AI-901 cheat sheet
Last-minute reference for AI-901. This is a fundamentals exam: it tests whether you can match a scenario to the right capability, not whether you can build one.
Exam facts
| Level | Fundamentals |
| Passing score | 700 / 1000 |
| Price | Set per country or region; shown at booking |
| Replaces | AI-900 |
Skill areas
| Area | Weight |
|---|---|
| Implement AI solutions by using Microsoft Foundry | 55–60% |
| Identify AI concepts and capabilities | 40–45% |
Only two areas, and the Foundry one is the bigger half.
Workload types
| Scenario | Workload |
|---|---|
| Predict a number from history | Regression |
| Predict a category from labelled data | Classification |
| Find groups with no labels | Clustering |
| Flag unusual behaviour, no labelled examples | Anomaly detection |
| Count objects in images | Computer vision |
| Analyse review sentiment | Natural language processing |
Responsible AI principles
Fairness · Reliability and safety · Privacy and security · Inclusiveness · Transparency · Accountability
| Situation | Principle |
|---|---|
| Model disadvantages a group | Fairness |
| Must explain an individual decision | Transparency |
| Someone must answer for outcomes | Accountability |
Key terms
- Foundation model — broad training, adapted to many tasks
- Context window — text considered in one request, input and output, in tokens
- Hallucination — fluent but unsupported output
- Few-shot prompting — examples inside the prompt; no retraining
- RAG — retrieve documents at query time so answers stay current
- Model drift — accuracy declining as the world moves away from training data
Recurring right answers
- RAG, not monthly fine-tuning, when source content changes
- Lower temperature for consistent factual answers
- Content safety filters on prompts and completions, not prompt wording
- Managed identity, not a key in code
- Both input and output tokens are billed
The main trap: over-preparation
You do not need transformer architecture, Python, or service pricing. Candidates lose weeks studying at associate depth. If you score sixteen or more on the practice test cold, book it.
Night-before checklist
- Two skill areas, Foundry is the larger
- Six responsible AI principles
- Be able to pick the workload type from a one-line scenario
- Check ID and proctoring rules — see exam day