Unit 6: Advanced Generative AI Applications - Practice Quiz

1 What is Vertex AI?

2 Which company provides the Vertex AI platform?

3 What is the main purpose of Generative AI Studio?

4 Which activity is commonly performed in Generative AI Studio?

5 What type of model is Gemini?

6 The term multimodal in Gemini models means the model can process:

7 What is Model Garden in Vertex AI?

8 Which task can you perform using Model Garden?

9 Which is typically the first step in an end-to-end AI development workflow?

10 What does model deployment refer to in the AI workflow?

11 What does text-to-image generation do?

12 In text-to-image generation, the prompt is:

13 What is the goal of image understanding?

14 Which task is an example of image understanding?

15 What does video analysis typically involve?

16 Which of the following is a common video analysis task?

17 Which is an example of an audio processing task?

18 Speech-to-text technology converts:

19 What is cross-modal reasoning?

20 A multimodal application is one that:

21 A data science team wants a single managed platform to train custom models, deploy them to endpoints, and also access pre-trained foundation models without stitching together separate services. Which characteristic of Vertex AI best addresses this need?

22 A developer wants to quickly experiment with prompts, tune generation parameters like temperature, and test model responses through a visual interface before writing production code. Which tool is most appropriate?

23 Gemini models are frequently described as natively multimodal. What does this most directly imply for an application developer?

24 A team is evaluating whether to use a Google first-party model, an open-source model, or a third-party partner model for their project. Which Vertex AI feature lets them discover, compare, and deploy from all these categories in one place?

25 In a typical end-to-end AI development workflow on Vertex AI, which sequence correctly orders the core phases?

26 When using a text-to-image model, a user's prompt produces images that ignore a key requested detail. Which prompting adjustment most directly improves adherence to that detail?

27 A retail app sends a product photo to a multimodal model and asks, "How many items are on the shelf and what colors are they?" This task is an example of which capability?

28 A model is asked to identify the moment a goal is scored in a soccer clip and describe what happened just before it. Which capability does this require beyond simple frame classification?

29 A voice-note app needs to convert spoken meeting recordings into searchable text and then summarize action items. Which pipeline correctly reflects the audio processing steps involved?

30 An app receives an image of a recipe card and an audio question asking, "Can I make this if I'm allergic to nuts?" Answering requires combining information from both inputs. What is this ability called?

31 When designing a multimodal customer support assistant that accepts screenshots and typed questions, which design decision best improves reliability of responses?

32 A developer needs the lowest-latency, cost-efficient Gemini variant for high-volume, simple classification tasks, while another team needs maximum reasoning quality for complex analysis. What is the key trade-off guiding this choice?

33 A team wants to reduce hallucinations by having their Vertex AI model reference their internal document store when answering. Which approach fits within the Vertex AI ecosystem?

34 In Generative AI Studio, a developer lowers the temperature setting close to 0 for a factual Q&A assistant. What effect should they expect?

35 A designer wants generated images to avoid including any text or watermarks. Which prompting technique is most directly suited to this goal?

36 A startup wants to fine-tune an open-source model they found in Model Garden on their proprietary dataset before deploying. Which statement about this workflow is accurate?

37 After deploying a model, a team notices prediction quality degrading over weeks as real-world data shifts. Which workflow phase addresses this, and what is the phenomenon called?

38 A document-processing app uses a multimodal model to read scanned invoices and extract the total amount and vendor name into structured fields. What is the best description of this task?

39 A content team wants automatic chapter markers and a summary for hour-long tutorial videos. Which combination of capabilities does the model need?

40 A multimodal app must handle cases where a user uploads an image but sends no text, or sends text with no image. Which design principle best manages this variability?

41 A team migrates a custom training job from a raw Compute Engine setup to Vertex AI to gain managed MLOps capabilities. They require reproducible pipelines, metadata lineage tracking, and automated model registration on successful evaluation. Which combination of Vertex AI components most directly satisfies all three requirements?

42 A prompt engineer notices that identical prompts in Generative AI Studio produce highly variable outputs across runs, harming reproducibility for a compliance report. Which parameter adjustment most directly reduces this stochasticity while preserving the model?

43 An application must reason jointly over a -hour video, its audio track, and a set of PDF technical diagrams within a single request. Which capability of Gemini models is the primary enabler, and what is the key practical constraint?

44 A developer wants to deploy an open-weights model (e.g., a Llama variant) from Model Garden for low-latency, dedicated serving with full control over the endpoint. Which deployment path is most appropriate?

45 In a Vertex AI pipeline, a model passes offline evaluation but degrades sharply after deployment as real traffic shifts. Which monitoring signal most directly identifies this specific failure mode, and what is the correct corrective step?

46 A designer using a text-to-image model finds that adding a negative prompt like blurry, extra fingers improves quality. Mechanistically, in a diffusion model using classifier-free guidance, what does a negative prompt influence?

47 A multimodal model is asked to count the number of red cars in a dense parking-lot photo and returns an incorrect count. Which prompting or design strategy is most likely to improve counting accuracy without changing the model?

48 When sending a long video to a Gemini model, a developer observes that fine-grained fast motion events are sometimes missed. Which factor most directly explains this and what is the appropriate mitigation?

49 A speech application must transcribe overlapping speakers, attribute utterances to each, and detect the emotional tone. Which combination of capabilities is required?

50 A model is shown a chart image and a paragraph of text that contradict each other about a sales figure. When asked for the correct value, the model should ideally do what to demonstrate robust cross-modal reasoning?

51 A production multimodal assistant must minimize latency and cost while handling both quick text queries and occasional heavy video-analysis requests. Which architectural pattern best balances these goals?

52 A developer needs the Gemini model to return strictly parseable output that conforms to a predefined schema for downstream automation. Which approach is the most reliable?

53 A team wants to compare a fine-tuned model against the base model on a held-out set and only promote the challenger if it improves a business metric by a statistically meaningful margin. Which workflow construct best encodes this gate?

54 A user increases the guidance scale (CFG) very high to force stronger adherence to a detailed prompt. What is the most likely trade-off they will observe?

55 For an OCR-heavy document understanding task where layout matters (tables, columns), which failure is most characteristic of a general multimodal model versus a specialized layout-aware pipeline?

56 A developer wants a Gemini model to answer "at what timestamp does the speaker mention pricing?" for an uploaded video. What must the design account for to make timestamp answers reliable?

57 An audio understanding system must handle a -minute podcast but the model has a token budget that the full audio exceeds. Which strategy preserves the most useful information within the constraint?

58 In contrastive multimodal embedding models (e.g., CLIP-style), image and text are projected into a shared space. Which statement best characterizes what "alignment" means and a key limitation?

59 A multimodal medical-imaging assistant is being designed. Beyond accuracy, which design decision is most critical for safe and responsible deployment?

60 A regulated enterprise must ensure that data sent to Vertex AI Gemini endpoints stays within a specific region and is not used to train Google's foundation models. Which combination of controls addresses these requirements?