A building-permit portal tester pastes a wireframe into a multimodal chat and writes only “what do you see?” Which change best aligns the prompt with a testing goal?
Select an answer to reveal the explanation.
Short Explanation
“What do you see?” is tourist talk, not a test charter. Steer the multimodal prompt toward mismatches with the story or cases for unlabeled controls. Give the model a testing job, not a generic caption request.
Full Explanation
Rewrite a building-permit portal multimodal prompt from generic “what do you see?” to a concrete testing objective—list mismatches with the story or propose cases for unlabeled controls. Effective multimodal testing prompts state a testing job so the model’s vision serves analysis or design. Civic permit wireframes need condition discovery and case ideas, not tourist captioning that never becomes testware.
Keeping a caption-only prompt to remain “unbiased by any test objective” yields descriptions that may not advance test conditions or cases. Asking the model to report neuron activation maps for each pixel cluster treats the model as the system under test. Replacing the wireframe with a jailbreak string that forces unrestricted system output abandons the testing goal for unsafe elicitation.
Exam caveat: multimodal capability on this exam must be aimed at a testing goal—compare, cover, draft—not open-ended captioning, model internals, or jailbreaks.
Operational check: replace “what do you see?” with an explicit compare-or-case instruction; attach wireframe plus story; verify the reply contains mismatches or draft conditions usable in review; save the improved prompt template in the permit portal municipal prompt library.