Developing generative AI solutionsFull notesSummaryRingkasanStoriesPracticeSelecting an FM10 exam-style questions on this lesson.Question 1 of 10What is a drawback of starting from a pre-trained foundation model?AIt always costs more than training from scratchBIt takes longer to converge when fine-tunedCIt can't be customized at allDIt can carry bias from its training data or miss domain nuancesCheck answerQuestion 2 of 10A live speech translation app must respond while people are talking. Which selection criterion matters most?ALatencyBLicensingCModel sizeDInput lengthCheck answerQuestion 3 of 10A legal team needs to process 300-page contracts in one request and get long summaries back. Which criterion matters most?AModel complexity and architectureBInput/output length (maximum sequence length)CModality of the outputDMultilingual support for legal termsCheck answerQuestion 4 of 10A company serves customers in Japanese, Arabic, and Portuguese. Which criterion must it check?ALatencyBCost onlyCMultilingual supportDModel complexityCheck answerQuestion 5 of 10A company needs a model that produces both images and text. Which criterion is this?ACustomizationBLatencyCLicensingDModalityCheck answerQuestion 6 of 10Which trade-off applies to bigger foundation models?AThey do better on complex tasks but need more computeBThey always respond with lower latencyCThey are cheaper and faster to runDThey can't be fine-tuned with your dataCheck answerQuestion 7 of 10. Select two.Which TWO criteria relate to responsible model selection? (Select TWO.)AColor of the user interfaceBSources of the training dataCRisk of bias and misinformationDNumber of GPUs in the officeELength of the product nameCheck answer0 of 2 selectedQuestion 8 of 10A team must confirm that a model works with its existing infrastructure, tools, and libraries. Which criterion is this?AMultilingual supportBDeployment and integrationCModel sizeDModalityCheck answerQuestion 9 of 10. Select two.Which TWO are types of Amazon Titan foundation models? (Select TWO.)AImage generationBReinforcement learning agentsCDocument form extractionDSpeech-to-text transcriptionEEmbeddingsCheck answer0 of 2 selectedQuestion 10 of 10Why should a team revisit its model selection criteria regularly?AModels stop working after about a yearBCriteria never change once they are setCBetter or cheaper models appear oftenDAWS requires a new model selection monthlyCheck answerDefining a Use Case7 exam-style questions on this lesson.Improving the Performance of an FM13 exam-style questions on this lesson.