Microsoft AI-102 (Designing and Implementing a Microsoft Azure AI Solution) Practice Test 22

Question 210

In a multimodal content-generation product, engineers must have the model request a structured application-defined operation with arguments. The expected evidence is that application code receives a typed action request for a permitted function. Which implementation is the best match?

  1. Use a supported DALL-E image-generation model
  2. Use an embedding model
  3. Call a chat-completion or response-capable model
  4. Use a vision-capable multimodal model
  5. Use model tool or function calling

Explanation

Correct Answer

E:

Tool calling lets a model select a declared function and return arguments for the application to execute under its own authorization controls. For this multimodal content-generation product, the intended evidence is that application code receives a typed action request for a permitted function. The model's tool-call suggestion does not itself execute arbitrary privileged code.

Incorrect Answers

A:

Image-generation models produce visual assets from prompts where the model and deployment support image output; it may be used to synthesize a new image from a textual description in a multimodal content-generation product, but the current requirement is instead to have the model request a structured application-defined operation with arguments. That separate capability does not prove that application code receives a typed action request for a permitted function; They do not simply analyze an existing invoice into key-value fields, so selecting that alternative would not meet the stated acceptance check.

B:

Text embeddings produce numeric representations suited to vector search and semantic comparison; it may be used to turn document and query text into vectors for similarity retrieval in a multimodal content-generation product, but the current requirement is instead to have the model request a structured application-defined operation with arguments. That separate capability does not prove that application code receives a typed action request for a permitted function; They do not generate images or compose full conversational answers, so selecting that alternative would not meet the stated acceptance check.

C:

A generative language model consumes instructions and context and produces conversational or textual output; it may be used to generate a natural-language response to a contextual user request in a multimodal content-generation product, but the current requirement is instead to have the model request a structured application-defined operation with arguments. That separate capability does not prove that application code receives a typed action request for a permitted function; Chat generation is not a substitute for embedding creation when vector retrieval is required, so selecting that alternative would not meet the stated acceptance check.

D:

Multimodal models can interpret visual inputs along with a text instruction where the selected deployment supports images; it may be used to answer questions that depend on both a supplied image and text context in a multimodal content-generation product, but the current requirement is instead to have the model request a structured application-defined operation with arguments. That separate capability does not prove that application code receives a typed action request for a permitted function; A text-only model cannot be assumed to inspect image pixels, so selecting that alternative would not meet the stated acceptance check.

Question 211

An embedding request exceeds model input length for a long legal report. What preprocessing should be added?

  1. Split the report into length-appropriate passages and embed each passage separately
  2. Verify each supported modality separately rather than assuming image input implies video input
  3. Match the model's inference task and API operation to its documented capabilities
  4. Treat tool output as untrusted data rather than promoting it to higher-priority instructions

Explanation

Correct Answer

A:

Chunk-level embeddings fit model limits and preserve more retrievable detail than truncating the whole report. In the multimodal content-generation product, confirm the fix by retesting the affected behavior after splitting the report into length-appropriate passages and embed each passage separately.

Incorrect Answers

B:

Although useful elsewhere in the multimodal content-generation product, multimodal support is model- and API-specific; acceptance of still images does not guarantee video processing; the present case is different and calls for the team to split the report into length-appropriate passages and embed each passage separately.

C:

That change addresses another failure mode: model families are not interchangeable endpoints; chat, embedding, and image generation can require different API contracts; it does not address this case, for which the relevant intervention is to split the report into length-appropriate passages and embed each passage separately.

D:

Tool responses can carry prompt injection; integrating tools should preserve the application instruction boundary in its own workflow, but that is a different correction from the present requirement to split the report into length-appropriate passages and embed each passage separately. The reported issue would remain.

Question 212

A customer assistant must retain multiple conversational turns when composing a reply. What should its request include?

  1. Use a consistent supported embedding model and dimensionality for documents and queries
  2. Send the relevant conversation messages or state through a supported chat or responses interface
  3. Select a supported image size or aspect ratio appropriate to the publishing slot
  4. Declare the callable function schema and execute authorized tool calls in the application

Explanation

Correct Answer

B:

Conversational generation depends on supplied context or managed conversation state; independent stateless calls cannot infer missing history. In the multimodal content-generation product, confirm the fix by retesting the affected behavior after sending the relevant conversation messages or state through a supported chat or responses interface.

Incorrect Answers

A:

Cross-model vectors do not generally inhabit the same calibrated similarity space in its own workflow, but that is a different correction from the present requirement to send the relevant conversation messages or state through a supported chat or responses interface. The reported issue would remain.

C:

Although useful elsewhere in the multimodal content-generation product, image generation request options control supported output geometry; correct model selection alone does not make generated images fit a required placement; the present case is different and calls for the team to send the relevant conversation messages or state through a supported chat or responses interface.

D:

That change addresses another failure mode: function calling produces structured tool-intent data; actual external execution remains the host application's responsibility; it does not address this case, for which the relevant intervention is to send the relevant conversation messages or state through a supported chat or responses interface.

Question 213

A prompt says “edit this supplied photo” but the selected image-generation API only creates new images. What should be checked?

  1. Treat low-resolution visual transcription as uncertain and validate with OCR or higher-quality imagery
  2. Choose a model and endpoint that explicitly support image editing or image-conditioned generation
  3. Validate tool arguments and user authorization before invoking the side-effecting operation
  4. Use a compatible model and its supported structured-output facility

Explanation

Correct Answer

B:

Image creation and editing are distinct capabilities; support must be verified for the selected model/version. In the multimodal content-generation product, confirm the fix by retesting the affected behavior after choosing a model and endpoint that explicitly support image editing or image-conditioned generation.

Incorrect Answers

A:

Model image understanding can miss small text; confident wording is not proof of exact character recognition in its own workflow, but that is a different correction from the present requirement to choose a model and endpoint that explicitly support image editing or image-conditioned generation. The reported issue would remain.

C:

That change addresses another failure mode: a model's proposed tool arguments do not constitute proof that the user has permission to perform an action; it does not address this case, for which the relevant intervention is to choose a model and endpoint that explicitly support image editing or image-conditioned generation.

D:

Although useful elsewhere in the multimodal content-generation product, a general text model can generate prose, but reliable programmatic classification needs an enforceable response contract when available; the present case is different and calls for the team to choose a model and endpoint that explicitly support image editing or image-conditioned generation.

Question 214

An image API returns binary or encoded image data, which the client tries to parse as assistant text. What is the implementation error?

  1. Handle the model's documented image response format and content type
  2. Match the model's inference task and API operation to its documented capabilities
  3. Verify each supported modality separately rather than assuming image input implies video input
  4. Treat tool output as untrusted data rather than promoting it to higher-priority instructions

Explanation

Correct Answer

A:

Images are delivered through media-specific payloads or references; chat-text parsing is not a general image decoder. In the multimodal content-generation product, confirm the fix by retesting the affected behavior after handling the model's documented image response format and content type.

Incorrect Answers

B:

Although useful elsewhere in the multimodal content-generation product, model families are not interchangeable endpoints; chat, embedding, and image generation can require different API contracts; the present case is different and calls for the team to handle the model's documented image response format and content type.

C:

That change addresses another failure mode: multimodal support is model- and API-specific; acceptance of still images does not guarantee video processing; it does not address this case, for which the relevant intervention is to handle the model's documented image response format and content type.

D:

Tool responses can carry prompt injection; integrating tools should preserve the application instruction boundary in its own workflow, but that is a different correction from the present requirement to handle the model's documented image response format and content type. The reported issue would remain.

Question 215

The image generator returns the correct subject but crops product artwork unpredictably for a fixed website hero placement. What request property should be reviewed?

  1. Include the image using the model's documented multimodal message format
  2. Select a supported image size or aspect ratio appropriate to the publishing slot
  3. Use a consistent supported embedding model and dimensionality for documents and queries
  4. Declare the callable function schema and execute authorized tool calls in the application

Explanation

Correct Answer

B:

Image generation request options control supported output geometry; correct model selection alone does not make generated images fit a required placement. In the multimodal content-generation product, confirm the fix by retesting the affected behavior after selecting a supported image size or aspect ratio appropriate to the publishing slot.

Incorrect Answers

A:

That change addresses another failure mode: a vision-capable model still needs the image content or supported image reference alongside the question; it does not address this case, for which the relevant intervention is to select a supported image size or aspect ratio appropriate to the publishing slot.

C:

Although useful elsewhere in the multimodal content-generation product, cross-model vectors do not generally inhabit the same calibrated similarity space; the present case is different and calls for the team to select a supported image size or aspect ratio appropriate to the publishing slot.

D:

Function calling produces structured tool-intent data; actual external execution remains the host application's responsibility in its own workflow, but that is a different correction from the present requirement to select a supported image size or aspect ratio appropriate to the publishing slot. The reported issue would remain.

Question 216

A multimodal content-generation product must generate a natural-language response to a contextual user request. Reviewers will accept the implementation when the application receives a natural-language answer. Which approach best fulfills that requirement?

  1. Use model tool or function calling
  2. Use a vision-capable multimodal model
  3. Use an embedding model
  4. Call a chat-completion or response-capable model

Explanation

Correct Answer

D:

A generative language model consumes instructions and context and produces conversational or textual output. For this multimodal content-generation product, the intended evidence is that the application receives a natural-language answer. Chat generation is not a substitute for embedding creation when vector retrieval is required.

Incorrect Answers

A:

Tool calling lets a model select a declared function and return arguments for the application to execute under its own authorization controls; it may be used to have the model request a structured application-defined operation with arguments in a multimodal content-generation product, but the current requirement is instead to generate a natural-language response to a contextual user request. That separate capability does not prove that the application receives a natural-language answer; The model's tool-call suggestion does not itself execute arbitrary privileged code, so selecting that alternative would not meet the stated acceptance check.

B:

Multimodal models can interpret visual inputs along with a text instruction where the selected deployment supports images; it may be used to answer questions that depend on both a supplied image and text context in a multimodal content-generation product, but the current requirement is instead to generate a natural-language response to a contextual user request. That separate capability does not prove that the application receives a natural-language answer; A text-only model cannot be assumed to inspect image pixels, so selecting that alternative would not meet the stated acceptance check.

C:

Text embeddings produce numeric representations suited to vector search and semantic comparison; it may be used to turn document and query text into vectors for similarity retrieval in a multimodal content-generation product, but the current requirement is instead to generate a natural-language response to a contextual user request. That separate capability does not prove that the application receives a natural-language answer; They do not generate images or compose full conversational answers, so selecting that alternative would not meet the stated acceptance check.

Question 217

The assistant must call an inventory API, but generated text includes a pretend JSON command instead of a real tool request. What integration is needed?

  1. Select a supported image size or aspect ratio appropriate to the publishing slot
  2. Include the image using the model's documented multimodal message format
  3. Declare the callable function schema and execute authorized tool calls in the application
  4. Use a consistent supported embedding model and dimensionality for documents and queries

Explanation

Correct Answer

C:

Function calling produces structured tool-intent data; actual external execution remains the host application's responsibility. In the multimodal content-generation product, confirm the fix by retesting the affected behavior after declaring the callable function schema and execute authorized tool calls in the application.

Incorrect Answers

A:

Although useful elsewhere in the multimodal content-generation product, image generation request options control supported output geometry; correct model selection alone does not make generated images fit a required placement; the present case is different and calls for the team to declare the callable function schema and execute authorized tool calls in the application.

B:

That change addresses another failure mode: a vision-capable model still needs the image content or supported image reference alongside the question; it does not address this case, for which the relevant intervention is to declare the callable function schema and execute authorized tool calls in the application.

D:

Cross-model vectors do not generally inhabit the same calibrated similarity space in its own workflow, but that is a different correction from the present requirement to declare the callable function schema and execute authorized tool calls in the application. The reported issue would remain.

Question 218

A release for a multimodal content-generation product is blocked until the application can answer questions that depend on both a supplied image and text context. The tester will verify that the response refers to the contents of the attached picture. What should be implemented?

  1. Use model tool or function calling
  2. Use an embedding model
  3. Call a chat-completion or response-capable model
  4. Use a vision-capable multimodal model

Explanation

Correct Answer

D:

Multimodal models can interpret visual inputs along with a text instruction where the selected deployment supports images. For this multimodal content-generation product, the intended evidence is that the response refers to the contents of the attached picture. A text-only model cannot be assumed to inspect image pixels.

Incorrect Answers

A:

Tool calling lets a model select a declared function and return arguments for the application to execute under its own authorization controls; it may be used to have the model request a structured application-defined operation with arguments in a multimodal content-generation product, but the current requirement is instead to answer questions that depend on both a supplied image and text context. That separate capability does not prove that the response refers to the contents of the attached picture; The model's tool-call suggestion does not itself execute arbitrary privileged code, so selecting that alternative would not meet the stated acceptance check.

B:

Text embeddings produce numeric representations suited to vector search and semantic comparison; it may be used to turn document and query text into vectors for similarity retrieval in a multimodal content-generation product, but the current requirement is instead to answer questions that depend on both a supplied image and text context. That separate capability does not prove that the response refers to the contents of the attached picture; They do not generate images or compose full conversational answers, so selecting that alternative would not meet the stated acceptance check.

C:

A generative language model consumes instructions and context and produces conversational or textual output; it may be used to generate a natural-language response to a contextual user request in a multimodal content-generation product, but the current requirement is instead to answer questions that depend on both a supplied image and text context. That separate capability does not prove that the response refers to the contents of the attached picture; Chat generation is not a substitute for embedding creation when vector retrieval is required, so selecting that alternative would not meet the stated acceptance check.

Question 219

A team responsible for a multimodal content-generation product needs to synthesize a new image from a textual description; the acceptance check is that the response contains newly generated image data. Which capability should be selected?

  1. Use a supported DALL-E image-generation model
  2. Use model tool or function calling
  3. Use an embedding model
  4. Call a chat-completion or response-capable model

Explanation

Correct Answer

A:

Image-generation models produce visual assets from prompts where the model and deployment support image output. For this multimodal content-generation product, the intended evidence is that the response contains newly generated image data. They do not simply analyze an existing invoice into key-value fields.

Incorrect Answers

B:

Tool calling lets a model select a declared function and return arguments for the application to execute under its own authorization controls; it may be used to have the model request a structured application-defined operation with arguments in a multimodal content-generation product, but the current requirement is instead to synthesize a new image from a textual description. That separate capability does not prove that the response contains newly generated image data; The model's tool-call suggestion does not itself execute arbitrary privileged code, so selecting that alternative would not meet the stated acceptance check.

C:

Text embeddings produce numeric representations suited to vector search and semantic comparison; it may be used to turn document and query text into vectors for similarity retrieval in a multimodal content-generation product, but the current requirement is instead to synthesize a new image from a textual description. That separate capability does not prove that the response contains newly generated image data; They do not generate images or compose full conversational answers, so selecting that alternative would not meet the stated acceptance check.

D:

A generative language model consumes instructions and context and produces conversational or textual output; it may be used to generate a natural-language response to a contextual user request in a multimodal content-generation product, but the current requirement is instead to synthesize a new image from a textual description. That separate capability does not prove that the response contains newly generated image data; Chat generation is not a substitute for embedding creation when vector retrieval is required, so selecting that alternative would not meet the stated acceptance check.

Question 220

A vision model confidently reports a tiny serial number that cannot be read at the provided resolution. How should the result be handled?

  1. Validate tool arguments and user authorization before invoking the side-effecting operation
  2. Call an embedding model to produce numeric vectors for the index field
  3. Treat low-resolution visual transcription as uncertain and validate with OCR or higher-quality imagery
  4. Use a compatible model and its supported structured-output facility

Explanation

Correct Answer

C:

Model image understanding can miss small text; confident wording is not proof of exact character recognition. In the multimodal content-generation product, confirm the fix by retesting the affected behavior after treating low-resolution visual transcription as uncertain and validate with OCR or higher-quality imagery.

Incorrect Answers

A:

That change addresses another failure mode: a model's proposed tool arguments do not constitute proof that the user has permission to perform an action; it does not address this case, for which the relevant intervention is to treat low-resolution visual transcription as uncertain and validate with OCR or higher-quality imagery.

B:

Text responses are not automatically vector representations; vector indexes expect numeric arrays of configured dimensions in its own workflow, but that is a different correction from the present requirement to treat low-resolution visual transcription as uncertain and validate with OCR or higher-quality imagery. The reported issue would remain.

D:

Although useful elsewhere in the multimodal content-generation product, a general text model can generate prose, but reliable programmatic classification needs an enforceable response contract when available; the present case is different and calls for the team to treat low-resolution visual transcription as uncertain and validate with OCR or higher-quality imagery.

Designing and Implementing a Microsoft Azure AI Solution (AI-102) Practice Tests