AWS Certified AI Practitioner (AIF-C01) Vocabulary Flashcards

Vocabulary practice flashcards covering AWS services, GenAI, AI/ML concepts, responsible AI principles, and governance tools based on the AWS Certified AI Practitioner mock exam.

Amazon Comprehend

An AWS Natural Language Processing (NLP) service used to analyze text and automatically determine positive, negative, or neutral sentiment.

Amazon Textract

An AWS service designed to automatically extract text, forms, tables, and structured data like invoice numbers and totals from scanned documents.

Amazon Transcribe

An AWS service that converts spoken audio or recorded conversations into written text.

Amazon Translate

An AWS service that provides machine translation to automatically convert text from one language to another.

Amazon Polly

An AWS AI service that turns written text into lifelike spoken audio.

Amazon Rekognition

An AWS service used to analyze photographs and videos to identify objects, scenes, and visual elements.

Amazon Lex

An AWS service designed for building conversational interfaces for voice and text applications.

Supervised Learning

A machine learning paradigm where a model is trained on labeled training examples to predict labels or values for new data.

Unsupervised Learning

A machine learning approach used to discover natural patterns, segments, or groupings in unlabeled data.

Reinforcement Learning

A machine learning paradigm where an agent learns optimal behavior through trial and error by receiving rewards for success and penalties for errors.

Regression

A machine learning problem focused on predicting a continuous numerical value.

Batch Inference

An inference style suited for generating predictions on large volumes of records offline when immediate real-time responses are unnecessary.

Tokens

The fundamental smaller units of text processed by language models that dictate context windows and token-based pricing.

Embeddings

Numerical vector representations of text that capture semantic relationships, allowing semantically similar passages to be retrieved.

Chunking

The process of splitting long documents into smaller meaningful sections before generating embeddings for retrieval.

Hallucination

A Generative AI limitation where a model generates plausible-sounding statements that are inaccurate or unsupported by source documents.

Context Engineering

The practice of deliberately designing the information, instructions, tools, and background context supplied to a model for a specific task.

Agentic AI

An AI system that can reason about tasks, use tools/APIs, retrieve information, and coordinate actions across external systems.

Multimodal Capability

The ability of a foundation model to process and reason across multiple modalities, such as accepting both text and images as input.

Diffusion Models

A generative model family associated with generating images from text prompts through an iterative denoising process.

Amazon Bedrock

A fully managed AWS service providing API access to multiple foundation models and Generative AI application features.

Retrieval-Augmented Generation (RAG)

An approach that retrieves current external knowledge at inference time to ground foundation model responses without retraining.

Knowledge Bases for Amazon Bedrock

An Amazon Bedrock capability that connects foundation models to data sources in Amazon S3 to retrieve relevant passages and ground responses.

Vector Database

A specialized database designed to store high-dimensional embeddings and execute similarity searches to retrieve related text chunks.

Fine-Tuning

A customization approach that adapts a foundation model's behavior by training it on a specialized set of examples.

Zero-Shot Prompting

A prompting technique where the prompt clearly states the task and expected output without providing any task-specific examples.

Few-Shot Prompting

A prompting technique that includes a few exemplar demonstrations within the prompt before asking the model for a new answer.

Temperature

An inference parameter controlling randomness; reducing it makes model responses more deterministic and consistent.

Prompt Template

A reusable prompt structure that maintains standard instructions while allowing variable, user-specific fields to be inserted at runtime.

Negative Prompts

Prompt constructs used in image generation to inform the model which visual elements should not appear.

Multi-Agent System Patterns

Architectural patterns where multiple specialized AI agents collaborate on a complex workflow by passing context to one another.

Bias and Fairness

A responsible AI concern examining whether a model produces systematically unequal outcomes for specific demographic groups.

Explainability

The responsible AI principle focused on helping stakeholders understand why a model produced a particular decision or output.

Transparency

The principle of informing users when content is AI-generated and making them aware of the system's capabilities and limitations.

Robustness

The responsible AI property evaluating whether a model maintains appropriate behavior when input conditions vary slightly.

SageMaker Model Cards

Documented AWS artifacts describing a model's intended use, limitations, evaluation information, and parameters.

AWS CloudTrail

An AWS service that logs API activity, identifying who made calls to AWS resources and when those actions took place.

Amazon CloudWatch

An AWS operations service that collects metrics, logs, and sets alarms to monitor application health and performance.

AWS Config

An AWS compliance service that records resource configuration histories and evaluates them against defined rules over time.

AWS IAM

The core AWS permissions service used to control access rights, identities, and roles for workloads and users.

AWS KMS

An AWS service that provides managed cryptographic keys to encrypt data and control key usage across services.

AWS Secrets Manager

An AWS service designed to securely store, rotate, and manage sensitive secrets like API keys and database credentials.

Amazon Macie

A security service specifically designed to discover, classify, and protect sensitive data in Amazon S3.

VPC Endpoints / AWS PrivateLink

Technologies enabling workloads in private subnets to connect to supported AWS services privately without traversing the public internet.

AWS Artifact

An AWS service that provides access to AWS compliance reports and security agreements.