100% Latest & Updated Google Generative AI Leader Practice Test Questions, Exam Dumps & Verified Answers!
30 Days Free Updates, Instant Download!
Question 70
A public-sector contract restricts data access to named administrators and approved regions. Which criterion should dominate?
- Quality benchmark
- Modality compatibility
- Security and data residency
- Context-window suitability
Explanation
Correct Answer
C:Governed access and approved processing boundaries are non-negotiable. Security and residency determine whether data handling and locations satisfy obligations.
Incorrect Answers
A:Quality benchmarks measure task performance on representative cases rather than relying on popularity. This does not directly meet the case requirement to enforce contracted data security controls.
B:Modality compatibility concerns whether the model accepts and produces required text, audio, image or video types. This does not directly meet the case requirement to enforce contracted data security controls.
D:Context-window suitability concerns how much input and dialogue the model can consider at once. This does not directly meet the case requirement to enforce contracted data security controls.
Question 71
A writing assistant needs to use a specific persona and answer format, with no training budget. What is the first low-effort technique?
- Fine-tuning
- Prompt engineering
- Retrieval-augmented generation (RAG)
- Training from scratch
Explanation
Correct Answer
B:Changing the prompt can steer output presentation without modifying weights. Prompt engineering changes instructions and examples at inference time without changing model weights.
Incorrect Answers
A:Fine-tuning updates pretrained model parameters using a targeted training dataset. This does not directly meet the case requirement to improve output instructions without retraining.
C:RAG fetches relevant external evidence at answer time so responses can use current approved information. This does not directly meet the case requirement to improve output instructions without retraining.
D:Training from scratch fits a new model starting without a pretrained foundation and usually needs far more data and compute. This does not directly meet the case requirement to improve output instructions without retraining.
Question 72
A policy bot must answer from the latest internal handbook that changes weekly. Which technique directly supplies updated evidence?
- Retrieval-augmented generation (RAG)
- Post-generation human review
- Prompt engineering
- Prompt tuning
Explanation
Correct Answer
A:Retrieval accesses current approved passages at response time. RAG fetches relevant external evidence at answer time so responses can use current approved information.
Incorrect Answers
B:Human review checks and corrects an output after generation rather than adjusting the model itself. This does not directly meet the case requirement to use the newest authorized policy text in answers.
C:Prompt engineering changes instructions and examples at inference time without changing model weights. This does not directly meet the case requirement to use the newest authorized policy text in answers.
D:Prompt tuning learns task-specific prompt embeddings or parameters rather than rewriting all foundation weights. This does not directly meet the case requirement to use the newest authorized policy text in answers.
Question 73
A domain model consistently mishandles an uncommon specialized writing style despite clear examples, and a quality dataset is available. Which adaptation can teach the behavior?
- Retrieval-augmented generation (RAG)
- Training from scratch
- Prompt tuning
- Fine-tuning
Explanation
Correct Answer
D:Targeted parameter updates may improve a recurring domain task. Fine-tuning updates pretrained model parameters using a targeted training dataset.
Incorrect Answers
A:RAG fetches relevant external evidence at answer time so responses can use current approved information. This does not directly meet the case requirement to adapt learned behavior with curated training examples.
B:Training from scratch fits a new model starting without a pretrained foundation and usually needs far more data and compute. This does not directly meet the case requirement to adapt learned behavior with curated training examples.
C:Prompt tuning learns task-specific prompt embeddings or parameters rather than rewriting all foundation weights. This does not directly meet the case requirement to adapt learned behavior with curated training examples.
Question 74
An optimization team wants to learn compact task-specific soft prompts for a pretrained model. Which method is described?
- Prompt engineering
- Post-generation human review
- Fine-tuning
- Prompt tuning
Explanation
Correct Answer
D:Learned prompt embeddings influence an existing model. Prompt tuning learns task-specific prompt embeddings or parameters rather than rewriting all foundation weights.
Incorrect Answers
A:Prompt engineering changes instructions and examples at inference time without changing model weights. This does not directly meet the case requirement to fit trainable prompts without a full foundation retrain.
B:Human review checks and corrects an output after generation rather than adjusting the model itself. This does not directly meet the case requirement to fit trainable prompts without a full foundation retrain.
C:Fine-tuning updates pretrained model parameters using a targeted training dataset. This does not directly meet the case requirement to fit trainable prompts without a full foundation retrain.
Question 75
A startup proposes building a new foundation language model with random weights for a small internal FAQ. Which approach is it considering?
- Fine-tuning
- Retrieval-augmented generation (RAG)
- Prompt tuning
- Training from scratch
Explanation
Correct Answer
D:Full pretraining is resource-intensive and not proportional to that narrow task. Training from scratch fits a new model starting without a pretrained foundation and usually needs far more data and compute.
Incorrect Answers
A:Fine-tuning updates pretrained model parameters using a targeted training dataset. This does not directly meet the case requirement to build an entirely new model rather than reuse one.
B:RAG fetches relevant external evidence at answer time so responses can use current approved information. This does not directly meet the case requirement to build an entirely new model rather than reuse one.
C:Prompt tuning learns task-specific prompt embeddings or parameters rather than rewriting all foundation weights. This does not directly meet the case requirement to build an entirely new model rather than reuse one.
Question 76
A medical assistant produces a draft that a clinician must approve before a patient sees it. What control is being used?
- Prompt tuning
- Post-generation human review
- Fine-tuning
- Prompt engineering
Explanation
Correct Answer
B:Expert approval provides a gate on generated outputs. Human review checks and corrects an output after generation rather than adjusting the model itself.
Incorrect Answers
A:Prompt tuning learns task-specific prompt embeddings or parameters rather than rewriting all foundation weights. This does not directly meet the case requirement to apply human approval to a sensitive answer.
C:Fine-tuning updates pretrained model parameters using a targeted training dataset. This does not directly meet the case requirement to apply human approval to a sensitive answer.
D:Prompt engineering changes instructions and examples at inference time without changing model weights. This does not directly meet the case requirement to apply human approval to a sensitive answer.
Question 77
A procurement team needs JSON fields in a fixed order and has not tried a clear schema instruction. Which improvement should come first?
- Prompt engineering
- Fine-tuning
- Prompt tuning
- Retrieval-augmented generation (RAG)
Explanation
Correct Answer
A:Prompt design can express the expected structure. Prompt engineering changes instructions and examples at inference time without changing model weights.
Incorrect Answers
B:Fine-tuning updates pretrained model parameters using a targeted training dataset. This does not directly meet the case requirement to specify structured output requirements without updating weights.
C:Prompt tuning learns task-specific prompt embeddings or parameters rather than rewriting all foundation weights. This does not directly meet the case requirement to specify structured output requirements without updating weights.
D:RAG fetches relevant external evidence at answer time so responses can use current approved information. This does not directly meet the case requirement to specify structured output requirements without updating weights.
Question 78
An assistant invents citations to company reports it has not accessed. Which technique supplies the underlying evidence?
- Training from scratch
- Retrieval-augmented generation (RAG)
- Prompt tuning
- Prompt engineering
Explanation
Correct Answer
B:RAG provides the actual report excerpts and links. RAG fetches relevant external evidence at answer time so responses can use current approved information.
Incorrect Answers
A:Training from scratch fits a new model starting without a pretrained foundation and usually needs far more data and compute. This does not directly meet the case requirement to cite genuine current company sources.
C:Prompt tuning learns task-specific prompt embeddings or parameters rather than rewriting all foundation weights. This does not directly meet the case requirement to cite genuine current company sources.
D:Prompt engineering changes instructions and examples at inference time without changing model weights. This does not directly meet the case requirement to cite genuine current company sources.
Question 79
A firm has verified examples of its own legal classification language and seeks persistent adaptation across many calls. What option should it evaluate?
- Post-generation human review
- Prompt engineering
- Retrieval-augmented generation (RAG)
- Fine-tuning
Explanation
Correct Answer
D:Fine-tuning can encode task-specific patterns when retrieval alone is insufficient. Fine-tuning updates pretrained model parameters using a targeted training dataset.
Incorrect Answers
A:Human review checks and corrects an output after generation rather than adjusting the model itself. This does not directly meet the case requirement to persistently adapt model behavior to a domain dataset.
B:Prompt engineering changes instructions and examples at inference time without changing model weights. This does not directly meet the case requirement to persistently adapt model behavior to a domain dataset.
C:RAG fetches relevant external evidence at answer time so responses can use current approved information. This does not directly meet the case requirement to persistently adapt model behavior to a domain dataset.
Question 80
A journalism organization requires an editor to check every AI-suggested headline before it is published. Which stage is the safeguard?
- Training from scratch
- Fine-tuning
- Post-generation human review
- Retrieval-augmented generation (RAG)
Explanation
Correct Answer
C:Humans review the final content before release. Human review checks and corrects an output after generation rather than adjusting the model itself.
Incorrect Answers
A:Training from scratch fits a new model starting without a pretrained foundation and usually needs far more data and compute. This does not directly meet the case requirement to prevent unchecked publication of generated text.
B:Fine-tuning updates pretrained model parameters using a targeted training dataset. This does not directly meet the case requirement to prevent unchecked publication of generated text.
D:RAG fetches relevant external evidence at answer time so responses can use current approved information. This does not directly meet the case requirement to prevent unchecked publication of generated text.
Generative AI Leader Practice Tests
