100% Latest & Updated Google Generative AI Leader Practice Test Questions, Exam Dumps & Verified Answers!
30 Days Free Updates, Instant Download!
Question 550
A lending model's old features stop predicting defaults after regulations change eligibility. What should be reassessed?
- Concept drift assessment
- Knowledge cutoff
- Model versioning
- Live evidence retrieval
Explanation
Correct Answer
A:The underlying target relationship may have shifted. Concept drift examines whether the relationship between input and expected outcome has changed.
Incorrect Answers
B:Knowledge cutoff limits what can be learned from data available during model training. This does not directly meet the case requirement to investigate concept drift in prediction behavior.
C:Versioning tracks which model revision is running and helps compare releases. This does not directly meet the case requirement to investigate concept drift in prediction behavior.
D:Live retrieval accesses fresh authoritative sources at request time. This does not directly meet the case requirement to investigate concept drift in prediction behavior.
Question 551
A claims assistant succeeds on common receipts but fails on rare handwritten invoices. What targeted test should be added?
- Adversarial robustness testing
- Edge-case test suite
- Representative benchmark set
- User feedback capture
Explanation
Correct Answer
B:Rare but valid cases must be represented in evaluation. An edge-case suite exercises unusual but plausible inputs outside typical examples.
Incorrect Answers
A:Robustness tests evaluate behavior under confusing or intentionally challenging inputs. This does not directly meet the case requirement to exercise unusual document layouts.
C:A benchmark set tests broad expected cases and key quality measures. This does not directly meet the case requirement to exercise unusual document layouts.
D:Feedback collection records real user reports of poor or harmful outcomes. This does not directly meet the case requirement to exercise unusual document layouts.
Question 552
An agent that reads customer documents is deliberately attacked with malicious instructions embedded inside a supposedly informative document. Which security testing approach should challenge this hostile input boundary before launch?
- Error taxonomy analysis
- Edge-case test suite
- Fallback behavior
- Adversarial robustness testing
Explanation
Correct Answer
D:Adversarial tests check resistance to manipulative inputs. Robustness tests evaluate behavior under confusing or intentionally challenging inputs.
Incorrect Answers
A:An error taxonomy groups failure modes to guide corrective action. This does not directly meet the case requirement to test behavior under hostile instructions in data.
B:An edge-case suite exercises unusual but plausible inputs outside typical examples. This does not directly meet the case requirement to test behavior under hostile instructions in data.
C:A fallback safely handles unsupported situations or uncertain results. This does not directly meet the case requirement to test behavior under hostile instructions in data.
Question 553
A product team wants a balanced baseline of ordinary tasks before comparing two models. Which artifact should it create?
- Fallback behavior
- User feedback capture
- Error taxonomy analysis
- Representative benchmark set
Explanation
Correct Answer
D:Representative benchmarks assess typical expected use. A benchmark set tests broad expected cases and key quality measures.
Incorrect Answers
A:A fallback safely handles unsupported situations or uncertain results. This does not directly meet the case requirement to measure performance across standard use cases.
B:Feedback collection records real user reports of poor or harmful outcomes. This does not directly meet the case requirement to measure performance across standard use cases.
C:An error taxonomy groups failure modes to guide corrective action. This does not directly meet the case requirement to measure performance across standard use cases.
Question 554
A team records errors such as wrong citations, format violations and missing data in distinct categories. What activity is this?
- Representative benchmark set
- Error taxonomy analysis
- User feedback capture
- Edge-case test suite
Explanation
Correct Answer
B:Error classification helps identify systematic issues. An error taxonomy groups failure modes to guide corrective action.
Incorrect Answers
A:A benchmark set tests broad expected cases and key quality measures. This does not directly meet the case requirement to organize failures by underlying type.
C:Feedback collection records real user reports of poor or harmful outcomes. This does not directly meet the case requirement to organize failures by underlying type.
D:An edge-case suite exercises unusual but plausible inputs outside typical examples. This does not directly meet the case requirement to organize failures by underlying type.
Question 555
An assistant receives a file format it cannot process. What runtime behavior is safer than inventing a summary?
- User feedback capture
- Representative benchmark set
- Fallback behavior
- Error taxonomy analysis
Explanation
Correct Answer
C:Fallback responses disclose limitations and permit recovery. A fallback safely handles unsupported situations or uncertain results.
Incorrect Answers
A:Feedback collection records real user reports of poor or harmful outcomes. This does not directly meet the case requirement to handle unsupported input safely.
B:A benchmark set tests broad expected cases and key quality measures. This does not directly meet the case requirement to handle unsupported input safely.
D:An error taxonomy groups failure modes to guide corrective action. This does not directly meet the case requirement to handle unsupported input safely.
Question 556
A model passes lab tests but customers report repeated misleading replies. What feedback mechanism should exist?
- Representative benchmark set
- User feedback capture
- Edge-case test suite
- Error taxonomy analysis
Explanation
Correct Answer
B:Collecting user reports captures unanticipated production failures. Feedback collection records real user reports of poor or harmful outcomes.
Incorrect Answers
A:A benchmark set tests broad expected cases and key quality measures. This does not directly meet the case requirement to receive and triage model error reports.
C:An edge-case suite exercises unusual but plausible inputs outside typical examples. This does not directly meet the case requirement to receive and triage model error reports.
D:An error taxonomy groups failure modes to guide corrective action. This does not directly meet the case requirement to receive and triage model error reports.
Question 557
An OCR assistant cannot handle a rotated page with handwritten notes. What test case should be included?
- Adversarial robustness testing
- User feedback capture
- Error taxonomy analysis
- Edge-case test suite
Explanation
Correct Answer
D:Unusual orientations challenge normal-case accuracy. An edge-case suite exercises unusual but plausible inputs outside typical examples.
Incorrect Answers
A:Robustness tests evaluate behavior under confusing or intentionally challenging inputs. This does not directly meet the case requirement to test atypical document presentations.
B:Feedback collection records real user reports of poor or harmful outcomes. This does not directly meet the case requirement to test atypical document presentations.
C:An error taxonomy groups failure modes to guide corrective action. This does not directly meet the case requirement to test atypical document presentations.
Question 558
A support bot is asked to ignore all rules by text inside a webpage. Which test category targets this threat?
- Representative benchmark set
- Adversarial robustness testing
- Edge-case test suite
- User feedback capture
Explanation
Correct Answer
B:Robustness tests can simulate prompt injection. Robustness tests evaluate behavior under confusing or intentionally challenging inputs.
Incorrect Answers
A:A benchmark set tests broad expected cases and key quality measures. This does not directly meet the case requirement to assess resistance to instruction manipulation.
C:An edge-case suite exercises unusual but plausible inputs outside typical examples. This does not directly meet the case requirement to assess resistance to instruction manipulation.
D:Feedback collection records real user reports of poor or harmful outcomes. This does not directly meet the case requirement to assess resistance to instruction manipulation.
Question 559
A model fails in three different ways, and the team wants to know which defect family to fix first. What should it produce?
- User feedback capture
- Edge-case test suite
- Error taxonomy analysis
- Fallback behavior
Explanation
Correct Answer
C:An error taxonomy enables prioritization by cause. An error taxonomy groups failure modes to guide corrective action.
Incorrect Answers
A:Feedback collection records real user reports of poor or harmful outcomes. This does not directly meet the case requirement to classify distinct failure mechanisms.
B:An edge-case suite exercises unusual but plausible inputs outside typical examples. This does not directly meet the case requirement to classify distinct failure mechanisms.
D:A fallback safely handles unsupported situations or uncertain results. This does not directly meet the case requirement to classify distinct failure mechanisms.
Question 560
A document summarizer lacks permission to read the requested file. Which behavior should the design support?
- Error taxonomy analysis
- Adversarial robustness testing
- User feedback capture
- Fallback behavior
Explanation
Correct Answer
D:A safe fallback should acknowledge access limitations. A fallback safely handles unsupported situations or uncertain results.
Incorrect Answers
A:An error taxonomy groups failure modes to guide corrective action. This does not directly meet the case requirement to decline to invent content from inaccessible material.
B:Robustness tests evaluate behavior under confusing or intentionally challenging inputs. This does not directly meet the case requirement to decline to invent content from inaccessible material.
C:Feedback collection records real user reports of poor or harmful outcomes. This does not directly meet the case requirement to decline to invent content from inaccessible material.
Generative AI Leader Practice Tests
