Amazon AIF-C01 Online Practice Questions and Exam Preparation

Amazon AIF-C01 Online Questions & Answers

• Question 361:

  HOTSPOT

  A company wants to build an ML application.

  Select and order the correct steps from the following list to develop a well-architected ML workload. Each step should be selected one time.

  

  

  Explanation:

  The typical sequence of steps in building an ML application involves:

  1. Defining the business goal and framing the ML problem - This step involves understanding the business need and determining how ML can address it.

  2. Developing the model - This involves selecting the appropriate algorithm, training the model, and evaluating it.

  3. Deploying the model - Once the model is trained, it is deployed to production to serve predictions.

  4. Monitoring the model - After deployment, the model is monitored to ensure it performs well over time and to detect any issues like model drift.

• Question 362:

  Which technique can a company use to lower bias and toxicity in generative AI applications during the post-processing ML lifecycle?

  A. Human-in-the-loop
  B. Data augmentation
  C. Feature engineering
  D. Adversarial training
  A. Human-in-the-loop

  ---

  Human-in-the-loop (HITL) is a technique where human oversight is involved in the decision-making process of AI systems. In the context of generative AI applications, HITL can be used during the post- processing phase to identify and mitigate biases or toxic outputs. Humans can review and intervene when the model generates inappropriate or biased content, providing corrections or adjustments that help reduce the likelihood of toxicity and bias. This feedback loop helps refine and improve the model's outputs over time.

• Question 363:

  In which stage of the generative AI model lifecycle are tests performed to examine the model's accuracy?

  A. Deployment
  B. Data selection
  C. Fine-tuning
  D. Evaluation
  D. Evaluation

  ---

  The evaluation stage is when you run tests and benchmarks, such as accuracy, precision, and other performance metrics, to measure how well the generative AI model performs on hold-out or validation data before moving on to deployment.

• Question 364:

  A company is using a pre-trained large language model (LLM). The LLM must perform multiple tasks that require specific domain knowledge. The LLM does not have information about several technical topics in the domain. The company has unlabeled data that the company can use to fine-tune the model. Which fine-tuning method will meet these requirements?

  A. Full training
  B. Supervised fine-tuning
  C. Continued pre-training
  D. Retrieval Augmented Generation (RAG)
  C. Continued pre-training

  ---

  Continued pre-training uses the company's unlabeled domain data to further train a pre-trained LLM, expanding its knowledge in specific technical areas without requiring labeled datasets. This makes it the best fit for the scenario.

• Question 365:

  A company is using Amazon Bedrock for a generative AI solution. The solution must integrate a service with vector database storage and vector search capabilities. Which AWS service will meet these requirements?

  A. Amazon DynamoDB
  B. Amazon OpenSearch Service
  C. Amazon ElastiCache
  D. Amazon Redshift
  B. Amazon OpenSearch Service

  ---

  Amazon OpenSearch Service supports vector database storage and vector search capabilities, which are essential for integrating with generative AI solutions like those on Amazon Bedrock that require efficient similarity search over embeddings.

• Question 366:

  A financial company uses a generative AI model to assign credit limits to new customers. The company wants to make the decision-making process of the model more transparent to its customers. Which solution meets these requirements?

  A. Use a rule-based system instead of an ML model.
  B. Apply explainable AI techniques to show customers which factors influenced the model's decision.
  C. Develop an interactive UI for customers and provide clear technical explanations about the system.
  D. Increase the accuracy of the model to reduce the need for transparency.
  B. Apply explainable AI techniques to show customers which factors influenced the model's decision.

  ---

  Explainable AI techniques provide transparency by identifying and displaying the specific factors that influenced the generative AI model's credit limit decisions, making the decision-making process understandable to customers.

• Question 367:

  Which scenario describes a potential risk and limitation of prompt engineering in the context of a generative AI model?

  A. Prompt engineering does not ensure that the model always produces consistent and deterministic outputs, eliminating the need for validation.
  B. Prompt engineering could expose the model to vulnerabilities such as prompt injection attacks.
  C. Properly designed prompts reduce but do not eliminate the risk of data poisoning or model hijacking.
  D. Prompt engineering does not ensure that the model will consistently generate highly reliable outputs when working with real-world data.
  B. Prompt engineering could expose the model to vulnerabilities such as prompt injection attacks.

  ---

  A key risk in prompt engineering is prompt injection, where attackers manipulate input prompts to alter a model's behavior or produce unintended outputs. This vulnerability arises from the model's sensitivity to input structure and content, making it a critical limitation in secure prompt design.

• Question 368:

  A company is using a pre-trained large language model (LLM) to extract information from documents. The company noticed that a newer LLM from a different provider is available on Amazon Bedrock. The company wants to transition to the new LLM on Amazon Bedrock. What does the company need to do to transition to the new LLM?

  A. Create a new labeled dataset
  B. Perform feature engineering.
  C. Adjust the prompt template.
  D. Fine-tune the LLM.
  C. Adjust the prompt template.

• Question 369:

  A company is using Amazon Bedrock to build an assistant for its online store. The company wants to ensure that the assistant does not generate harmful responses based on hate speech, insults, sexual content, or violence.

  Which strategy will prevent harmful responses in Amazon Bedrock?

  A. Use Amazon SageMaker built-in algorithms to filter harmful content.
  B. Use Amazon Comprehend toxicity detection to identify harmful content.
  C. Configure Guardrails for Amazon Bedrock to filter harmful content.
  D. Train a custom model according to the company's responsible AI policies.
  C. Configure Guardrails for Amazon Bedrock to filter harmful content.

  ---

  Explanation

  Guardrails for Amazon Bedrock provides content filters that can detect and block harmful model inputs and outputs across categories including hate, insults, sexual content, and violence.

• Question 370:

  A company wants to document important details about an ML model, including intended use, training data considerations, and limitations, so stakeholders can review the information.

  Which tool best supports this need?

  A. SageMaker Model Cards
  B. Amazon CloudFront
  C. Amazon Polly
  D. AWS Secrets Manager
  A. SageMaker Model Cards

  ---

  Explanation

  SageMaker Model Cards is the correct answer because it is used to document critical information about ML models, including their purpose, risk considerations, and lifecycle details.

  Option A (Correct): "SageMaker Model Cards": This is correct because it supports model transparency and documentation.

  Option B: "Amazon CloudFront" is incorrect because it is a content delivery network.

  Option C: "Amazon Polly" is incorrect because it converts text to speech.

  Option D: "AWS Secrets Manager" is incorrect because it stores and manages secrets.