AI & Machine Learning Practice Questions
Metrics, MLOps, serving and generative AI techniques. 7 original questions with full explanations.
Question 1 of 7
0 answered · 7 remaining
AI & Machine Learningmedium
A support assistant built on a foundation model gives confident but incorrect answers about the company's own refund policy, which is documented internally. What is the appropriate fix?
Select an answer to see the explanation.
These questions are written by GCP Prep to test understanding of publicly documented concepts. They are not real exam questions.
What this test covers
One concept per question. If you can explain each of these without hesitating, you are in good shape on AI & Machine Learning.
- Missing knowledge is fixed by retrieval, not by tuning or prompting
- Hallucination is the characteristic failure mode of generative models
- Gradual degradation with unchanged code points to drift
- Batch prediction suits large offline scoring jobs where per-record latency is irrelevant
- Metric choice must reflect the cost of each error type on imbalanced data
- Fine-tuning changes systematic behaviour that prompting cannot reliably achieve
- Training–serving skew arises when the same feature is computed two different ways
Other topics
Related Certifications
- Generative AI LeaderBusiness-level fluency in generative AI, without writing code.
- Professional Machine Learning EngineerProduction machine learning: pipelines, serving and monitoring.
- Cloud Digital LeaderThe business-literacy exam. No hands-on experience required.
- Professional Data EngineerPipelines, warehouses and the systems analytics runs on.
Related Cloud Topics
- Generative AIFoundation models, prompting, grounding, retrieval augmentation and evaluation — and choosing the right technique for a failure.
- Vertex AI and the ML PlatformTraining, serving, pipelines and monitoring in one place — the MLOps surface that production machine learning needs.
- Machine Learning FundamentalsThe concepts every ML question assumes: training, evaluation metrics, overfitting, and knowing when not to use machine learning.