AIF-C01 Practice Test 33 — 15 Questions

Question 1

Which component of the Transformer architecture allows the model to weigh the importance of different words in the input sequence when generating output?

Accepted Answer

Self-attention mechanism. The self-attention mechanism computes attention scores between every pair of words in the input sequence, allowing the model to dynamically weigh the relevance of each word when producing an output. This enables the Transformer to capture long-range dependencies and contextual relationships, which is the core innovation over earlier sequence models like RNNs.

Answer

Layer normalization

Answer

Positional encoding

Answer

Feed-forward neural network

Question 2

A team trained a binary classifier to detect fraudulent transactions. The dataset is highly imbalanced (1% fraud). The model achieves 99% accuracy but only catches 5% of actual fraud cases. Which metric should the team primarily optimize?

Accepted Answer

Recall. In fraud detection with 1% fraud rate, 99% accuracy is misleading because the model can achieve it by simply predicting 'not fraud' for all transactions. The model catches only 5% of actual fraud cases, meaning recall (true positive rate) is critically low at 5%. Optimizing recall directly increases the proportion of actual fraud cases correctly identified, which is the primary business requirement in fraud detection where missing a fraud (false negative) is far more costly than a false alarm.

Answer

F1 score

Answer

Accuracy

Answer

Precision

Question 3

Refer to the exhibit. A developer has attached this IAM policy to their user. When trying to invoke the Anthropic Claude v2 model using the Bedrock runtime, they receive an AccessDeniedException. Which change to the policy would resolve the issue?

Accepted Answer

Change the Resource to arn:aws:bedrock:us-east-1::foundation-model/anthropic.claude-v2. The IAM policy's Resource element must specify the exact ARN of the foundation model being invoked. The original policy likely used a wildcard or incorrect ARN pattern. The correct ARN for Anthropic Claude v2 in us-east-1 is arn:aws:bedrock:us-east-1::foundation-model/anthropic.claude-v2, which grants permission to invoke that specific model via the Bedrock runtime.

Answer

Add the bedrock:InvokeModelWithResponseStream action

Answer

Change the Action to bedrock:ListFoundationModels

Answer

Remove the Resource element and set Effect to Deny

Question 4

A healthcare organization wants to use generative AI to draft clinical notes from patient-physician conversations. They must comply with HIPAA and minimize false medical information. Which approach should they take?

Accepted Answer

Use Amazon Bedrock with a HIPAA-eligible foundation model and connect it to a medical knowledge base via RAG.. It combines a HIPAA-eligible foundation model via Amazon Bedrock with Retrieval-Augmented Generation (RAG) to ground responses in a curated medical knowledge base. This approach ensures compliance with HIPAA by using a service that supports Protected Health Information (PHI) processing under a Business Associate Agreement (BAA), while RAG reduces hallucination risk by retrieving factual clinical data rather than relying solely on the model's training.

Answer

Use Amazon SageMaker JumpStart with a publicly available clinical model and no additional modifications.

Answer

Use a generic open-source LLM hosted on Amazon EC2 with manual prompt engineering.

Answer

Use Amazon Bedrock with a large foundation model and a high temperature setting for creativity.

Question 5

A company wants to automatically summarize customer support tickets into a short paragraph. Which AWS service is MOST appropriate for this task?

Accepted Answer

Amazon Bedrock. Amazon Bedrock provides access to foundation models (FMs) from providers like Anthropic and AI21 that excel at natural language generation tasks, including summarization. By invoking a model such as Claude or Jurassic-2 via Bedrock's API, you can pass the customer support ticket text and receive a concise paragraph summary. This makes Bedrock the most appropriate service for generative summarization.

Answer

Amazon Rekognition

Answer

Amazon Polly

Answer

Amazon Comprehend

Question 6

Refer to the exhibit. A data scientist is training a model in SageMaker using a KMS-encrypted dataset. The training job fails with the error shown. Which action should be taken to resolve this issue?

Accepted Answer

Add the SageMaker execution role to the KMS key policy with the kms:Decrypt permission.. The error indicates that the SageMaker training job cannot access the KMS-encrypted dataset. Since the S3 bucket uses server-side encryption with a customer-managed KMS key (SSE-KMS), the SageMaker execution role must be explicitly granted kms:Decrypt permission in the KMS key policy. Option A correctly adds this permission to the key policy, which is required for SageMaker to decrypt the data during training.

Answer

Create a new KMS key and update the bucket policy to use the new key.

Answer

Attach an IAM policy to the SageMaker execution role that allows kms:Decrypt on the key.

Answer

Disable server-side encryption on the S3 bucket and use client-side encryption.

Question 7

A developer is using Amazon Bedrock to build a chatbot that answers questions about a large internal knowledge base. The knowledge base contains documents with varying lengths, some exceeding 10,000 tokens. The chatbot must provide accurate answers and handle queries about multiple topics. Which THREE strategies should the developer implement? (Select THREE)

Accepted Answer

Chunk documents into smaller segments with overlap, and index them in a vector database. Option A is correct because chunking long documents into smaller segments with overlap and indexing them in a vector database is the standard RAG approach for handling documents that exceed the model's context window while preserving semantic continuity across chunk boundaries. Option B is correct because setting maxTokens appropriately ensures the model has enough room to generate complete, accurate answers without truncation or exceeding the model's context window when combined with retrieved context. Option E is correct because Amazon Bedrock Knowledge Bases can use Amazon OpenSearch Serverless with the vector engine as the vector store to persist and retrieve document chunk embeddings for semantic search. Option C is not selected because simply using a larger context window model does not by itself solve retrieval accuracy for a large knowledge base and is not one of the required strategies here. Option D is not selected because lowering temperature to 0 affects randomness, not retrieval accuracy or handling of long documents, and deterministic output is not a stated requirement.

Answer

Use a model with a larger context window, such as Anthropic Claude 2.1 (200K tokens)

Answer

Lower the temperature to 0 to ensure deterministic responses

Question 8

A company wants to build a customer service chatbot that answers questions about their internal policy documents. The documents are updated monthly, and the team cannot afford to retrain a model each time. Which approach is MOST appropriate?

Accepted Answer

Use Retrieval-Augmented Generation (RAG) with the policy documents indexed in a vector store. Retrieval-Augmented Generation (RAG) is the most appropriate approach because it retrieves relevant policy document chunks from a vector store at inference time, allowing the chatbot to answer questions without retraining the model. This addresses the requirement for monthly updates without the cost and time of retraining, as only the vector store needs to be re-indexed with the new documents.

Answer

Train a custom model from scratch on the policy documents each month

Answer

Use a larger foundation model with a longer context window and paste all documents into each prompt

Answer

Fine-tune a base LLM on the policy documents monthly

Question 9

A generative AI application occasionally produces factually incorrect responses. The team has already tried prompt engineering and increasing the temperature parameter. Which next step is MOST effective to improve factual accuracy?

Accepted Answer

Implement a Retrieval Augmented Generation (RAG) pipeline. Retrieval Augmented Generation (RAG) is the most effective next step because it grounds the model's responses in a verified external knowledge base, directly addressing factual inaccuracies without requiring retraining. Unlike prompt engineering or temperature adjustments, RAG provides real-time access to authoritative documents, reducing hallucinations by constraining the model's output to retrieved evidence.

Answer

Use a larger foundation model

Answer

Fine-tune the model on company data

Answer

Reduce the temperature to 0

Question 10

A machine learning engineer normalizes numerical features to have mean 0 and standard deviation 1 before training a linear regression model. What is the PRIMARY benefit of this normalization?

Accepted Answer

It ensures all features are on the same scale so that no feature dominates the model training. Normalizing numerical features to have mean 0 and standard deviation 1 (z-score normalization) ensures all features are on the same scale. In linear regression, features with larger magnitudes can disproportionately influence the gradient descent updates and the learned coefficients, leading to unstable training or suboptimal convergence. By standardizing, each feature contributes equally to the model, improving both training stability and interpretability of the coefficients.

Answer

It eliminates the need for a bias term

Answer

It converts categorical variables into numerical values

Answer

It reduces the number of features

Question 11

A company wants to build a customer service chatbot that answers questions about their internal policy documents. The documents are updated monthly, and the team cannot afford to retrain a model each time. Which approach is MOST appropriate?

Accepted Answer

Use Retrieval-Augmented Generation (RAG) with the policy documents indexed in a vector store. Retrieval-Augmented Generation (RAG) is the most appropriate approach because it allows the chatbot to answer questions based on the latest policy documents without retraining the model. By indexing the documents in a vector store and retrieving relevant chunks at inference time, the system can incorporate monthly updates simply by re-indexing the new documents, keeping the model static and avoiding costly retraining.

Answer

Fine-tune a base LLM on the policy documents monthly

Answer

Use a larger foundation model with a longer context window and paste all documents into each prompt

Answer

Train a custom model from scratch on the policy documents each month

Question 12

A data scientist is building a model to predict customer churn. They have 10,000 samples with 20 features. After training a decision tree, they observe high variance. Which ensemble method would best reduce variance without significantly increasing bias?

Accepted Answer

Random forest. A random forest reduces variance by training multiple decision trees on bootstrapped subsets of the data and averaging their predictions. This ensemble approach leverages the low bias of individual trees while smoothing out their high variance through aggregation, directly addressing the overfitting issue without substantially increasing bias.

Answer

k-nearest neighbors

Answer

Gradient boosting

Answer

Logistic regression

Question 13

A company wants to build a customer service chatbot that answers questions about their internal policy documents. The documents are updated monthly, and the team cannot afford to retrain a model each time. Which approach is MOST appropriate?

Accepted Answer

Use Retrieval-Augmented Generation (RAG) with the policy documents indexed in a vector store. Retrieval-Augmented Generation (RAG) is the most appropriate approach because it allows the chatbot to answer questions by retrieving relevant chunks from the policy documents stored in a vector store at inference time, without requiring any model retraining. When documents are updated monthly, only the vector store needs to be re-indexed, while the underlying LLM remains unchanged, making it cost-effective and scalable for frequently changing knowledge bases.

Answer

Use a larger foundation model with a longer context window and paste all documents into each prompt

Answer

Fine-tune a base LLM on the policy documents monthly

Answer

Train a custom model from scratch on the policy documents each month

Question 14

A machine learning team is using prompt engineering to guide a large language model on Amazon Bedrock. They want the model to follow a specific reasoning process step-by-step. Which THREE prompt engineering techniques are most relevant? (Select THREE.)

Accepted Answer

Few-shot prompting with examples of step-by-step reasoning. Option B (few-shot prompting with examples of step-by-step reasoning) is correct because providing worked examples that demonstrate each reasoning step conditions the model to imitate that structured reasoning pattern for new inputs. Option C (system prompts that describe the desired reasoning process) is correct because a system prompt sets persistent instructions and context for the model, so explicitly describing the required step-by-step reasoning process steers all subsequent responses accordingly. Option D (chain-of-thought prompting) is correct because CoT explicitly elicits intermediate reasoning steps before the final answer, which is exactly the step-by-step reasoning process the team wants. Option A (zero-shot prompting) does not belong because it only supplies a task instruction with no reasoning examples or process guidance, so it does not specifically enforce a step-by-step reasoning procedure. Option E (adjusting temperature to a high value) does not belong because temperature controls randomness/creativity of sampling, not the structure or presence of a reasoning process, and a high value would typically make outputs less deterministic rather than more step-by-step.

Answer

Zero-shot prompting

Answer

Adjusting the temperature to a high value

Question 15

A developer is using the Amazon Bedrock Converse API to build a conversational agent. The agent needs to maintain context across multiple turns of dialogue. Which parameter should be used to provide the conversation history?

Accepted Answer

The 'messages' parameter with an array of previous messages. The Amazon Bedrock Converse API uses the 'messages' parameter to pass an array of previous message objects, each with a 'role' (user or assistant) and 'content'. This array represents the full conversation history, allowing the model to maintain context across multiple turns of dialogue.

Answer

The 'additionalModelRequestFields' parameter with a custom field

Answer

The 'system' parameter with a concatenated history

Answer

The 'inferenceConfig' parameter with maxTokens set to a high value