Foundational Exam
Entry-level certification validating basic AWS knowledge.
Getting Started: Exam Essentials
Free knowledge base
AWS Certified AI Practitioner — key terms, tricks & tips
Everything from the course in one searchable place: 210 entries. Use it to review before a practice test or look up a word you forgot.
210 results
Exam Domain
A specific knowledge area covered by the certification exam.
Getting Started: Exam Essentials
Weighting
The percentage contribution of a domain to the total exam score.
Getting Started: Exam Essentials
Multiple-Choice
Question type with one correct answer from several options.
Getting Started: Exam Essentials
Multiple-Response
Question type with multiple correct answers from several options.
Getting Started: Exam Essentials
Exam Guide
Official AWS document detailing exam content and objectives.
Getting Started: Exam Essentials
Online Proctoring
Taking an exam remotely under supervision via webcam.
Getting Started: Exam Essentials
Understanding the AWS Certified AI Practitioner Exam
To remember the exam's focus: 'AI Practitioner is about *What* and *Why*, not *How*.'
Getting Started: Exam Essentials
Understanding the AWS Certified AI Practitioner Exam
The exam focuses on identifying the *right* AWS AI/ML service for a given business problem, not on how to implement or configure it. Keywords like 'identify,' 'describe,' 'explain business value,' and 'use case' are common.
Getting Started: Exam Essentials
Understanding the AWS Certified AI Practitioner Exam
Spending too much time on deep technical details of services like SageMaker, which are beyond the scope of this foundational exam.
Getting Started: Exam Essentials
Understanding the AWS Certified AI Practitioner Exam
Not reviewing the official exam guide, leading to studying irrelevant topics or missing key objectives.
Getting Started: Exam Essentials
Understanding the AWS Certified AI Practitioner Exam
Underestimating the importance of understanding the business value and use cases of AWS AI/ML services.
Getting Started: Exam Essentials
Multiple-Choice Question
Question with one correct answer from several options.
Getting Started: Exam Essentials
Multiple-Response Question
Question requiring selection of multiple correct answers.
Getting Started: Exam Essentials
Scenario-Based Question
Presents a problem, asks for the best AWS solution.
Getting Started: Exam Essentials
Mark for Review
Feature to flag questions for later revisit.
Getting Started: Exam Essentials
Exam Interface
The on-screen environment for taking the certification exam.
Getting Started: Exam Essentials
Time Management
Strategically allocating time to complete all exam questions.
Getting Started: Exam Essentials
Navigating the Exam Interface and Question Types
To remember question types: 'M'ultiple-Choice is 'M'ain, 'M'ultiple-Response is 'M'ore, 'S'cenario is 'S'tory.
Getting Started: Exam Essentials
Navigating the Exam Interface and Question Types
The exam will test your ability to identify the 'most appropriate' or 'best' solution among several plausible options. Pay close attention to qualifiers like 'most cost-effective,' 'highest security,' or 'least operational overhead' in scenario questions.
Getting Started: Exam Essentials
Navigating the Exam Interface and Question Types
Spending too much time on a single difficult question, leading to not finishing the exam.
Getting Started: Exam Essentials
Navigating the Exam Interface and Question Types
Not reading the entire question or all answer options carefully, leading to misinterpretation.
Getting Started: Exam Essentials
Navigating the Exam Interface and Question Types
Failing to select the exact number of answers specified in multiple-response questions.
Getting Started: Exam Essentials
Artificial Intelligence (AI)
Systems mimicking human intelligence to solve problems.
AI/ML & Generative AI Core Concepts
Machine Learning (ML)
AI subset enabling systems to learn from data without explicit programming.
AI/ML & Generative AI Core Concepts
Deep Learning (DL)
ML subset using multi-layered neural networks for complex patterns.
AI/ML & Generative AI Core Concepts
Supervised Learning
ML with labeled data to predict outputs (classification, regression).
AI/ML & Generative AI Core Concepts
Unsupervised Learning
ML with unlabeled data to find hidden patterns (clustering, dimensionality reduction).
AI/ML & Generative AI Core Concepts
Reinforcement Learning
ML agent learns by interacting with environment, maximizing rewards.
AI/ML & Generative AI Core Concepts
Features
Input variables or attributes used by an ML model.
AI/ML & Generative AI Core Concepts
Model Training
Process of adjusting model parameters using data to minimize error.
AI/ML & Generative AI Core Concepts
Key AI/ML Concepts: Definitions and Applications
AIM-D: AI is the Aim, ML provides the Method, Deep Learning Delivers. Or, Supervised = Teacher, Unsupervised = Explorer, Reinforcement = Gamer.
AI/ML & Generative AI Core Concepts
Key AI/ML Concepts: Definitions and Applications
The exam often tests your ability to distinguish between AI, ML, and DL. Remember the hierarchy: AI is the umbrella, ML is a subset of AI, and DL is a subset of ML. Keywords like 'labeled data' point to supervised learning, while 'unlabeled data' or 'pattern discovery' suggest unsupervised learning.
AI/ML & Generative AI Core Concepts
Key AI/ML Concepts: Definitions and Applications
Confusing AI, ML, and DL as interchangeable terms; they are hierarchical.
AI/ML & Generative AI Core Concepts
Key AI/ML Concepts: Definitions and Applications
Assuming all ML requires labeled data; unsupervised learning does not.
AI/ML & Generative AI Core Concepts
Key AI/ML Concepts: Definitions and Applications
Thinking Deep Learning is always the best solution; simpler ML models can be more efficient for certain problems.
AI/ML & Generative AI Core Concepts
Generative AI
AI that creates new, original content.
AI/ML & Generative AI Core Concepts
Discriminative AI
AI that classifies or predicts based on existing data.
AI/ML & Generative AI Core Concepts
Latent Space
Compressed representation of data's features.
AI/ML & Generative AI Core Concepts
GANs
Generative Adversarial Networks; generator vs. discriminator.
AI/ML & Generative AI Core Concepts
Transformer
Neural network for sequential data, uses self-attention.
AI/ML & Generative AI Core Concepts
Diffusion Models
Generative models that denoise data iteratively.
AI/ML & Generative AI Core Concepts
Large Language Model (LLM)
Generative AI trained on vast text data.
AI/ML & Generative AI Core Concepts
Deepfake
Synthetic media created by AI, often misleading.
AI/ML & Generative AI Core Concepts
Generative AI Concepts: How it Works and Impact
Imagine 'Genie-AI' that can *generate* anything you wish, versus 'Detective-AI' that can only *discriminate* between existing clues.
AI/ML & Generative AI Core Concepts
Generative AI Concepts: How it Works and Impact
The exam frequently asks to differentiate between discriminative and generative AI. Memorize that discriminative models classify or predict (e.g., 'Is this a cat?'), while generative models create new content (e.g., 'Draw me a cat').
AI/ML & Generative AI Core Concepts
Generative AI Concepts: How it Works and Impact
Confusing generative AI with discriminative AI: Generative creates, discriminative classifies.
AI/ML & Generative AI Core Concepts
Generative AI Concepts: How it Works and Impact
Underestimating the data requirements: Generative models need massive, diverse datasets for effective training.
AI/ML & Generative AI Core Concepts
Generative AI Concepts: How it Works and Impact
Ignoring ethical implications: Always consider bias, misuse, and copyright when deploying generative AI solutions.
AI/ML & Generative AI Core Concepts
ML Lifecycle
Iterative process from problem definition to model deployment and monitoring.
AI/ML & Generative AI Core Concepts
Data Preparation
Cleaning, transforming, and organizing raw data for machine learning.
AI/ML & Generative AI Core Concepts
Feature Engineering
Creating new input features from existing data to improve model performance.
AI/ML & Generative AI Core Concepts
Model Evaluation
Assessing a model's performance on unseen data using specific metrics.
AI/ML & Generative AI Core Concepts
Model Deployment
Making a trained ML model available for use in a production environment.
AI/ML & Generative AI Core Concepts
Data Drift
Changes in the statistical properties of the input data over time.
AI/ML & Generative AI Core Concepts
Concept Drift
Changes in the relationship between input features and the target variable.
AI/ML & Generative AI Core Concepts
The Machine Learning Lifecycle: Data to Deployment
P-D-P-T-E-D-M: 'Panda's Don't Play Tag, Except During Mating!' (Problem, Data Collection, Preparation, Training, Evaluation, Deployment, Monitoring)
AI/ML & Generative AI Core Concepts
The Machine Learning Lifecycle: Data to Deployment
The exam often tests your understanding of the *order* of these stages and the *purpose* of each. Look for keywords like 'data cleaning' (preparation), 'generalization' (evaluation), or 'production environment' (deployment).
AI/ML & Generative AI Core Concepts
The Machine Learning Lifecycle: Data to Deployment
Skipping or rushing the problem definition stage, leading to building the wrong solution.
AI/ML & Generative AI Core Concepts
The Machine Learning Lifecycle: Data to Deployment
Neglecting data preparation, which results in 'garbage in, garbage out' for model performance.
AI/ML & Generative AI Core Concepts
The Machine Learning Lifecycle: Data to Deployment
Failing to continuously monitor deployed models, allowing performance to degrade silently.
AI/ML & Generative AI Core Concepts
Structured Data
Highly organized data, typically in tables with defined rows and columns.
AI/ML & Generative AI Core Concepts
Unstructured Data
Data without a predefined schema, like text, images, audio, video.
AI/ML & Generative AI Core Concepts
Data Quality
Accuracy, completeness, consistency, and relevance of data.
AI/ML & Generative AI Core Concepts
Data Cleaning
Process of fixing errors, inconsistencies, and missing values in data.
AI/ML & Generative AI Core Concepts
Data Labeling
Assigning meaningful tags or categories to raw data for supervised learning.
AI/ML & Generative AI Core Concepts
Data Augmentation
Creating synthetic variations of existing data to expand dataset size.
AI/ML & Generative AI Core Concepts
Tokenization
Breaking down text into smaller units (tokens) for NLP models.
AI/ML & Generative AI Core Concepts
Data for AI/ML and Generative AI: Types and Preparation
Imagine a 'DATA CLEANING CREW' (Cleaning, Labeling, Augmentation, Normalization, Engineering, New Features, Ground Truth) making your data sparkling clean and ready for AI!
AI/ML & Generative AI Core Concepts
Data for AI/ML and Generative AI: Types and Preparation
The exam often asks about the importance of data quality and common preprocessing steps. Keywords to spot include 'missing values,' 'outliers,' 'imputation,' 'normalization,' and 'feature engineering.' Remember that Amazon SageMaker Ground Truth is AWS's service for data labeling.
AI/ML & Generative AI Core Concepts
Data for AI/ML and Generative AI: Types and Preparation
Ignoring data quality issues, leading to 'garbage in, garbage out' models.
AI/ML & Generative AI Core Concepts
Data for AI/ML and Generative AI: Types and Preparation
Using insufficient or unrepresentative data, causing models to perform poorly on real-world scenarios.
AI/ML & Generative AI Core Concepts
Data for AI/ML and Generative AI: Types and Preparation
Skipping data preparation steps, assuming raw data is ready for model training.
AI/ML & Generative AI Core Concepts
Foundation Model (FM)
Large AI model pre-trained on vast data, adaptable to many tasks.
Understanding Foundation Models
Transformer Architecture
Neural network architecture using self-attention, common for FMs.
Understanding Foundation Models
Self-Attention
Mechanism allowing models to weigh importance of different input parts.
Understanding Foundation Models
Pre-training
Initial, unsupervised training of an FM on massive datasets.
Understanding Foundation Models
Fine-tuning
Adapting a pre-trained FM to a specific task with labeled data.
Understanding Foundation Models
Emergent Capabilities
Complex behaviors arising from model scale, not explicitly programmed.
Understanding Foundation Models
Foundation Model Concepts: Architecture & Training
Imagine a 'TRANSFORMER' robot: it first 'PRE-TRAINS' by observing the entire world (massive data) to learn general skills, then 'FINE-TUNES' for a specific mission (task-specific data).
Understanding Foundation Models
Foundation Model Concepts: Architecture & Training
The exam often tests your understanding of the training lifecycle: pre-training, fine-tuning, and deployment. Be ready to distinguish between these phases and their purpose. Keywords to spot include 'general knowledge,' 'domain-specific,' and 'unsupervised vs. supervised learning.'
Understanding Foundation Models
Foundation Model Concepts: Architecture & Training
Confusing pre-training (general learning) with fine-tuning (task-specific adaptation).
Understanding Foundation Models
Foundation Model Concepts: Architecture & Training
Believing all foundation models are LLMs; LLMs are a type of FM, but FMs can also be multimodal.
Understanding Foundation Models
Foundation Model Concepts: Architecture & Training
Underestimating the computational resources required for pre-training, even if fine-tuning is more accessible.
Understanding Foundation Models
Encoder-Only Model
Processes input to create contextualized representations for understanding tasks.
Understanding Foundation Models
Decoder-Only Model
Generates new content sequentially, predicting the next token.
Understanding Foundation Models
Encoder-Decoder Model
Transforms an input sequence into a different output sequence.
Understanding Foundation Models
Multimodal Model
Processes and generates across multiple data types (text, image, audio).
Understanding Foundation Models
Autoregressive
A model that predicts future values based on past values in a sequence.
Understanding Foundation Models
BERT
Bidirectional Encoder Representations from Transformers, an encoder-only model.
Understanding Foundation Models
GPT
Generative Pre-trained Transformer, a decoder-only model.
Understanding Foundation Models
T5
Text-to-Text Transfer Transformer, an encoder-decoder model.
Understanding Foundation Models
Exploring Different Foundation Model Types
Think 'E' for Encoder-only = 'Evaluate' (understand). 'D' for Decoder-only = 'Draft' (generate). 'E-D' for Encoder-Decoder = 'Exchange' (translate/transform). 'M' for Multimodal = 'Mix' (different data types).
Understanding Foundation Models
Exploring Different Foundation Model Types
The exam often presents scenarios and asks which type of foundation model is most appropriate. Look for keywords: 'understanding,' 'classification,' 'sentiment' point to encoder-only. 'Generating,' 'writing,' 'summarizing' (long to short) point to decoder-only. 'Translating,' 'summarizing' (transforming) point to encoder-decoder. 'Image captioning,' 'visual Q&A' point to multimodal.
Understanding Foundation Models
Exploring Different Foundation Model Types
Confusing encoder-only models (understanding) with decoder-only models (generation).
Understanding Foundation Models
Exploring Different Foundation Model Types
Assuming a single model type can efficiently handle all AI tasks; often, a combination is best.
Understanding Foundation Models
Exploring Different Foundation Model Types
Underestimating the power of multimodal models for tasks involving diverse data inputs.
Understanding Foundation Models
Analytical AI
AI that processes data to extract insights, classify, or predict.
Understanding Foundation Models
API
Application Programming Interface; allows software components to communicate.
Understanding Foundation Models
Content Moderation
Monitoring and filtering user-generated content to ensure compliance.
Understanding Foundation Models
Sentiment Analysis
Determining the emotional tone or opinion expressed in text.
Understanding Foundation Models
Code Generation
Using AI to automatically write or suggest programming code.
Understanding Foundation Models
Personalized Learning
Tailoring educational content and pace to individual student needs.
Understanding Foundation Models
Practical Foundation Model Use Cases and Examples
To remember FM uses: 'G.A.I.N.' - Generative, Analytical, Integration, New Industries. FMs help you GAIN value!
Understanding Foundation Models
Practical Foundation Model Use Cases and Examples
The exam often asks about specific AWS services that leverage FMs, such as Amazon Bedrock for accessing various FMs or Amazon CodeWhisperer for code generation. Understand the 'what' and 'why' of these services.
Understanding Foundation Models
Practical Foundation Model Use Cases and Examples
Assuming FMs are 'set it and forget it' solutions; human oversight and validation are crucial.
Understanding Foundation Models
Practical Foundation Model Use Cases and Examples
Trying to use a general-purpose FM for highly specialized tasks without fine-tuning or prompt engineering.
Understanding Foundation Models
Practical Foundation Model Use Cases and Examples
Underestimating the importance of data privacy and ethical considerations when deploying FMs.
Understanding Foundation Models
F1-score
Harmonic mean of precision and recall, balances false positives/negatives.
Understanding Foundation Models
BLEU
Metric for evaluating the quality of machine-translated text.
Understanding Foundation Models
ROUGE
Set of metrics for evaluating automatic summarization and machine translation.
Understanding Foundation Models
Latency
Time taken for a model to process an input and generate an output.
Understanding Foundation Models
Scalability
Ability of a system to handle increasing workloads or demands.
Understanding Foundation Models
Bias
Systematic error in a model's output due to skewed training data.
Understanding Foundation Models
Explainability
Degree to which a human can understand the cause of a model's output.
Understanding Foundation Models
Benchmarking
Comparing model performance against standardized datasets and tasks.
Understanding Foundation Models
Evaluating and Selecting Foundation Models
To pick the right FM, remember 'PERFECT': Performance, Ethics, Resources (Cost/Latency), Features, Ecosystem, Customization, Transparency.
Understanding Foundation Models
Evaluating and Selecting Foundation Models
The exam often presents scenarios where you must choose the 'best' model. Look for keywords indicating specific requirements like 'low latency,' 'high accuracy for medical imaging,' or 'cost-effective.' Your answer should align with these explicit constraints, considering trade-offs.
Understanding Foundation Models
Evaluating and Selecting Foundation Models
Selecting a model based solely on its size or perceived 'intelligence' without considering specific project requirements.
Understanding Foundation Models
Evaluating and Selecting Foundation Models
Ignoring ethical implications and potential biases, leading to unintended negative consequences in deployment.
Understanding Foundation Models
Evaluating and Selecting Foundation Models
Underestimating the total cost of ownership, including inference, fine-tuning, and infrastructure expenses.
Understanding Foundation Models
Responsible AI (RAI)
Framework ensuring ethical, fair, and accountable AI development and use.
Responsible AI Practices
Beneficence
Ethical principle stating AI should aim to do good and benefit humanity.
Responsible AI Practices
Non-maleficence
Ethical principle requiring AI systems to avoid causing harm.
Responsible AI Practices
Autonomy
Principle emphasizing respect for human choice and control over AI.
Responsible AI Practices
Justice
Ethical principle ensuring fair distribution of AI benefits and impacts.
Responsible AI Practices
Explicability
The ability to understand and explain how an AI system makes decisions.
Responsible AI Practices
Human-centric AI
Approach placing human well-being and values at the core of AI design.
Responsible AI Practices
Core Responsible AI Concepts: Ethics and Principles
Remember 'BANJ-E': Beneficence, Autonomy, Non-maleficence, Justice, Explicability – the core ethical principles for AI.
Responsible AI Practices
Core Responsible AI Concepts: Ethics and Principles
The exam often tests your understanding of core ethical principles. Look for keywords like 'fairness,' 'transparency,' 'accountability,' and 'human-centric' in questions about responsible AI design and deployment.
Responsible AI Practices
Core Responsible AI Concepts: Ethics and Principles
Overlooking the societal impact of AI in favor of purely technical performance.
Responsible AI Practices
Core Responsible AI Concepts: Ethics and Principles
Assuming that an AI model is 'neutral' simply because it's data-driven.
Responsible AI Practices
Core Responsible AI Concepts: Ethics and Principles
Failing to establish clear lines of accountability for AI system outcomes.
Responsible AI Practices
Fairness
AI systems treating all individuals equitably.
Responsible AI Practices
Transparency
Making AI system workings and decisions understandable.
Responsible AI Practices
Explainable AI (XAI)
Techniques to interpret AI model predictions.
Responsible AI Practices
Demographic Parity
Equal positive outcome rates across groups.
Responsible AI Practices
SHAP
A method for explaining individual AI predictions.
Responsible AI Practices
LIME
Explains individual predictions of any classifier.
Responsible AI Practices
Fairness and Transparency in AI
Imagine a 'FAIR'ground with a 'TRANSPARENT' tent. Everyone gets equal rides (Fairness), and you can see exactly how the rides work (Transparency).
Responsible AI Practices
Fairness and Transparency in AI
For the exam, memorize that Amazon SageMaker Clarify is a key AWS service for detecting bias in datasets and models, and for providing model explainability. Look for questions about identifying and mitigating bias or explaining model predictions.
Responsible AI Practices
Fairness and Transparency in AI
Assuming that a model trained on a large dataset is inherently fair without explicit bias checks.
Responsible AI Practices
Fairness and Transparency in AI
Believing that transparency only means making the code open source, rather than explaining model behavior.
Responsible AI Practices
Fairness and Transparency in AI
Ignoring the need for continuous monitoring of fairness and transparency metrics after deployment.
Responsible AI Practices
AI Bias
Systematic errors leading to unfair outcomes.
Responsible AI Practices
Fairness Metrics
Quantifiable measures of AI system fairness.
Responsible AI Practices
Pre-processing
Adjusting data before model training.
Responsible AI Practices
Feature Importance
Measures input impact on model output.
Responsible AI Practices
Interpretability
Degree to which a human can understand.
Responsible AI Practices
Mitigating Bias and Ensuring Explainability in AI
BIAS: B-Balance (data), I-Inspect (outputs), A-Adjust (model), S-Supervise (continuously). EXPLAIN: E-Examine (features), X-eXplain (predictions), P-Provide (reasons), L-Look (for insights), A-Analyze (impact), I-Interpret (results), N-Navigate (complexity).
Responsible AI Practices
Mitigating Bias and Ensuring Explainability in AI
The exam often tests your ability to identify different types of bias (e.g., historical, selection) and the stages where they can be introduced. Be ready to distinguish between pre-processing, in-processing, and post-processing mitigation techniques. Also, know the core purpose of XAI and common techniques like LIME and SHAP.
Responsible AI Practices
Mitigating Bias and Ensuring Explainability in AI
Assuming that a large dataset automatically guarantees fairness; unrepresentative large datasets can still contain significant biases.
Responsible AI Practices
Mitigating Bias and Ensuring Explainability in AI
Focusing solely on model accuracy without evaluating fairness metrics, leading to high-performing but biased systems.
Responsible AI Practices
Mitigating Bias and Ensuring Explainability in AI
Treating explainability as an afterthought rather than integrating it into the AI development process from the beginning.
Responsible AI Practices
Differential Privacy
Adds noise to data to protect individual privacy.
Responsible AI Practices
Federated Learning
Trains models on decentralized data without sharing it.
Responsible AI Practices
Homomorphic Encryption
Performs computations on encrypted data directly.
Responsible AI Practices
Adversarial Attack
Manipulates AI input to cause incorrect outputs.
Responsible AI Practices
Data Poisoning
Injects malicious data into training sets.
Responsible AI Practices
Membership Inference
Determines if data was in a model's training set.
Responsible AI Practices
Data Minimization
Collecting only essential data for a purpose.
Responsible AI Practices
Pseudonymization
Replaces identifiers with artificial ones.
Responsible AI Practices
Privacy and Security in AI Systems
P.S. I Love AI: **P**rivacy, **S**ecurity, **I**nference attacks, **L**earning (Federated), **A**dversarial attacks, **I**nfrastructure security.
Responsible AI Practices
Privacy and Security in AI Systems
The exam often tests your understanding of privacy-preserving AI techniques. Look for questions about differential privacy, federated learning, and homomorphic encryption as methods to protect sensitive data while still enabling AI functionality.
Responsible AI Practices
Privacy and Security in AI Systems
Assuming anonymized data is always fully private and cannot be re-identified.
Responsible AI Practices
Privacy and Security in AI Systems
Focusing only on traditional cybersecurity and neglecting AI-specific threats like adversarial attacks.
Responsible AI Practices
Privacy and Security in AI Systems
Not implementing 'privacy by design' from the initial stages of AI development.
Responsible AI Practices
Amazon SageMaker
Fully managed service for building, training, and deploying ML models.
AWS Services for AI/ML & Generative AI
Amazon Rekognition
AI service for image and video analysis using computer vision.
AWS Services for AI/ML & Generative AI
AI Services
Pre-trained, ready-to-use services for common AI tasks (e.g., vision, speech).
AWS Services for AI/ML & Generative AI
ML Services
Managed platforms and tools for the entire machine learning workflow.
AWS Services for AI/ML & Generative AI
Computer Vision
Field of AI enabling computers to 'see' and interpret digital images/videos.
AWS Services for AI/ML & Generative AI
Overview of AWS AI/ML Services (SageMaker, Rekognition)
Remember 'Sage' for 'Smart Algorithms Get Engineered' (SageMaker) and 'Rekog' for 'Recognize Everything Known' (Rekognition).
AWS Services for AI/ML & Generative AI
Overview of AWS AI/ML Services (SageMaker, Rekognition)
The exam often asks to identify the correct AWS service for a given AI/ML task. Remember that Rekognition is for pre-trained computer vision, while SageMaker is for building and managing custom ML models.
AWS Services for AI/ML & Generative AI
Overview of AWS AI/ML Services (SageMaker, Rekognition)
Confusing SageMaker (custom model building) with Rekognition (pre-trained vision API).
AWS Services for AI/ML & Generative AI
Overview of AWS AI/ML Services (SageMaker, Rekognition)
Assuming AI services require ML expertise; they are designed for developers without it.
AWS Services for AI/ML & Generative AI
Overview of AWS AI/ML Services (SageMaker, Rekognition)
Underestimating the scope of SageMaker's capabilities across the entire ML lifecycle.
AWS Services for AI/ML & Generative AI
Amazon Bedrock
Managed service for accessing and customizing FMs via a single API.
AWS Services for AI/ML & Generative AI
Amazon CodeWhisperer
AI coding companion that generates code suggestions in real-time.
AWS Services for AI/ML & Generative AI
Agents for Bedrock
Bedrock feature enabling FMs to perform multi-step tasks autonomously.
AWS Services for AI/ML & Generative AI
Knowledge Bases for Bedrock
Bedrock feature allowing FMs to retrieve information from private data sources.
AWS Services for AI/ML & Generative AI
Exploring AWS Generative AI Services
Imagine a 'Bedrock' of powerful models you build upon, and a 'CodeWhisperer' quietly suggesting perfect lines of code in your ear.
AWS Services for AI/ML & Generative AI
Exploring AWS Generative AI Services
Memorize the core function of Bedrock (managed FM access) and CodeWhisperer (AI code generation). The exam often tests your ability to choose the right service for a given generative AI task.
AWS Services for AI/ML & Generative AI
Exploring AWS Generative AI Services
Confusing Bedrock's role (FM access and customization) with a service that directly provides pre-built generative AI applications.
AWS Services for AI/ML & Generative AI
Exploring AWS Generative AI Services
Underestimating CodeWhisperer's security scanning capabilities; it's not just about code generation.
AWS Services for AI/ML & Generative AI
Exploring AWS Generative AI Services
Assuming you need to manage the underlying infrastructure for FMs when using Bedrock; it's a fully managed service.
AWS Services for AI/ML & Generative AI
Shared Responsibility Model
AWS secures the cloud, customer secures in the cloud.
AWS Services for AI/ML & Generative AI
Least Privilege
Granting only necessary permissions for a task.
AWS Services for AI/ML & Generative AI
AWS KMS
Manages encryption keys used across AWS services.
AWS Services for AI/ML & Generative AI
Amazon VPC
Logically isolated section of AWS Cloud for resources.
AWS Services for AI/ML & Generative AI
AWS IAM
Manages access to AWS services and resources securely.
AWS Services for AI/ML & Generative AI
AWS CloudTrail
Records API calls and events for auditing and governance.
AWS Services for AI/ML & Generative AI
Data Governance
Policies and procedures for managing data lifecycle.
AWS Services for AI/ML & Generative AI
PrivateLink
Private connectivity between VPCs and AWS services.
AWS Services for AI/ML & Generative AI
Security and Compliance for AI/ML Workloads on AWS
SECURE AI: S-Shared responsibility, E-Encryption, C-Control access, U-Understand network, R-Regulations, E-Ethical AI.
AWS Services for AI/ML & Generative AI
Security and Compliance for AI/ML Workloads on AWS
The exam frequently tests the Shared Responsibility Model. Remember that customers are always responsible for their data, configurations, and network security *within* their AWS resources, even if AWS manages the underlying service.
AWS Services for AI/ML & Generative AI
Security and Compliance for AI/ML Workloads on AWS
Over-provisioning IAM permissions, leading to potential security vulnerabilities.
AWS Services for AI/ML & Generative AI
Security and Compliance for AI/ML Workloads on AWS
Neglecting to encrypt data at rest or in transit, especially for sensitive information.
AWS Services for AI/ML & Generative AI
Security and Compliance for AI/ML Workloads on AWS
Not isolating AI/ML resources within a private network, exposing them to the public internet.
AWS Services for AI/ML & Generative AI
AWS Cost Explorer
Tool to visualize, understand, and manage AWS costs and usage.
AWS Services for AI/ML & Generative AI
AWS Budgets
Service to set custom cost/usage budgets and receive alerts.
AWS Services for AI/ML & Generative AI
Managed Spot Training
SageMaker feature using spare EC2 capacity for lower-cost training.
AWS Services for AI/ML & Generative AI
Data Transfer Costs
Charges for moving data in/out of AWS regions or services.
AWS Services for AI/ML & Generative AI
Resource Tagging
Adding metadata labels to AWS resources for cost allocation.
AWS Services for AI/ML & Generative AI
Provisioned Throughput
Dedicated capacity for Bedrock models, billed hourly.
AWS Services for AI/ML & Generative AI
Cost Optimization & Monitoring for AI/ML on AWS
To remember cost tools: 'C-B-C' for Cost Explorer, Budgets, CUR – 'See Big Costs!'
AWS Services for AI/ML & Generative AI
Cost Optimization & Monitoring for AI/ML on AWS
The exam often tests your understanding of cost-saving features within specific services, such as SageMaker's Managed Spot Training or S3 Lifecycle Policies. Be prepared to identify the most cost-effective solution for a given scenario, and know that tagging is key for cost allocation.
AWS Services for AI/ML & Generative AI
Cost Optimization & Monitoring for AI/ML on AWS
Forgetting to terminate unused SageMaker instances or endpoints, leading to continuous charges.
AWS Services for AI/ML & Generative AI
Cost Optimization & Monitoring for AI/ML on AWS
Not using S3 lifecycle policies, resulting in paying for expensive storage tiers for old, infrequently accessed data.
AWS Services for AI/ML & Generative AI
Cost Optimization & Monitoring for AI/ML on AWS
Ignoring cross-region data transfer costs by placing data and compute in different AWS regions.
AWS Services for AI/ML & Generative AI