Developing generative AI solutionsFull notesSummaryRingkasanStoriesPracticeImproving the Performance of an FM13 exam-style questions on this lesson.Question 1 of 13A team wants to improve an LLM's output as quickly as possible, without any extra training or infrastructure. What should it try first?ABuilding a vector databaseBFine-tuningCPre-training a new modelDPrompt engineeringCheck answerQuestion 2 of 13Which aspect of prompt engineering means combining several prompts to get a better result?AEnsemblingBTuningCAugmentationDMiningCheck answerQuestion 3 of 13A company's support chatbot must answer questions from its latest product documentation, which changes every week. Which approach fits best?ARaise the temperature for more varied answersBRAG with Knowledge Bases for Amazon BedrockCPre-train a new model on the docs each monthDFine-tune the model on the docs every weekCheck answerQuestion 4 of 13What are the two main parts of a RAG system?AA load balancer that routes and a firewall that filtersBA labeling workforce and a reward model that ranks answersCA retriever that finds passages and a generator that answersDAn encoder that compresses and a discriminator that judgesCheck answerQuestion 5 of 13Which AWS feature gathers a company's data sources into a repository that RAG applications draw on?AAmazon KinesisBAWS ArtifactCSageMaker Feature StoreDKnowledge Bases for Amazon BedrockCheck answerQuestion 6 of 13What does fine-tuning change?AThe model's weightsBOnly the retrieved documentsCNothing about the modelDOnly the promptCheck answerQuestion 7 of 13A medical company wants its model to understand clinical terminology deeply. It has a large set of labeled medical texts. Which approach fits?AAgents onlyBFine-tuning on the medical dataCLowering the temperatureDZero-shot promptingCheck answerQuestion 8 of 13. Select two.Which TWO are types of fine-tuning named in the course? (Select TWO.)ACross-validationBBatch transformCInstruction fine-tuningDDimensionality reductionEReinforcement learning from human feedback (RLHF)Check answer0 of 2 selectedQuestion 9 of 13When is building a foundation model from scratch appropriate?AWhen the budget is small and time is shortBFor every new chatbot the company launchesCFor research, or when no pre-trained model fits the needDWhen the team wants the fastest, simplest optionCheck answerQuestion 10 of 13Which order lists approaches from cheapest to most expensive?AFine-tuning, RAG, prompt engineering, pre-trainingBPre-training, fine-tuning, RAG, prompt engineeringCRAG, prompt engineering, pre-training, fine-tuningDPrompt engineering, RAG, fine-tuning, pre-training from scratchCheck answerQuestion 11 of 13A travel app needs an assistant that checks flight availability, books a seat, charges the card, and emails a confirmation, in the right order. Which capability fits?AAgents for Amazon BedrockBA larger context windowCA lower temperatureDKnowledge Bases onlyCheck answerQuestion 12 of 13What is the core role of agents in Amazon Bedrock?AEncrypting data at rest for each taskBCoordinating subtasks in the right orderCStoring embeddings for later retrievalDLabeling training data for fine-tuningCheck answerQuestion 13 of 13. Select two.Which TWO prompt engineering techniques are named in the course? (Select TWO.)ASHAPBChain-of-thought (CoT)CReActDSMOTEEPCACheck answer0 of 2 selectedSelecting an FM10 exam-style questions on this lesson.Evaluating an FM11 exam-style questions on this lesson.