AIF-C01 Generative AI and Foundation Models — 50 Questions

Question 1

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. RAG (Retrieval-Augmented Generation) allows the LLM to retrieve relevant document sections at inference time, so knowledge stays current without retraining. The other options either require expensive retraining for each update or lack document grounding.

Answer

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

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

Question 2

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 3

A data scientist is using Amazon Bedrock to generate product descriptions. They notice the output is often repetitive and lacks creativity. Which combination of parameter adjustments is MOST likely to produce more diverse and less repetitive output?

Accepted Answer

Increase temperature and increase top-p. Increasing temperature raises the probability of sampling lower-probability tokens, which increases randomness and diversity. Increasing top-p (nucleus sampling) expands the set of tokens considered for sampling, further reducing repetitiveness. Together, these adjustments encourage the model to explore a wider range of possible continuations, producing more creative and less repetitive output.

Answer

Decrease temperature and decrease top-p

Answer

Decrease temperature and increase top-p

Answer

Increase temperature and decrease top-p

Question 4

A developer is building a multimodal application that needs to process both text and images to generate a description. Which Amazon Bedrock model provider offers a multimodal foundation model capable of accepting images and text as input?

Accepted Answer

Anthropic. Anthropic's Claude 3 and Claude 3.5 models on Amazon Bedrock are multimodal, accepting both text and image inputs to generate descriptions. This capability is built into the model's architecture, which processes images alongside text prompts natively, unlike other providers that focus solely on text or image generation.

Answer

Mistral AI

Answer

Cohere

Answer

Stability AI

Question 5

A company is using Amazon Bedrock to generate embeddings for a semantic search application. They want to ensure that semantically similar phrases (e.g., "car" and "vehicle") produce similar vector representations. Which type of model should they use?

Accepted Answer

An embedding model like Amazon Titan Embeddings. Amazon Titan Embeddings is a text embedding model specifically designed to convert textual input into dense vector representations that capture semantic meaning. By mapping semantically similar phrases like 'car' and 'vehicle' to nearby points in the embedding space, it enables accurate similarity comparisons for semantic search applications.

Answer

An image generation model like Stable Diffusion

Answer

A text generation model like Anthropic Claude

Answer

A multimodal model that supports both text and image input

Question 6

A team is fine-tuning an Amazon Titan Text model on a small dataset of legal documents. After fine-tuning, the model produces outputs that are factually incorrect and sometimes contradicts the training data. What is the most likely cause?

Accepted Answer

The model was fine-tuned for too many epochs on a small dataset, leading to overfitting. Option C is correct because fine-tuning a large language model like Amazon Titan Text on a very small dataset for too many epochs causes overfitting. The model memorizes the limited training examples instead of learning generalizable patterns, leading to outputs that are factually incorrect or contradictory when faced with inputs that differ slightly from the training data.

Answer

The batch size was too large, reducing gradient noise and causing mode collapse

Answer

The learning rate was set too high, causing the model to diverge

Answer

The base model was not designed for legal domain tasks

Question 7

Which of the following is a key advantage of using a pre-trained foundation model over training a model from scratch?

Accepted Answer

Reduces the amount of labeled data and compute resources needed. Pre-trained foundation models, such as those based on transformer architectures, have already learned general language patterns from vast, diverse datasets during their initial training. This transfer learning approach drastically reduces the need for large amounts of labeled data and extensive compute resources when adapting the model to a specific downstream task, as only a relatively small fine-tuning step is required.

Answer

Eliminates the need for any fine-tuning

Answer

Guarantees perfect accuracy on domain-specific tasks

Answer

Allows the model to work offline without any cloud infrastructure

Question 8

A developer is using the Amazon Bedrock Converse API to build a chat application. They want the model to maintain context across multiple turns. Which parameter should they set to ensure the conversation history is included?

Accepted Answer

messages. The Converse API's `messages` parameter accepts an array of message objects, each with a `role` (user or assistant) and `content`. By including previous turns in this array, the model receives the full conversation history as context, enabling coherent multi-turn dialogue. Without it, each API call is stateless and the model has no memory of prior exchanges.

Answer

inferenceConfig

Answer

additionalModelRequestFields

Answer

system

Question 9

A company needs to transcribe and summarize customer support calls in real time. They want to use a large language model (LLM) for summarization but the audio input is streaming. Which approach should they use?

Accepted Answer

Transcribe the audio using Amazon Transcribe and then pass the text to an LLM for summarization. Option C is correct because Amazon Transcribe is designed to handle streaming audio input and convert it to text in real time, which can then be passed to an LLM for summarization. This two-step pipeline leverages the strengths of each service: Transcribe for accurate speech-to-text conversion and an LLM for natural language summarization, without requiring a multimodal model that directly processes audio.

Answer

Use a multimodal LLM that directly accepts audio input

Answer

Use an image generation model to create a visual summary of the audio

Answer

Use an embedding model to convert audio directly into searchable vectors

Question 10

Which prompt engineering technique involves providing the model with a few examples of desired input-output pairs before asking it to complete a new instance?

Accepted Answer

Few-shot prompting. Few-shot prompting (Option D) is the correct technique because it explicitly involves providing the model with a few examples of desired input-output pairs before asking it to complete a new instance. This helps the model understand the pattern, format, or task from the examples, improving performance on tasks like classification, translation, or formatting without requiring fine-tuning.

Answer

Zero-shot prompting

Answer

Chain-of-thought prompting

Answer

System prompting

Question 11

A startup wants to generate high-quality images from text descriptions using Amazon Bedrock. They need to create realistic images of products for an e-commerce catalog. Which model provider should they choose?

Accepted Answer

Stability AI. Stability AI is the correct choice because it specializes in image generation models, such as Stable Diffusion, which are designed to produce high-quality, photorealistic images from text prompts. Amazon Bedrock offers Stability AI's models for text-to-image tasks, making it ideal for generating realistic product images for an e-commerce catalog.

Answer

Anthropic

Answer

Mistral AI

Answer

Cohere

Question 12

A company is using Amazon Bedrock to build an application that requires very low latency responses (under 100ms). They are currently using a large model but need faster inference. Which model selection strategy is MOST appropriate?

Accepted Answer

Switch to a smaller model that can meet the latency requirement while still providing acceptable quality. Option C is correct because smaller models have fewer parameters, which reduces the computational cost per inference, directly lowering latency. For latency-sensitive applications requiring under 100ms responses, a smaller model can often provide acceptable quality while meeting the strict timing requirement, whereas a larger model would introduce higher inference latency due to increased matrix operations and memory bandwidth demands.

Answer

Increase the temperature parameter to speed up generation

Answer

Use a larger model with more parameters for better accuracy

Answer

Use batch processing instead of real-time streaming

Question 13

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 a curated dataset, which can teach it to follow a specific output format, tone, or style (e.g., legal citations or structured summaries). This is a core use case for fine-tuning when the desired behavior is stylistic or structural, not when the model needs new facts or real-time data.

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 14

A company is deploying a chatbot using Amazon Bedrock and wants to ensure that the model does not generate offensive or inappropriate content. Which THREE measures can they apply?

Accepted Answer

Use a system prompt to define ethical guidelines and constraints. A system prompt is a directive provided to the foundation model at inference time that sets the context, tone, and behavioral constraints for the model's responses. By explicitly defining ethical guidelines and content restrictions in the system prompt, the company can instruct the model to avoid generating offensive or inappropriate content without modifying the underlying model weights. This is a lightweight, non-invasive control that works with any Bedrock model and can be iterated quickly.

Answer

Increase the temperature to make outputs more creative and less likely to repeat offensive phrases

Answer

Fine-tune the model on a dataset of safe conversations only

Question 15

A developer wants to use Amazon Bedrock to build a text summarization application. Which TWO of the following are required steps?

Accepted Answer

Request access to a foundation model in Amazon Bedrock. Amazon Bedrock provides access to pre-trained foundation models via API, so the developer does not need to host models on SageMaker or train from scratch. The two required steps are requesting access to a foundation model (D) and invoking it with a prompt via the Bedrock API or SDK (E), which directly enables text summarization.

Answer

Create an Amazon SageMaker endpoint for hosting the model

Answer

Train a custom model from scratch on summarization data

Answer

Set up an Amazon EMR cluster to preprocess the text data

Question 16

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. RAG (Retrieval-Augmented Generation) allows the LLM to retrieve relevant document sections at inference time, so knowledge stays current without retraining. The other options either require expensive retraining for each update or lack document grounding.

Answer

Fine-tune a base LLM on the policy documents monthly

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

Question 17

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

Question 18

A data scientist is using Amazon Bedrock to generate product descriptions. They discover that the model frequently repeats phrases and produces overly deterministic outputs. Which parameter adjustment would MOST likely introduce more diversity?

Accepted Answer

Increase the temperature from 0.7 to 1.2. Increasing the temperature from 0.7 to 1.2 raises the randomness of token selection by flattening the probability distribution, which makes lower-probability tokens more likely to be chosen. This directly counters the overly deterministic and repetitive outputs by introducing more diversity into the generated text.

Answer

Increase the top_p value from 0.9 to 1.0

Answer

Decrease the top_k value from 50 to 10

Answer

Decrease the temperature from 0.7 to 0.2

Question 19

A company uses Amazon Titan Text Express to summarize customer support tickets. The model often misses key details when the ticket exceeds 4,000 tokens. The team needs to process tickets up to 8,000 tokens without losing important information. Which strategy is MOST effective?

Accepted Answer

Split the ticket into chunks of 3,500 tokens, summarize each chunk, then summarize the summaries. Option C is correct because it uses a hierarchical summarization approach: splitting the 8,000-token ticket into chunks within the model's 4,000-token limit, summarizing each chunk independently, then combining those summaries into a final summary. This preserves key details across the entire ticket without exceeding the model's context window, which is a common technique for handling long documents with fixed-context models.

Answer

Switch to a model with a 100,000 token context window like Claude 3

Answer

Truncate the ticket to 4,000 tokens before sending to the model

Answer

Use Amazon Bedrock's Converse API with the entire 8,000-token ticket

Question 20

A developer is building an image generation application using Stability AI's Stable Diffusion model on Amazon Bedrock. The application needs to generate high-resolution images (1024x1024) with consistent style across multiple prompts. Which approach should they use?

Accepted Answer

Set the same seed value for all invocations. Setting the same seed value ensures that the random noise generator used by Stable Diffusion produces identical initial noise for each invocation. This guarantees consistent style and composition across multiple prompts, as the model's output is deterministic given the same seed, prompt, and parameters.

Answer

Use a higher number of inference steps

Answer

Increase the CFG scale to 20

Answer

Invoke the model through the Converse API with a system prompt