100% Latest & Updated Google Generative AI Leader Practice Test Questions, Exam Dumps & Verified Answers!
30 Days Free Updates, Instant Download!
Question 40
An AI team uses millions of unannotated sentences to learn language representations by reconstructing missing words. Which training signal is used?
- Inference
- Self-supervised pretraining
- Reinforcement learning
- Human evaluation
Explanation
Correct Answer
B:Automatically masked text supplies self-generated objectives. Self-supervised learning derives training targets from the input data itself without manually annotated labels.
Incorrect Answers
A:Inference runs an already trained model on new input without updating its training weights. This does not directly meet the case requirement to learn representations without manual annotation.
C:Reinforcement learning adapts a policy through actions and reward signals over time. This does not directly meet the case requirement to learn representations without manual annotation.
D:Human review judges or scores outputs but is not by itself a model-training regime. This does not directly meet the case requirement to learn representations without manual annotation.
Question 41
An analytics pipeline begins pulling transaction logs into a governed storage area. Which lifecycle activity is this?
- Data ingestion
- Model training
- Model monitoring
- Model evaluation
Explanation
Correct Answer
A:The data is being collected before it can be processed. Ingestion gathers data from source systems and transports it into an ML workflow.
Incorrect Answers
B:Training adjusts model parameters to fit patterns in prepared examples. This does not directly meet the case requirement to collect source data for model development.
C:Monitoring observes quality, latency, costs, drift and safety after deployment. This does not directly meet the case requirement to collect source data for model development.
D:Evaluation tests whether the candidate meets relevant accuracy, safety and quality criteria. This does not directly meet the case requirement to collect source data for model development.
Question 42
Before training, a team deduplicates customer records and fixes missing values. Which lifecycle stage is this?
- Model evaluation
- Model monitoring
- Data preparation
- Model training
Explanation
Correct Answer
C:The dataset is being improved prior to training. Preparation cleans, transforms, labels or splits data before modeling.
Incorrect Answers
A:Evaluation tests whether the candidate meets relevant accuracy, safety and quality criteria. This does not directly meet the case requirement to clean and prepare examples.
B:Monitoring observes quality, latency, costs, drift and safety after deployment. This does not directly meet the case requirement to clean and prepare examples.
D:Training adjusts model parameters to fit patterns in prepared examples. This does not directly meet the case requirement to clean and prepare examples.
Question 43
A team updates model weights using examples and an optimization algorithm. Which stage is being executed?
- Data ingestion
- Model monitoring
- Data preparation
- Model training
Explanation
Correct Answer
D:Parameter learning happens during model training. Training adjusts model parameters to fit patterns in prepared examples.
Incorrect Answers
A:Ingestion gathers data from source systems and transports it into an ML workflow. This does not directly meet the case requirement to fit model parameters using data.
B:Monitoring observes quality, latency, costs, drift and safety after deployment. This does not directly meet the case requirement to fit model parameters using data.
C:Preparation cleans, transforms, labels or splits data before modeling. This does not directly meet the case requirement to fit model parameters using data.
Question 44
A team tests a candidate model on held-out cases for bias and factuality before release. Which activity does this represent?
- Model evaluation
- Data preparation
- Model training
- Model monitoring
Explanation
Correct Answer
A:Assessing readiness on test cases is evaluation. Evaluation tests whether the candidate meets relevant accuracy, safety and quality criteria.
Incorrect Answers
B:Preparation cleans, transforms, labels or splits data before modeling. This does not directly meet the case requirement to measure candidate quality before deployment.
C:Training adjusts model parameters to fit patterns in prepared examples. This does not directly meet the case requirement to measure candidate quality before deployment.
D:Monitoring observes quality, latency, costs, drift and safety after deployment. This does not directly meet the case requirement to measure candidate quality before deployment.
Question 45
A proven model is exposed behind a managed endpoint so employees can call it. Which stage is this?
- Data preparation
- Model evaluation
- Model deployment
- Data ingestion
Explanation
Correct Answer
C:The model is being served for production inference. Deployment makes a trained model available for application use or inference.
Incorrect Answers
A:Preparation cleans, transforms, labels or splits data before modeling. This does not directly meet the case requirement to make the trained model accessible to applications.
B:Evaluation tests whether the candidate meets relevant accuracy, safety and quality criteria. This does not directly meet the case requirement to make the trained model accessible to applications.
D:Ingestion gathers data from source systems and transports it into an ML workflow. This does not directly meet the case requirement to make the trained model accessible to applications.
Question 46
After launch, analysts track response latency and harmful-output reports every week. Which stage is active?
- Model monitoring
- Model training
- Model deployment
- Data preparation
Explanation
Correct Answer
A:Operational monitoring identifies issues in production behavior. Monitoring observes quality, latency, costs, drift and safety after deployment.
Incorrect Answers
B:Training adjusts model parameters to fit patterns in prepared examples. This does not directly meet the case requirement to track deployed quality and performance over time.
C:Deployment makes a trained model available for application use or inference. This does not directly meet the case requirement to track deployed quality and performance over time.
D:Preparation cleans, transforms, labels or splits data before modeling. This does not directly meet the case requirement to track deployed quality and performance over time.
Question 47
An engineer separates customer rows into training and test sets to avoid evaluating on seen examples. Which stage owns this split?
- Model monitoring
- Model deployment
- Data preparation
- Model evaluation
Explanation
Correct Answer
C:Dataset splitting is preprocessing for reliable evaluation. Preparation cleans, transforms, labels or splits data before modeling.
Incorrect Answers
A:Monitoring observes quality, latency, costs, drift and safety after deployment. This does not directly meet the case requirement to prepare independent train and evaluation data.
B:Deployment makes a trained model available for application use or inference. This does not directly meet the case requirement to prepare independent train and evaluation data.
D:Evaluation tests whether the candidate meets relevant accuracy, safety and quality criteria. This does not directly meet the case requirement to prepare independent train and evaluation data.
Question 48
Two candidate models are compared against a business acceptance threshold before traffic is switched. What stage informs the release decision?
- Data ingestion
- Data preparation
- Model evaluation
- Model training
Explanation
Correct Answer
C:A deployment decision depends on test metrics and validation. Evaluation tests whether the candidate meets relevant accuracy, safety and quality criteria.
Incorrect Answers
A:Ingestion gathers data from source systems and transports it into an ML workflow. This does not directly meet the case requirement to compare models against readiness thresholds.
B:Preparation cleans, transforms, labels or splits data before modeling. This does not directly meet the case requirement to compare models against readiness thresholds.
D:Training adjusts model parameters to fit patterns in prepared examples. This does not directly meet the case requirement to compare models against readiness thresholds.
Question 49
An application team configures a scalable serving route for an approved model version. What activity is taking place?
- Model evaluation
- Model deployment
- Data preparation
- Model monitoring
Explanation
Correct Answer
B:Serving integration makes a candidate available for requests. Deployment makes a trained model available for application use or inference.
Incorrect Answers
A:Evaluation tests whether the candidate meets relevant accuracy, safety and quality criteria. This does not directly meet the case requirement to publish a model endpoint for clients.
C:Preparation cleans, transforms, labels or splits data before modeling. This does not directly meet the case requirement to publish a model endpoint for clients.
D:Monitoring observes quality, latency, costs, drift and safety after deployment. This does not directly meet the case requirement to publish a model endpoint for clients.
Question 50
A model was accurate at launch but now fails on newer customer language. Which lifecycle activity should detect this?
- Model training
- Model evaluation
- Model deployment
- Model monitoring
Explanation
Correct Answer
D:Drift is a post-deployment observation requiring ongoing measurement. Monitoring observes quality, latency, costs, drift and safety after deployment.
Incorrect Answers
A:Training adjusts model parameters to fit patterns in prepared examples. This does not directly meet the case requirement to detect performance changes after launch.
B:Evaluation tests whether the candidate meets relevant accuracy, safety and quality criteria. This does not directly meet the case requirement to detect performance changes after launch.
C:Deployment makes a trained model available for application use or inference. This does not directly meet the case requirement to detect performance changes after launch.
Generative AI Leader Practice Tests
