AWS Certified AI Practitioner Flashcards

The exam has 65 total questions, of which 50 are scored and 15 are unscored (unidentified pretest items used to evaluate future questions).

You need a minimum scaled score of 700, on a scale that runs from 100 to 1,000.

A scoring approach where your overall score determines pass/fail rather than requiring you to pass each domain individually, so strong performance in one domain can offset weaker performance in another.

AI is the broad field of building systems that perform tasks requiring human-like intelligence; ML is a subset of AI where models learn patterns from data rather than being explicitly programmed; deep learning is a subset of ML using multi-layered neural networks.

A machine learning approach where a model is trained on labeled data (input-output pairs) so it learns to map inputs to known correct outputs, such as classification or regression.

A machine learning approach where a model finds patterns or structure in unlabeled data, such as clustering or dimensionality reduction, without predefined output labels.

A learning paradigm where an agent learns to take actions in an environment by receiving rewards or penalties, optimizing for cumulative reward over time.

A large model pretrained on broad, diverse data that can be adapted to many downstream tasks through prompting, fine-tuning, or other customization techniques.

A category of AI that creates new content — text, images, audio, code, or other media — by learning patterns from training data, typically powered by foundation models.

A type of foundation model trained on massive text corpora to understand and generate human language, used for tasks like summarization, question answering, and content generation.

A neural network architecture that uses self-attention mechanisms to weigh the relevance of different parts of an input sequence, forming the basis of most modern LLMs.

A unit of text (a word, subword, or character sequence) that a language model processes; model context windows and costs are typically measured in tokens.

The practice of crafting input prompts to guide a foundation model toward producing more accurate, relevant, or useful outputs without changing the model's underlying weights.

A prompting technique where the model is asked to perform a task with no prior examples given in the prompt, relying solely on its pretrained knowledge.

A prompting technique where a small number of example input-output pairs are included in the prompt to guide the model toward the desired response format or behavior.

A prompting technique that encourages a model to break down its reasoning into intermediate steps before producing a final answer, often improving accuracy on complex tasks.

A technique that retrieves relevant external documents or data at inference time and feeds them into the model's context, grounding responses in up-to-date or proprietary information without retraining the model.

The process of further training a pretrained foundation model on a smaller, task-specific or domain-specific labeled dataset to improve its performance on that task.

A fully managed AWS service that provides access to foundation models from multiple providers through a single API, along with tools for customization, RAG, and building generative AI applications.

A fully managed AWS service for building, training, and deploying machine learning models across the entire ML lifecycle.

An AWS generative AI-powered assistant that helps with tasks such as answering questions, generating content, and writing or troubleshooting code, tailored to business or developer contexts.

A phenomenon where a generative AI model produces output that is factually incorrect, fabricated, or not grounded in its training data or provided context, while appearing plausible.

Systematic errors or unfair skew in model outputs that favor or disadvantage particular groups, often resulting from imbalanced or unrepresentative training data.

The degree to which a human can understand the reasons behind a model's predictions or decisions, which is important for trust, debugging, and regulatory compliance.

AWS frames Responsible AI around dimensions such as fairness, explainability, privacy and security, safety, controllability, veracity and robustness, governance, and transparency.

A capability that helps detect potential bias in datasets and models and provides explainability insights into model predictions.

A capability that continuously monitors deployed models in production for data quality issues and model drift over time.

The policies, processes, and controls an organization puts in place to ensure AI systems are developed and used responsibly, ethically, and in compliance with relevant regulations.

AWS secures the underlying cloud infrastructure and managed AI/ML services, while the customer is responsible for securing their data, access controls, model configurations, and appropriate use of the AI systems they build.

Guardrails are configurable safeguards that filter or block undesired content, topics, or PII in prompts and model responses to help keep generative AI applications safe and compliant.