CertiFive ideas
AI is the field. ML learns from data. Deep learning uses deep neural networks. Generative AI creates content. Agentic AI plans and uses tools.
Study guide · Flashcards & memory notes
AWS AI Practitioner Flashcards & Memory Notes
50 recall cards for the facts AIF-C01 exam guide version 1.1 keeps asking you to separate: learning types, generative AI, foundation models, responsibility, and controls.
How to use these cards: Say the answer out loud before you flip the card, and mark it honestly. Revisit the cards you missed tomorrow rather than rereading them right away. Progress is kept only while this page is open. For the full explanations behind each card, see the core notes.
Study mode
Choose a deck, flip each card, and mark what you already know.
You marked every card in this deck as known. Shuffle and run through it again, or choose another deck.
All flashcards
Select a question to reveal its answer.
Deck 1 · AI and ML fundamentals
Learning types, inference styles, and the metric that matches the mistake.
1.1How do AI, ML, and deep learning relate?
Deep learning is a subset of ML, and ML is a subset of AI. Generative AI creates new content. Agentic AI uses a model to plan and take actions.
1.2What is supervised learning?
Learning from labeled examples. Regression predicts a number. Classification predicts a category.
1.3What is unsupervised learning?
Finding structure in unlabeled data, such as clustering. It is not a labeled fraud classifier.
1.4What is reinforcement learning?
An agent learns from rewards and penalties as it acts. It is not a static table of labels.
1.5What is the difference between training and inference?
Training builds or updates the model. Inference uses it to produce a result.
1.6Name the four inference styles on AIF-C01.
Batch, real-time, asynchronous, and serverless. Batch is offline. Real-time waits. Asynchronous returns later. Serverless means you do not manage the servers.
1.7What do precision and recall measure?
Precision is how many flagged items were truly positive. Recall is how many true positives you found. Accuracy can look fine when the event is rare.
1.8When is a traditional model a better choice than a foundation model?
When you need an explanation, a narrow labeled task, or a constraint that rules out an opaque generative model.
1.9Which service turns speech into text, and which turns text into speech?
Amazon Transcribe is speech to text. Amazon Polly is text to speech. Amazon Translate changes the language of text.
1.10What is overfitting?
The model memorizes the training data and fails on new examples. Underfitting is too simple and fails on both.
Deck 2 · Generative AI
Tokens, context, agents, and the AWS service that hosts them.
2.1What is a token, and why does it matter to the bill?
A token is a piece of text the model reads or writes. Token pricing charges for input and output. Longer prompts and answers cost more.
2.2What is an embedding?
A vector that places similar meanings near each other so retrieval can search by meaning, not only by keywords.
2.3What is context engineering?
Choosing and arranging what the model sees: instructions, retrieved passages, history, and tool results. It is broader than one user sentence.
2.4What is the Model Context Protocol?
A standard way for an agent to connect to external tools and data. It is not a foundation model or an IAM policy.
2.5Name two disadvantages of generative AI.
Hallucinations, weak interpretability, inaccuracy, and nondeterminism. The same prompt can produce a different answer.
2.6What does Amazon Bedrock provide?
A managed API for foundation models, including Amazon Nova. You are not training the base model from scratch.
2.7What is Amazon Bedrock AgentCore for?
Running agents, including identity and policy. Strands Agents is the developer-tool framework. Kiro is the agentic development tool.
2.8What is Amazon Quick on this exam?
The in-scope assistant for workplace knowledge and generative business intelligence. The version 1.1 revision also adds Amazon Q to the in-scope list.
2.9What is prompt caching?
Reuse of a repeated prompt prefix so you do not pay the full input-token price every time.
2.10What is a diffusion model used for?
Generating images by refining noise over steps. A transformer is the architecture behind most large language models.
Deck 3 · Foundation models
Retrieval, prompts, customization, and how you score an answer.
3.1Does Retrieval Augmented Generation update model weights?
No. It places retrieved passages in the prompt. Fine-tuning updates weights. Amazon Bedrock Knowledge Bases is the managed retrieval path.
3.2Where can AIF-C01 embeddings be stored?
Amazon OpenSearch Service, Amazon Aurora, Amazon Neptune, or Amazon RDS for PostgreSQL. Amazon MemoryDB left the in-scope list in version 1.1.
3.3What does a higher temperature do?
It makes output more varied. A lower temperature is more stable. Temperature does not make an answer more factual.
3.4What is the difference between zero-shot and few-shot prompting?
Zero-shot gives no example. Few-shot gives several. Single-shot gives one. Chain-of-thought asks for steps.
3.5What is model distillation?
A smaller student model learns from a larger teacher, so later inference is cheaper. It is not the same as putting examples in a prompt.
3.6What is Amazon Bedrock Prompt Management for?
Storing and versioning prompts. It does not fine-tune the model and it is not a content filter.
3.7Name a prompt risk that is not a bad writing style.
Exposure, poisoning, hijacking, or jailbreaking. A user can override instructions or try to bypass a safety control.
3.8What do ROUGE and BLEU score?
ROUGE scores summary overlap. BLEU scores translation overlap. BERTScore compares embeddings. LLM-as-a-judge asks a model to score an output.
3.9Name three business-alignment metrics from exam guide 1.1.
Task completion rate, user satisfaction, and cost per interaction. A strong BLEU score is not one of them.
3.10What is in-context learning?
Examples inside the prompt, with no weight update. Pre-training from scratch is the expensive opposite.
Deck 4 · Responsible AI
Bias, guardrails, and what a reviewer can actually see.
4.1Name the responsible-AI features in the exam guide.
Bias, fairness, inclusivity, robustness, safety, and veracity.
4.2What do Amazon Bedrock Guardrails do?
They filter inputs and outputs, including denied topics and sensitive information. They do not replace IAM.
4.3What dataset properties reduce biased outputs?
Inclusivity, diversity, curated sources, and balance. A set that represents one group encodes that gap.
4.4What legal and trust risks does the guide name for generative AI?
Intellectual-property claims, biased outputs, loss of customer trust, end-user harm, and hallucinations.
4.5What is a SageMaker Model Card for?
Documenting intended use, data notes, metrics, and limitations so a reviewer can see what the model is.
4.6What is the transparency tradeoff?
A simpler, explainable model can be less capable. A large foundation model is harder to explain.
4.7How do you check that a model treats groups differently?
Subgroup analysis, plus label-quality checks and human audits.
4.8Why does sustainability show up in model selection?
Large models have an environmental cost. A distilled model can be the responsible choice when quality is enough.
4.9What is human-centered design for explainable AI?
A way for users to give feedback, and a decision they can see. Hiding the reason is the opposite.
4.10What is the difference between bias as unfairness and underfitting?
Unfairness is a systematic harm to a group. Underfitting is a model that is too simple and inaccurate for everyone.
Deck 5 · Security and governance
The control that matches the risk. You are recognizing it, not building a program.
5.1Who secures the data and the prompts in a managed AI service?
The customer. AWS secures the infrastructure of the managed service. That is shared responsibility applied to AI.
5.2How do you ground a hallucinated answer?
Retrieval grounding, output validation, and confidence scoring. Raising the temperature does not add facts.
5.3What does Amazon Macie do?
It finds sensitive data, including in Amazon S3. AWS KMS holds encryption keys. AWS PrivateLink keeps traffic private.
5.4What do AgentCore Identity and Policy control?
Which identity an agent uses and what that agent is allowed to do. Guardrails still filter the content.
5.5Where do you download AWS compliance reports?
AWS Artifact. AWS CloudTrail is API history. AWS Config is configuration. Amazon Inspector is vulnerabilities.
5.6What is data residency?
Keeping data in a required geography. Retention is how long you keep prompts, outputs, and logs.
5.7What is the Generative AI Security Scoping Matrix?
A governance framework named in the exam guide for scoping how an AI workload is secured. You recognize it. You are not asked to author one.
5.8What should you log for an AI interaction?
An audit trail of the interaction, including enough context to review a bad output later. CloudTrail covers the API call.
5.9What is prompt injection?
Untrusted text that tries to override the instruction. Filter and validate inputs and outputs. It is not solved by a larger context window alone.
5.10Are Amazon GuardDuty and AWS WAF in scope for AIF-C01?
No. The exam guide lists them out of scope. Use IAM, encryption, Macie, PrivateLink, Guardrails, and the governance services that are named.
Memory notes
Groupings that make the highest-yield AIF-C01 facts easier to recall.
Speech and text
- Transcribe speech to text
- Polly text to speech
- Translate language to language
- Comprehend meaning in text · Lex a bot
Change the model or the prompt
- Prompt or RAG weights stay put
- Fine-tune weights change
- Distill a smaller student
- Pre-train the expensive start
Metrics
Precision is false alarms. Recall is misses. F1 balances them. ROUGE and BLEU score text overlap. Task completion, satisfaction, and cost per interaction score the business.
Stop a hallucination
Retrieve, validate, and score confidence. Do not raise temperature. Guardrails filter. Citations show the source.
Controls
- IAM who can call
- KMS keys · Macie sensitive data
- CloudTrail API history · Artifact reports
- AgentCore what the agent may do
Continue the AWS AI Practitioner study path
Test what you have learned with a timed practice exam.
Overview
Exam format, the 700 passing score, domain weights, and a study plan.
Core Notes
Reference notes for learning types, generative AI, prompts, evaluation, and governance.
Flashcards & Memory Notes
50 recall cards in five decks, plus memory notes for models, prompts, and controls.
Practice Exams
200 original questions with custom exams, explanations, and a score report by domain.
Related Tools
Useful companions while you study.
Password Generator
Generate strong random passwords with custom length, characters, symbols, and security options.
JSON YAML Converter
Convert structured data between JSON and YAML with formatting and validation.
Base64 Converter
Encode text and files to Base64 or decode Base64 data with UTF-8 support.
Diff Checker
Compare two text blocks and highlight added, removed, and changed content.
Legal Disclaimer
Certicab provides general informational, educational, and convenience resources. Calculations, conversions, estimates, and learning materials may contain errors or become outdated. Financial, tax, medical, legal, and travel information is not professional advice. Verify important results and current requirements with qualified professionals or authoritative sources. Protect sensitive files and personal information, review each tool’s privacy limitations, and use only content you are authorized to process. You are responsible for how you use and share results. Study resources are independent and do not guarantee exam success or imply certification-provider endorsement. Amazon Web Services, AWS, and related marks are trademarks of Amazon.com, Inc. or its affiliates. Certicab is not affiliated with, endorsed by, or sponsored by Amazon.