AI use cases and applicationsFull notesSummaryRingkasanStoriesPracticeChallenges of Generative AI8 exam-style questions on this lesson.Question 1 of 8A user sends the same prompt to a generative AI model three times and gets three different answers. What is this called?AHallucinationBNondeterminismCInterpretabilityDToxicityCheck answerQuestion 2 of 8How can a team reduce nondeterminism in a model's answers?ARemove all guardrailsBRaise the temperatureCLower the temperature, test repeatedly, and validate outputsDAdd more toxic examples to the training dataCheck answerQuestion 3 of 8A chatbot confidently cites a research paper that does not exist. What is this problem, and how is it best mitigated?ANondeterminism, mitigated by raising the temperature for every requestBToxicity, mitigated by switching to a larger foundation modelCRegulatory violation, mitigated by encrypting the model weights at restDHallucination, mitigated by grounding answers with RAG and human reviewCheck answerQuestion 4 of 8. Select two.A model sometimes produces offensive language. Which TWO mitigations fit? (Select TWO.)ARaise the temperatureBTurn off loggingCCurate the training dataDUse guardrails to filter contentEIncrease the maximum output lengthCheck answer0 of 2 selectedQuestion 5 of 8A model's output reveals customers' personally identifiable information (PII), which breaks privacy law. Which challenge is this, and what is a mitigation?ARegulatory violation; anonymize data and audit training dataBNondeterminism; test repeatedly and lower temperatureCSocial risk; set content policies and monitor useDInterpretability; document model behaviorCheck answerQuestion 6 of 8A bank's regulators ask why the model rejected a loan, and the team can't explain the decision. Which challenge is this?AData efficiencyBInterpretabilityCToxicityDHallucinationCheck answerQuestion 7 of 8Sensitive business data could leak through the prompts employees send or the answers the model gives. Which mitigations fit?ARaise the temperature and shorten every promptBPublish all prompts so users can review themCEncrypt data, control access, and filter inputs and outputsDUse a bigger model with stronger built-in privacyCheck answerQuestion 8 of 8A company's AI-generated marketing content damages its reputation with the public. Which challenge is this, and what helps?AInterpretability; explain each output with SHAP valuesBHallucination; ground each answer with RAG sourcesCNondeterminism; lower the temperature and retestDSocial risk; test outputs, set content policies, and monitor useCheck answerCapabilities of Generative AI7 exam-style questions on this lesson.Factors to Consider When Selecting a Generative AI Model7 exam-style questions on this lesson.