AIF-C01 Practice Test 49 — 15 Questions

Question 1

A startup is deploying a foundation model on Amazon SageMaker for real-time inference. They notice high latency (over 2 seconds per request). Which action is most likely to reduce latency?

Accepted Answer

Switch to a smaller, distilled version of the model.. Using a smaller, distilled version of the model directly reduces the computational complexity per inference request. Distillation compresses the model by training a smaller student network to mimic a larger teacher model, resulting in fewer parameters and faster forward passes. This is the most direct way to cut latency when the model size is the bottleneck, as it reduces the number of floating-point operations (FLOPs) required per request.

Answer

Enable auto-scaling on the SageMaker endpoint to handle more concurrent requests.

Answer

Deploy the model on a CPU-based instance instead of GPU.

Answer

Increase the batch size parameter in the inference request.

Question 2

Which TWO of the following are valid methods to reduce the risk of foundation models generating harmful or biased content?

Accepted Answer

Use a content filter. Option B (Use a content filter) is correct because content filters act as a post-processing guardrail that screens both prompts and model completions for harmful, violent, hateful, or otherwise policy-violating content, blocking or redacting it before it reaches users. Option C (Apply prompt engineering to guide output) is correct because carefully crafted system prompts, few-shot examples, and instructions can steer the foundation model toward safe, neutral, and on-topic responses, reducing the likelihood of biased or harmful generations. Option A (Use a smaller model) is not a valid mitigation because model size does not determine safety or bias; smaller models can still produce harmful or biased content. Option D (Fine-tune the model on a biased dataset) would actually increase the risk by reinforcing biased patterns in the model's outputs. Option E (Disable all logging) does not reduce harmful content generation and instead removes the audit trail needed to detect, monitor, and remediate such issues.

Answer

Use a smaller model

Answer

Fine-tune the model on a biased dataset

Answer

Disable all logging

Question 3

Which TWO factors are most important when selecting a foundation model in Amazon Bedrock for a text summarization task with strict latency requirements?

Accepted Answer

Average response latency per request.. Option A is correct because with strict latency requirements the deciding metric is the model's average response latency per request, which directly determines whether the summarization workload can meet its service-level objective. Option D is correct because output quality and token efficiency determine whether the summary is usable and how many tokens must be generated, and fewer generated tokens translate directly into lower end-to-end latency. Option B is not decisive on its own: parameter count correlates loosely with latency but does not guarantee it, since architecture, hardware, and serving optimizations also matter. Option C, the maximum input token limit, matters for handling long documents but does not address the strict latency constraint. Option E, fine-tuning availability, supports domain adaptation and accuracy rather than latency, so it is not one of the two most important factors here.

Answer

Model size in billions of parameters.

Answer

Maximum input token limit.

Answer

Availability of fine-tuning capability for domain adaptation.

Question 4

A data scientist is evaluating whether to use fine-tuning or Retrieval-Augmented Generation (RAG) for a legal document analysis application. Which TWO statements correctly describe when to use each approach?

Accepted Answer

Use fine-tuning to teach the model a specific output format or style. Option B is correct because fine-tuning adjusts the model's weights on labeled examples, which is the right technique for teaching a consistent output format, tone, or style (e.g., always returning structured legal citations in a fixed schema). Option E is correct because RAG retrieves documents from an external knowledge base at inference time, so when the underlying corpus changes frequently, you simply update the index/vector store rather than retraining the model. Option A is wrong because RAG reduces hallucinations by grounding responses in retrieved external data, not by working 'without any external data.' Option C is wrong because real-time information access is achieved through retrieval (RAG) or tool calls, not fine-tuning, which bakes in static weights at training time. Option D is wrong because fine-tuning is unreliable for injecting new factual knowledge; RAG is the preferred approach for adding or updating facts.

Answer

Use RAG to reduce model hallucinations without any external data

Answer

Use fine-tuning to enable the model to access real-time information

Answer

Use fine-tuning to inject new factual knowledge into the model

Question 5

A team is designing a RAG system on Amazon Bedrock. They need to chunk a large set of PDF documents into smaller pieces for embedding. Which THREE considerations should guide their chunking strategy? (Choose three.)

Accepted Answer

Chunk size should be small enough to fit multiple chunks within the model's context window after including the query. Option B is correct because chunk size must be small enough that several retrieved chunks plus the user query and prompt fit within the foundation model's context window; otherwise retrieved content will be truncated or the request will fail. Option C is correct because each chunk is passed to the embedding model, so it must not exceed that model's maximum input token limit (for example, Amazon Titan Text Embeddings V2 supports up to 8,192 tokens), or the chunk will be rejected or silently truncated. Option E is correct because splitting on natural semantic boundaries such as paragraphs, headings, and sections preserves coherent meaning, which produces more accurate embeddings and better retrieval relevance than arbitrary fixed-length cuts. Option A is not required: uniform chunk length is not a performance requirement, and forcing equal sizes often breaks semantic boundaries. Option D is not correct: overlap is commonly recommended because it preserves context across chunk boundaries and prevents information loss at split points, even though it adds some redundancy.

Answer

All chunks must be exactly the same length for optimal performance

Answer

Overlapping chunks should be avoided to reduce redundancy

Question 6

A team has built a regression model to predict house prices. The RMSE is 50,000 on the test set. Which action is most appropriate to improve model performance?

Accepted Answer

Add more relevant features. Adding more relevant features can provide the model with additional predictive signals, potentially reducing bias and lowering RMSE if the new features have genuine correlation with house prices. Since RMSE is already 50,000, the model may be underfitting due to insufficient input variables, and enriching the feature set is a direct way to capture more variance in the target variable.

Answer

Remove outliers from training data

Answer

Apply feature scaling

Answer

Use a different evaluation metric

Question 7

What is the primary role of the self-attention mechanism in the Transformer architecture?

Accepted Answer

To allow each token to attend to all other tokens in the sequence, capturing long-range dependencies. The self-attention mechanism allows each token in the input sequence to directly attend to every other token, computing a weighted sum of all token representations. This enables the model to capture long-range dependencies and contextual relationships regardless of distance, which is the fundamental innovation of the Transformer architecture over recurrent or convolutional models.

Answer

To process tokens in parallel by alternating attention and feed-forward layers

Answer

To reduce the vocabulary size by mapping tokens to embeddings

Answer

To generate the next token one at a time in an autoregressive manner

Question 8

An AI practitioner is evaluating a text generation model and notices that the model sometimes produces plausible-sounding but factually incorrect statements. What is this phenomenon called?

Accepted Answer

Hallucination. Hallucination in LLMs refers to generating content that is not grounded in the training data or provided context. It is a known challenge for generative models.

Answer

Catastrophic forgetting

Answer

Bias amplification

Answer

Overfitting

Question 9

Which TWO of the following are key components of a responsible AI governance framework?

Accepted Answer

Develop and enforce AI ethics policies and standards. A responsible AI governance framework must include the development and enforcement of AI ethics policies and standards to ensure alignment with societal values, fairness, and accountability. These policies guide the design, deployment, and monitoring of AI systems, embedding ethical principles such as transparency, privacy, and non-discrimination into the AI lifecycle. Without such policies, organizations risk deploying AI that violates ethical norms or regulatory expectations.

Answer

Focus solely on compliance with legal regulations

Answer

Minimize human involvement in AI lifecycle decisions

Answer

Deploy AI models as black boxes to avoid scrutiny

Question 10

A company is building a document summarization application using Amazon Bedrock. They want to prototype the application quickly by testing different models and prompts interactively. Which AWS service or feature should they use?

Accepted Answer

Amazon Bedrock Playground. Amazon Bedrock Playground is the correct choice because it provides an interactive, no-code environment within the AWS Management Console specifically designed for experimenting with different foundation models and prompts. It allows you to quickly test and compare model outputs, adjust parameters, and iterate on prompts without writing any code or setting up infrastructure, making it ideal for rapid prototyping of a document summarization application.

Answer

Amazon SageMaker Studio

Answer

Amazon Bedrock Studio

Answer

AWS Cloud9

Question 11

A company uses an AI system to automate loan approvals. The model uses demographic features and achieves high accuracy, but the company wants to ensure compliance with responsible AI guidelines. Which practice best balances performance and fairness?

Accepted Answer

Remove sensitive attributes and monitor for proxy bias. Removing sensitive attributes (e.g., race, gender) from the training data directly addresses fairness by preventing the model from explicitly using these features. However, simply removing them is insufficient; monitoring for proxy bias (e.g., zip code or income correlating with race) is critical to ensure the model does not inadvertently learn discriminatory patterns through correlated features. This approach balances performance by retaining predictive power from non-sensitive features while actively auditing for fairness violations.

Answer

Use demographic features but with minimal monitoring

Answer

Use a complex black-box model and rely on post-hoc explanations

Answer

Optimize the model solely for accuracy on historical data

Question 12

A security engineer is configuring logging for Amazon Bedrock model invocations. They need to capture both the input and output of all API calls for compliance audits. Which set of steps should they take?

Accepted Answer

Enable Bedrock model invocation logging and specify an S3 bucket and optionally CloudWatch Logs as the destination. Bedrock model invocation logging captures inputs and outputs to S3 and/or CloudWatch Logs. CloudTrail records API calls but not the model inputs/outputs.

Answer

Enable CloudTrail for the Bedrock API and configure S3 event notifications

Answer

Use VPC Flow Logs to capture network traffic and reconstruct model inputs from packet data

Answer

Enable AWS Config rules for Bedrock and stream logs to Amazon Kinesis

Question 13

An IAM policy allows creation of SageMaker training jobs only if they use a specific VPC security group. A user tries to create a training job without specifying that security group. What will happen?

Accepted Answer

The request will be denied with an AccessDenied error. IAM policies are evaluated before any AWS API action is executed. If the policy includes a condition that requires a specific VPC security group for SageMaker training jobs, and the user's request does not include that security group, the condition is not met, resulting in an explicit deny (AccessDenied error). AWS IAM denies the request by default if the condition in a policy is not satisfied, regardless of whether the condition is marked as optional in the API.

Answer

The request will succeed but SageMaker will ignore the condition

Answer

The request will succeed because the condition is optional

Answer

The request will be denied because the training job resource ARN is invalid

Question 14

Which TWO of the following are types of unsupervised learning? (Select TWO.)

Accepted Answer

Dimensionality reduction. Dimensionality reduction (B) is a type of unsupervised learning because it seeks to compress or transform unlabeled data into a lower-dimensional representation (e.g., via PCA or t-SNE) without any target labels. Clustering (C) is also unsupervised, as algorithms like k-means or DBSCAN group unlabeled data points by similarity without predefined classes. Classification (A) and regression (E) are supervised learning tasks, since they require labeled training data with discrete or continuous target values, respectively. Reinforcement learning (D) is a separate paradigm in which an agent learns from reward signals via interaction with an environment, not from unlabeled data in the unsupervised sense.

Answer

Classification

Answer

Reinforcement learning

Answer

Regression

Question 15

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

Accepted Answer

Self-attention mechanism. The self-attention mechanism (option D) is the core component of the Transformer that computes attention scores between every pair of tokens in the input sequence, allowing the model to dynamically weigh the importance of each token when generating an output token. This enables the model to capture long-range dependencies and contextual relationships without the sequential constraints of RNNs.

Answer

Layer normalization

Answer

Feed-forward network

Answer

Positional encoding