Chuyển đến nội dung chính

Lesson 12: Exam Strategy & Complete Cheat Sheet

AWS AI Practitioner exam strategy. Time management. Complete cheat sheet for all 5 domains. Key services mapping. Common exam traps.

📝 Exam Prep — Lesson 12 Lesson 12: Exam Strategy & Complete Cheat Sheet

AWS Certified AI Practitioner (AIF-C01) Exam Prep

Domain 5: Security, Compliance & Governance for AI (14%)

xdev.asia

1. Exam Overview — AIF-C01

DetailValue
Exam codeAIF-C01
Questions65 scored + 15 unscored = 80 total
Duration90 minutes
Passing score700 / 1000
Cost$100 USD
FormatMultiple choice + Multiple response
DeliveryPearson VUE (testing center or online proctored)
Validity3 years

2. Domain Weights

Domain 1: Fundamentals of AI and ML ████████████ 20%  (~13 questions)
Domain 2: Fundamentals of Generative AI █████████████████ 24%  (~16 questions)
Domain 3: Applications of Foundation Models ████████████████████ 28%  (~18 questions)
Domain 4: Guidelines for Responsible AI █████████ 14%  (~9 questions)
Domain 5: Security, Compliance & Governance █████████ 14%  (~9 questions)

Strategy: Domain 3 (28%) is the largest — focus most on Bedrock, RAG, Prompt Engineering, Fine-tuning. Domains 1+2 (44%) are foundational — understanding the concepts is sufficient.

3. Time Management Strategy

80 questions / 90 minutes = ~67 seconds per question

Strategy:
┌─────────────────────────────────────────────┐
│ Pass 1 (0-60 min): Answer easy questions    │
│   → Skip difficult ones (flag for review)   │
│   → Target: 60+ questions answered          │
├─────────────────────────────────────────────┤
│ Pass 2 (60-85 min): Review flagged questions│
│   → Eliminate wrong answers first           │
│   → Use process of elimination              │
├─────────────────────────────────────────────┤
│ Pass 3 (85-90 min): Final review            │
│   → Never leave questions blank             │
│   → No penalty for guessing                 │
└─────────────────────────────────────────────┘

4. Domain 1 Cheat Sheet — AI/ML Fundamentals

ConceptKey Points
AI vs ML vs DLAI ⊃ ML ⊃ DL. DL uses neural networks.
SupervisedLabeled data → Classification (discrete) or Regression (continuous)
UnsupervisedUnlabeled data → Clustering, dimensionality reduction
ReinforcementAgent + Environment + Rewards → Learn by trial and error
OverfittingGood on training, bad on test → More data, regularization, dropout
UnderfittingBad on both → More complex model, more features, longer training
PrecisionOf predicted positives, how many are correct? (avoid false positives)
RecallOf actual positives, how many did we find? (avoid false negatives)
F1 ScoreHarmonic mean of Precision and Recall
AUC-ROCModel's ability to distinguish classes (higher = better)

5. Domain 2 Cheat Sheet — Generative AI

ConceptKey Points
Foundation ModelsLarge models trained on broad data, adaptable to many tasks
TransformerEncoder-only (BERT, classify), Decoder-only (GPT, generate), Both (T5, translate)
Tokens~4 chars = 1 token. Context window = max tokens. Pricing = per token.
TemperatureLow (0) = deterministic/factual. High (1) = creative/diverse.
Top-pNucleus sampling. 0.1 = focused. 0.9 = diverse.
HallucinationModel generates false info. Mitigate with: RAG, guardrails, lower temp.
EmbeddingsText → vectors. Similar meaning = close vectors. Used for search, RAG.
Diffusion modelsImage gen by denoising. Stable Diffusion, Titan Image Generator.
Multi-modalProcess multiple data types. Claude 3 (text+image), Titan MM Embeddings.

6. Domain 3 Cheat Sheet — Applications

ConceptKey Points
Zero-shotNo examples. Model uses its knowledge.
Few-shot2-5 examples. Pattern following.
Chain-of-Thought"Think step by step" → better reasoning/math.
RAGRetrieve docs → augment prompt → generate. Reduces hallucination.
Bedrock Knowledge BasesManaged RAG: S3 docs → auto chunk → embed → vector store → retrieve.
Fine-tuningFurther train FM with your data. Need 1000+ labeled examples. LoRA = cheaper.
Bedrock AgentsMulti-step tasks: plan → call APIs → return results. Uses action groups.
Bedrock GuardrailsContent filter, denied topics, PII filter, word filter, grounding check.
PartyRockFree, no-code Bedrock playground. No AWS account needed.
Amazon Q DeveloperAI code assistant (was CodeWhisperer). IDE integration.
Amazon Q BusinessEnterprise Q&A on company data. Respects ACLs.

7. Domain 4 Cheat Sheet — Responsible AI

ConceptKey Points
Bias typesSelection, sampling, measurement, label, algorithmic, confirmation.
SageMaker ClarifyPre-training bias + post-training bias + SHAP explainability.
Amazon A2IHuman-in-the-loop. Trigger human review when confidence low.
SHAPFeature importance per prediction. "Why was this denied?"
AI Service CardsAWS public docs: intended use, limitations, fairness for AWS AI services.
Model CardsSageMaker: document YOUR model's details, metrics, ethics.
Toxicity detectionGuardrails content filters + Amazon Comprehend toxicity.

8. Domain 5 Cheat Sheet — Security & Compliance

ConceptKey Points
IAMLeast privilege. bedrock:InvokeModel + resource ARN = restrict models.
Encryption at restKMS keys for Bedrock, SageMaker, S3. Customer-managed for compliance.
VPC endpointsPrivateLink = no internet. Bedrock + SageMaker support this.
PII detectionComprehend (text), Macie (S3), Guardrails (FM I/O).
CloudTrailWho called what API when. Audit trail.
Model Invocation LoggingLog actual prompts & responses (to S3/CloudWatch).
Bedrock data privacyYour data NOT used to train base FMs. Stays in region. Encrypted.
Shared ResponsibilityCustomer: data, bias, IAM, compliance. AWS: infra, service security.

9. "Which AWS Service?" Quick Reference

If you need to...Use this service
Access multiple FMs via single APIAmazon Bedrock
Build RAG with company docsBedrock Knowledge Bases
Create AI that calls APIsBedrock Agents
Block harmful/PII contentBedrock Guardrails
Fine-tune FMBedrock Custom Models or SageMaker
Train custom ML modelAmazon SageMaker
Auto-ML (no code)SageMaker Canvas or SageMaker Autopilot
Detect bias in modelSageMaker Clarify
Human review for AI predictionsAmazon A2I
Extract text from documentsAmazon Textract
Analyze images/videoAmazon Rekognition
Translate textAmazon Translate
Speech-to-textAmazon Transcribe
Text-to-speechAmazon Polly
NLP (sentiment, entities)Amazon Comprehend
Build chatbotAmazon Lex
Enterprise searchAmazon Kendra
RecommendationsAmazon Personalize
Time-series forecastingAmazon Forecast
Fraud detectionAmazon Fraud Detector
Code assistantAmazon Q Developer
Enterprise Q&A on company dataAmazon Q Business
No-code GenAI playgroundPartyRock
Find PII in S3Amazon Macie

10. Common Exam Traps

Watch out for these:

  1. "MOST suitable" / "BEST": Multiple answers may work — choose the one that best fits ALL constraints.
  2. RAG vs Fine-tuning: If question mentions "latest data" or "company docs" → RAG. If "specific style" or "writing pattern" → Fine-tuning.
  3. System prompt vs Guardrails: If question says "guarantee" or "enforce" → Guardrails (system prompts can be bypassed).
  4. Supervised vs Unsupervised: "Labels" = supervised. "No labels" = unsupervised. "Reward" = reinforcement.
  5. SageMaker vs Bedrock: Custom ML model from scratch → SageMaker. Use/customize existing FM → Bedrock.
  6. Comprehend vs Macie: PII in text (runtime) → Comprehend. PII in S3 files (discovery) → Macie.
  7. Clarify vs Guardrails: Bias detection → Clarify. Content safety → Guardrails.
  8. Temperature: "Consistent answers" → Low temperature. "Creative responses" → High temperature.
  9. "Minimum effort" / "least overhead": Usually means managed service (Bedrock > SageMaker).
  10. CloudTrail vs CloudWatch vs Invocation Logging: API audit → CloudTrail. Metrics → CloudWatch. Prompt/response logs → Invocation Logging.

11. Exam Day Tips

  • Read the ENTIRE question before looking at answers
  • Identify keywords: "MOST", "LEAST", "BEST", "FIRST", "guarantee"
  • Eliminate wrong answers first — usually 2 are obviously wrong
  • Don't overthink — if you know the answer, select it and move on
  • Flag and skip hard questions — come back in Pass 2
  • Never leave blank — no penalty for guessing
  • 15 unscored questions — you don't know which ones, so answer all seriously
  • Manage time — 67 seconds per question average

12. Next Steps

  1. Review all 11 lessons in this series
  2. Take the mock exam in the "Practice Exam" section
  3. AWS Skill Builder: Free "Exam Prep" course for AIF-C01
  4. AWS PartnerCast: Additional exam prep resources
  5. Practice with Bedrock: Use AWS Free Tier / PartyRock
  6. Schedule the exam at aws.amazon.com/certification

Remember: You don't need hands-on experience — this is a Practitioner level exam. Focus on concepts, use cases, and choosing the right AWS service.