Responsible AI practicesFull notesSummaryRingkasanStoriesPracticeResponsible AI Challenges in Traditional AI and Generative AI12 exam-style questions on this lesson.Question 1 of 12A model scores 99% accuracy on its training data but only 60% on new data. What is happening?ABalanced fitBUnderfittingCOverfittingDData driftCheck answerQuestion 2 of 12A model performs poorly even on the data it was trained on. What is the likely problem?ABalanced fit: the model has low bias and low varianceBOverfitting: the model memorized noiseCData leakage: the test set is too large for trainingDUnderfitting: the model is too simple and has high biasCheck answerQuestion 3 of 12Which combination describes the goal for a well-trained model?ALow bias, low varianceBHigh bias, high varianceCHigh bias, low varianceDLow bias, high varianceCheck answerQuestion 4 of 12What does variance measure in a model?AThe number of features the model uses as inputsBHow much the model's predictions change with different training dataCThe size of the test set compared with the training setDThe gap between the predictions and the true valuesCheck answerQuestion 5 of 12Which technique trains on subsets of the data and tests on the remaining data to detect overfitting?AEarly stoppingBRegularizationCCross-validationDData augmentationCheck answerQuestion 6 of 12. Select two.Which TWO techniques help reduce overfitting? (Select TWO.)AStopping training earlyBAdding more features without more dataCTraining for many more epochs after validation error starts risingDRegularization, which penalizes extreme weightsEMaking the model more complexCheck answer0 of 2 selectedQuestion 7 of 12A team uses principal component analysis (PCA) on a dataset. What is PCA an example of?ACross-validationBRegularizationCData augmentationDDimension reductionCheck answerQuestion 8 of 12A model writes a report that includes a convincing but invented scientific citation. Which generative AI challenge is this, and what causes it?AHallucination, caused by sampling likely next words instead of checking factsBToxicity, caused by offensive examples in the training dataCPlagiarism, caused by copying text from a real published paperDIntellectual property, caused by the model license termsCheck answerQuestion 9 of 12A generative AI model reproduces paragraphs from a copyrighted book word for word. Which challenge is this?AToxicityBIntellectual propertyCHallucinationDNondeterminismCheck answerQuestion 10 of 12A university worries that students are submitting essays written by AI and it can't verify authorship. Which challenge is this?AUnderfittingBDisruption of the nature of workCPlagiarism and cheatingDToxicityCheck answerQuestion 11 of 12Why is toxicity hard to manage in generative AI?AToxicity only affects image models, not text modelsBLanguage models can't produce offensive textCRegulators require models to allow all outputDWhat counts as offensive depends on context and cultureCheck answerQuestion 12 of 12Employees worry that generative AI will replace or transform their jobs. Which challenge does the course call this?ADisruption of the nature of workBRegularizationCIntellectual propertyDHallucinationCheck answerResponsible AI6 exam-style questions on this lesson.Core Dimensions of Responsible AI10 exam-style questions on this lesson.