AIF-C01 cheat sheet
Last-minute reference for AIF-C01. A foundational exam: match the business scenario to the right concept.
Exam facts
| Duration | 90 minutes |
| Questions | 65 (50 scored) |
| Passing score | 700 / 1000 |
| Price | 100 USD |
Domain weights
| Domain | Weight |
|---|---|
| Applications of Foundation Models | 28% |
| Fundamentals of GenAI | 24% |
| Fundamentals of AI and ML | 20% |
| Guidelines for Responsible AI | 14% |
| Security, Compliance, and Governance | 14% |
Responsible AI plus governance is 28% combined — as much as the largest single domain, and the most under-studied part.
Concept pairs
| A | B | When A wins |
|---|---|---|
| RAG | Fine-tuning | Content changes; no training budget |
| Few-shot | Zero-shot | You can supply worked examples |
| Groundedness | Fluency | Asking whether the answer is supported |
| Smaller model | Larger model | Narrow task, cost and latency matter |
| Regression | Classification | Predicting a number, not a category |
| Anomaly detection | Classification | No labelled examples of the bad class |
Key definitions
- Foundation model — broad training, adapted to many tasks
- Embedding — vector capturing meaning, enabling similarity search
- Hallucination — fluent but unsupported output
- Overfitting — great on training data, poor on unseen data
- Chunking — oversized chunks dilute embeddings and wreck retrieval
Recurring right answers
- Index and retrieve at query time when the manual changes monthly
- Reduce chunk size and add overlap for topically related but useless results
- Human confirms before the write executes for a first agent release
- Redact or tokenise personal data before the prompt leaves your boundary
- Model documentation with data lineage and deployment approval for auditors
- Untrusted web content can carry prompt injection
Traps
- This is not a maths exam. No gradient descent, no hyperparameter tuning.
- No deep service configuration. Instance types and IAM syntax belong to MLA-C02.
- Do not use ML Specialty material. Wrong level entirely.
Night-before checklist
- Five domains and their weights
- Responsible AI and governance are 28% together
- RAG versus fine-tuning, cold
- Check ID and proctoring rules — see exam day
Take the 20-question practice test.