and answers with the most likely picture. Most disappointment is the gap between what you meant and what you typed, and that gap is yours to close. The Lab teaches this first because it is the move that makes every other card on this strand work.

The strongest objection

Always suspecting your own brief first teaches user-blame. A failed batch belongs to the whole interaction — the brief, the model, its training coverage, the interface, the conditioning you gave it, the settings, the task — and blaming the brief by reflex is no more rigorous than blaming the machine by reflex. Worse, it breeds prompt superstition: if the machine fails, keep rewriting yourself until it behaves. Engines have blind spots no rewording closes (under-photographed regions, drawing conventions, dimensions), and the student sent back to the brief every time learns to doubt a brief that was fine, for a tool that cannot do the job.

What would make us revise it

The position is about the first suspect, and "diagnose the brief" is closer to what we mean than "blame" it. Read the batch, find the word; if there is no word to find, the tool goes to the bake-off. We would revise the card if workshop logs showed students spending more cycles rewriting briefs that were fine than they saved by catching loose ones. So far the loose brief has been the commoner cause, but the log is where that gets checked, not the slogan.

Try it

Take a failed batch and find the fault in the brief before you touch the model: the word that produced the wrong building. Then rewrite the brief to fix a failure you had first blamed on the engine, and run it once.

Take it to crit

When the student's outputs miss, do they reach for the re-roll, or for the brief? Ask them to show the word in the brief that caused the miss. If they cannot find it, ask them to read the batch again.

How it works

A model answers the words you gave it, read through the habits of its training. "Contemporary" is a word with a strong default in the photographs. "Local materials" is a phrase with almost none, because captions rarely name the material. So the strong word wins and the weak phrase vanishes. Reading the batch tells you which of your words were strong and which were weak. The rewrite replaces the weak phrases with specific nouns the model has seen captioned: laterite, Mangalore tile, verandah.

What this idea builds on

What this idea opens up

Sources

Open this idea on the map · The complete map · Logika · RBDS AI Lab, India · revised every edition.

Age grows from 11 at the centre to 22 at the edge, and six sectors show the learning strands. Tab into the map and the arrow keys step from idea to idea, following the links where there is one. Enter opens the idea under the cursor, and E reads out its links and the reason recorded on each. Press slash for Search, question mark for the full key list, and Escape to leave. Open Ideas for the complete readable list, including what each idea builds on and what it opens up.

LOGIKA · RBDS AI LAB INDIA
ON-RAMP · AGE 11 · FIRST ENCOUNTERS, NOT GATES — IDEAS · — DEPENDENCIES
DONE
OPENS NEXT
SOLID — STANDS ON