AWS Certified AI Practitioner (AIF-C01) Mock Test

This is a classification problem because the goal is to assign a predefined category (spam or not spam) to each email.

Latency is the time delay between sending a request to the model and receiving a response. Low latency is critical for real-time applications.

Vector databases are optimized for storing and querying embeddings (vector representations of data). In RAG, they store embeddings of the external knowledge source, allowing for efficient similarity searches to find relevant context.

Generative AI models, especially LLMs, can sometimes produce 'hallucinations,' which are confident but incorrect or nonsensical outputs.

Tokens are basic units of text (like words, subwords, or characters) that LLMs process and generate.

Prompt engineering is the process of designing and refining input prompts to guide a generative AI model to produce desired outputs.

Supervised learning algorithms require labeled data, where each data point is tagged with a correct output or target variable.

AWS is responsible for the security 'of' the cloud, including the underlying infrastructure, hardware, software, networking, and facilities that run AWS Cloud services. Customers are responsible for security 'in' the cloud, such as data encryption, IAM configurations, and network traffic protection.

Generative AI, particularly LLMs, excels at understanding natural language and generating human-like responses, making them ideal for building chatbots and virtual assistants.

While the original research paper's citation count might indicate influence, factors like cost, latency, model size, and performance on relevant benchmarks are more direct and practical considerations for selecting a model for a business application.

The ML development lifecycle typically begins with defining the problem you want to solve and understanding the objectives.

AWS CloudTrail records AWS API calls for your account and delivers log files to an Amazon S3 bucket, enabling security analysis, resource change tracking, and compliance auditing.

Robustness means the AI system can maintain its level of performance even when faced with adversarial attacks or unexpected, noisy, or out-of-distribution inputs.

In traditional programming, humans write explicit rules for the computer to follow. In ML, the system learns patterns from data to make predictions or decisions without being explicitly programmed for each case.

The evaluation stage involves testing the foundation model (either pre-trained or fine-tuned) against various benchmarks and metrics to understand its capabilities and limitations.

Amazon SageMaker is a comprehensive ML platform that supports the entire ML workflow, including capabilities for working with foundation models.

AWS Identity and Access Management (IAM) enables you to securely control access to AWS services and resources for your users and applications.

While Generative AI can assist in analyzing data that might lead to fraud detection, fraud detection itself often relies more on discriminative models (classification) to identify anomalous patterns from existing data.

Data minimization involves collecting and retaining only the necessary personal data for a specific, legitimate purpose, reducing privacy risks.

Generative AI models trained on copyrighted material may inadvertently produce outputs that infringe on existing intellectual property rights, leading to legal challenges.

AI refers to the simulation of human intelligence in machines that are programmed to think like humans and mimic their actions. This can include learning, problem-solving, and decision-making.

Unstructured data does not have a predefined format or organization, making it more difficult to collect, process, and analyze. Examples include text documents, images, and videos.

This parameter directly controls the maximum number of tokens the model will process as input or generate as output, impacting resource usage and the scope of the interaction.

Deep Learning is a specialized subfield of Machine Learning that uses neural networks with many layers (deep neural networks) to analyze various factors of data.

Deployment is the process of making a trained and evaluated foundation model available for use in applications, often via an API endpoint.

Multimodal AI models are designed to process and understand information from multiple types of data, such as text, images, audio, and video, simultaneously.

Model size and complexity directly impact inference speed (latency). Smaller, optimized models are generally preferred for low-latency applications.

Explainability (or interpretability) is the ability to explain how an AI model arrived at a particular decision or prediction in terms understandable to humans.

Content creation, such as generating text, images, audio, and video, is a primary use case for Generative AI.

While pre-deployment mitigation is best, post-deployment strategies can include continuous monitoring for bias, collecting feedback, and potentially re-training or fine-tuning the model with more diverse and debiased data, or implementing fairness-aware post-processing techniques.

Reinforcement Learning from Human Feedback (RLHF) is a technique used to fine-tune language models by incorporating human feedback into the training process, helping the model generate outputs that are more helpful, harmless, and honest.

Ensuring the training dataset is diverse and representative of the population the AI will affect is a key strategy. This can involve augmenting underrepresented groups or carefully sampling data.

Pre-training foundation models typically relies on self-supervised learning on massive amounts of unlabeled text and/or code, where the model learns to predict parts of the input data itself.

The Transformer architecture, introduced in the paper 'Attention Is All You Need,' is highly effective for sequence-to-sequence tasks and is the basis for many modern LLMs due to its use of attention mechanisms.

Data preparation is the stage where data is cleaned, transformed, and split into appropriate sets for training and evaluating the model.

Transformer models process entire input sequences simultaneously using attention mechanisms, rather than sequentially like RNNs, allowing for better parallelization and capturing long-range dependencies.

A foundation model is a large AI model pre-trained on a vast quantity of broad data, designed to be adapted (e.g., fine-tuned) to a wide range of downstream tasks.

Amazon SageMaker is a fully managed service that provides developers and data scientists with the ability to build, train, and deploy machine learning models quickly.

Temperature is a parameter that controls the randomness of the model's output. Higher temperatures lead to more creative and diverse outputs, while lower temperatures produce more focused and deterministic outputs.

Amazon SageMaker Clarify helps improve machine learning models by detecting potential bias and helping explain how these models make predictions.

Transparency involves providing clear information about how an AI system works, its capabilities, limitations, and the data it uses, which helps build trust and allows for accountability.

Sentiment analysis, which involves determining the emotional tone behind a series of words, is a common application of NLP.

The ability of generative AI to create convincing deepfakes raises significant concerns about misinformation, disinformation, and the potential for malicious use like impersonation or defamation.

RAG enhances foundation models by providing them with access to external, up-to-date knowledge sources. The model retrieves relevant information from these sources to generate more accurate and contextually appropriate responses.

Amazon Bedrock is a fully managed service that offers a choice of high-performing foundation models via a single API, along with a broad set of capabilities to build generative AI applications.

Model versioning is the practice of tracking different versions of trained models. It's important for reproducibility, rollback capabilities, auditing, and understanding how model performance changes over time.

Clustering is an unsupervised learning technique used to group similar data points together based on their characteristics, without prior knowledge of the groups.

Data lineage tracks the origin, movement, transformations, and usage of data throughout its lifecycle. This is crucial for auditing, ensuring data quality, and understanding model behavior in AI systems.

Retrieval-Augmented Generation (RAG) allows the model to retrieve relevant information from the company's documents (the external knowledge base) and use that information to generate answers, reducing hallucinations and improving factual accuracy.

Pre-training foundation models involves training them on massive, diverse datasets to learn general patterns, language structures, and world knowledge.

ROUGE (Recall-Oriented Understudy for Gisting Evaluation) is a set of metrics commonly used for evaluating automatic summarization and machine translation by comparing an automatically produced summary against reference summaries.

Human-in-the-loop systems combine machine and human intelligence, where humans can review, validate, or correct AI decisions, especially in critical or ambiguous cases.

Predicting a continuous value (like price) based on input features is a regression problem.

Batch inferencing involves making predictions on a large collection of data points at once, typically in a scheduled manner, rather than in real-time.

Inference parameters like temperature, top-k, top-p, and max length significantly influence the style, diversity, length, and determinism of the generated output from a foundation model.

Embeddings are numerical vector representations of words, sentences, or other data types, where similar items have similar vector representations. They capture semantic meaning.

Data labeling is the process of annotating raw data (e.g., images, text) with informative labels or tags that provide the ground truth for the model to learn from during supervised fine-tuning.

Transfer learning is a technique where a model developed for a task is reused as the starting point for a model on a second, similar task. It's particularly useful when the dataset for the new task is small.

Data curation involves selecting, cleaning, and organizing data to ensure it is high-quality, relevant, and suitable for the fine-tuning task.

While AWS provides compliant infrastructure and services, the customer is responsible for configuring those services and building their applications in a way that meets the specific requirements of regulations like HIPAA, including data handling, access controls, and audit logging.

RAG is highly valuable for building customer support chatbots that can access and use a company's latest product manuals or FAQs to provide accurate answers, reducing the need for frequent model retraining.

Encrypting data both at rest (e.g., in Amazon S3) and in transit (e.g., using TLS/SSL) is a fundamental security best practice to protect sensitive training data.

Fine-tuning involves taking a pre-trained foundation model and further training it on a smaller, task-specific dataset to adapt its capabilities to that particular task.

Securing the inference endpoint with proper authentication, authorization, and network controls (e.g., using VPCs and security groups) is crucial to prevent unauthorized access and protect sensitive data.

Fairness is a key principle of Responsible AI, ensuring that AI systems do not perpetuate or amplify existing biases and treat all individuals and groups equitably.