Exam Quick Reference
Key services, common traps, domain weightings, practice questions
The AWS Certified AI Practitioner (AIF-C01) validates foundational knowledge of AI/ML concepts and AWS AI services. It is a non-technical exam — no coding required. Focus on: which service solves which problem, responsible AI principles, and understanding generative AI terminology.
Key Points
- Exam format: 65 questions, 90 minutes, multiple choice and multiple response
- Passing score: 700/1000 (scaled)
- Domain 1 (20%): Fundamentals of AI and ML
- Domain 2 (24%): Fundamentals of Generative AI
- Domain 3 (28%): Applications of Foundation Models
- Domain 4 (14%): Guidelines for Responsible AI
- Domain 5 (14%): Security, Compliance and Governance for AI Solutions
| Scenario | Correct Service | Why Not Others |
|---|---|---|
| Extract tables from invoices | Amazon Textract | Rekognition = labels/faces; Comprehend = NLP on text |
| Build a customer chatbot | Amazon Lex | Comprehend = NLP analysis; Kendra = search, not conversation |
| Convert audio recordings to text | Amazon Transcribe | Polly = text→speech (opposite direction) |
| Personalised product recs | Amazon Personalize | Forecast = time series, not recommendations |
| Access Claude 3 on AWS | Amazon Bedrock | SageMaker = custom models; Bedrock = FMs via API |
| Detect bias in training data | SageMaker Clarify | Macie = PII; Comprehend = NLP; Clarify = bias |
| Human review of AI output | Amazon A2I | A2I adds human-in-the-loop; SageMaker = model hosting |
| Enterprise knowledge search | Amazon Kendra | Comprehend = text NLP; Kendra = indexed search |
| Detect PII in S3 data lake | Amazon Macie | Comprehend PII = in text; Macie = in S3 objects at scale |
| Fine-tune a foundation model | Amazon Bedrock / SageMaker JumpStart | Direct fine-tuning on managed FMs |
Quick-reference cheat sheet for the AI Practitioner exam
# Key Generative AI Terminology for the Exam
TOKENISATION
- Text split into subword units (tokens)
- ~4 chars per token in English
- GPT-4 context: 128K tokens ≈ ~96K words
INFERENCE PARAMETERS
Temperature = 0 → deterministic (factual tasks)
Temperature = 0.7 → balanced
Temperature = 1.0 → creative
Top-P = 0.9 → sample from top 90% probability mass
Top-K = 50 → pick from top 50 candidate tokens
HALLUCINATION
- Model generates plausible-but-false information
- Mitigation: RAG (ground in documents), Guardrails grounding check
RAG vs FINE-TUNING
RAG → dynamic knowledge, no retraining, cheaper
Fine-tune → behaviour/style/tone changes, more expensive
BEDROCK vs SAGEMAKER
Bedrock → pre-built FMs via API (GenAI)
SageMaker → custom model training & hosting (all ML)
RESPONSIBLE AI PILLARS (AWS)
1. Fairness 2. Explainability
3. Privacy & Security 4. Robustness
5. Governance 6. TransparencyReal-World Example
The AWS AI Practitioner exam launched in September 2024 and quickly became one of the fastest-growing AWS certifications. It is ideal for solution architects, product managers, and technical leads who work with AI teams but do not build models themselves.