2) Amazon Bedrock (GenAI)

Flashcards testing core terms and concepts related to Generative AI, Amazon Bedrock, fine-tuning, model evaluation metrics, and Amazon Nova.

Generative AI (Gen-AI)

A subset of Deep Learning used to generate new data—such as text, image, audio, code, and video—that is similar to the data it was trained on.

Foundation Model

An AI model trained on a wide variety of input data that serves as the base for generating content and executing a broad range of general tasks.

Large Language Model (LLM)

A type of AI designed to generate coherent human-like text, usually containing billions of parameters and trained on a large corpus of text data.

Non-deterministic

A property of generative language models where the generated text response may differ each time for users submitting the exact same prompt.

Diffusion Models

Generative image models that train via a forward diffusion process (adding noise) and generate images via a reverse diffusion process from text prompts.

Amazon Bedrock

A fully-managed AWS service for building Generative AI applications that provides access to multiple foundation models via unified APIs without needing server management.

Amazon Titan

A family of high-performing Foundation Models from AWS offering image, text, and multimodal choices through fully-managed APIs.

Supervised Fine-Tuning

A method to improve a model's performance on specific tasks by training it on labeled examples consisting of prompt and completion input-output pairs.

Reinforcement Fine-Tuning

A feedback-based learning technique that uses input prompts and a Reward Function to evaluate and score generated responses iteratively.

Distillation

A process for making models smaller, faster, and up to 75% less expensive by transferring knowledge from a larger teacher model to a lighter student model.

Automatic Evaluation

A feature in Amazon Bedrock to assess model quality for built-in task types (e.g., text summarization, Q&A, classification) using automated statistical metrics.

Human Evaluation

A Bedrock model assessment method where human reviewers (work teams/SMEs) rate generated responses using scoring methods like thumbs up/down or ranking.

ROUGE (Recall-Oriented Understudy for Gisting Evaluation)

An automated metric for text summarization and translation that measures matching n-grams or longest common subsequences.

BLEU (Bilingual Evaluation Understudy)

An automated metric that evaluates generated text translation quality by looking at n-gram precision and penalizing brevity.

BERTScore

An evaluation metric that measures semantic similarity by using pre-trained BERT models to compare contextualized embeddings and compute cosine similarity.

Perplexity

An automated evaluation metric that measures how well a model predicts the next token, where a lower score indicates better performance.

Retrieval-Augmented Generation (RAG)

A technique that allows a Foundation Model to reference external data sources outside its training data by storing and querying vector embeddings.

Tokenization

The process of converting raw input text into a sequence of individual word or subword tokens.

Context Window

The maximum number of tokens an LLM can process and consider at one time when generating text.

Embeddings

High-dimensional numerical vector representations of text, images, or audio that capture semantic meaning, syntactic roles, and sentiment.

Amazon Bedrock Guardrails

A feature in Amazon Bedrock that controls user interactions with models by filtering undesirable/harmful content, removing PII, and reducing hallucinations.

Amazon Bedrock Agents

A Bedrock feature that manages and executes multi-step tasks in sequence by integrating with APIs, AWS Lambda functions, action groups, and RAG knowledge bases.

Amazon Nova

A family of fast, cost-effective, enterprise-ready AI Foundation Models built by AWS and accessible through Amazon Bedrock.

Nova Premier

The most capable multimodal model in the Amazon Nova family, designed for complex reasoning tasks and for serving as a teacher model to distill custom models.

Nova Pro

Highly capable multimodal model delivering an optimal balance of accuracy, speed, and cost across enterprise tasks.

Nova Lite / Nova 2 LIte

Fast, low-cost multimodal models optimized for rapid text, document, image, and video processing.

Nova Micro

Text-only model delivering minimal latency at the lowest price point.

Nova Canvas

State-of-the-art image generation model.

Nova Reel

State-of-the-art video generation model.

Nova Sonic / Nova 2 Sonic

Real-time, bidirectional speech-to-speech conversational model.

Nova 2 Multimodal Embeddings

Advanced embedding model designed for agentic RAG and high-accuracy semantic search applications.

Nova 2 Omni

All-in-one multimodal reasoning and image generation model supporting context windows up to $1\text{M tokens}$ ($1,000,000$ tokens).