A pool-lane booking screenshot must inform a multimodal test prompt. A colleague only types lane.png into the chat. What is the key limitation of that approach?
Select an answer to reveal the explanation.
Short Explanation
Saying “lane.png” is like telling a friend the name of a photo album without opening a single picture. Multimodal use means attaching the actual image so it can be embedded or tokenized as vision input. A filename string alone gives the model nothing to see.
Full Explanation
Typing only lane.png into chat gives the model a text string, not image pixels, so it cannot analyze the visual content of a pool-lane booking screenshot. Multimodal use requires attaching the actual image so vision embeddings or vision tokens are prepared alongside any text. Civic recreation testers confuse file naming with file attachment surprisingly often; this item locks the distinction before false confidence spreads.
Claims that filenames are encrypted so models refuse any prompt mentioning PNG extensions are myth. Claims that typing a filename is enough because transformers automatically download every named file from the internet are myth and unsafe as a design assumption for municipal systems. Claims that screenshots cannot be used in testing aids and only API logs are valid multimodal input falsely narrow modality and ignore UI-oriented GenAI assistance.
Exam caveat: each modality must be supplied in processable form—text as tokens, images as visual input. A path or filename string remains text only, regardless of extension.
Operational check: retry the lane-booking prompt with the real image attached; ask the model to list visible controls and labels; confirm it references on-screen content absent from the filename string; update the municipal multimodal checklist to ban filename-only “attachments” in test prompts.