IBM C1000-185 問題集

質問 1：
Which of the following statements accurately describes a drawback of using soft prompts in generative AI model optimization?
A. Soft prompts can increase the model's interpretability by providing clear, user-defined input instructions.
B. Soft prompts offer improved performance for specific tasks but are harder to implement when fine-tuning models across multiple domains.
C. Soft prompts require additional computational resources during training, which can limit their scalability in real-time applications.
D. Soft prompts make it easier to control the model's behavior as the prompts are flexible and can be adjusted by the user during inference.
正解：C

質問 2：
A company is using IBM's InstructLab to fine-tune a large language model (LLM) to perform customer support tasks, such as answering frequently asked questions (FAQs) and troubleshooting product issues.
Which of the following components of InstructLab plays the most crucial role in ensuring the model learns to align its responses with the specific format and tone required for customer interactions?
A. The user simulation environment, which provides real-time testing of model outputs.
B. The instruction optimizer, which tunes hyperparameters to improve task-specific performance.
C. The prompt-tuning engine, which fine-tunes model outputs based on pre-defined instructions.
D. The evaluation metrics dashboard, which tracks model performance on customer interaction tasks.
正解：C

質問 3：
You are tasked with improving the performance of a Retrieval-Augmented Generation (RAG) system in IBM watsonx. Part of this improvement involves selecting the right embedding model for document retrieval.
Which of the following is the best description of the differences between various embedding models, and how would you choose the most suitable model for your task?
A. Word2Vec embeddings capture only the syntactic relationships between words, while BERT embeddings focus on both syntax and semantic context, making BERT more suitable for complex retrieval tasks in a RAG system.
B. TF-IDF is an advanced embedding model that captures both the frequency and semantic meaning of words, making it more effective than deep learning-based models like BERT for retrieval in RAG systems.
C. Word2Vec, GloVe, and BERT are all embedding models, but BERT embeddings capture richer context by considering the entire sentence rather than just the local context, making it more effective for generating semantically relevant embeddings.
D. BERT embeddings are context-independent, which makes them less useful for a RAG system than Word2Vec or GloVe, which focus on learning semantic relationships between words.
正解：C

質問 4：
A financial institution is deploying a generative AI model to generate loan approval recommendations based on applicant profiles, including factors like income, credit score, and employment history. The organization is concerned about ensuring that the model does not introduce bias in its recommendations, particularly related to gender and race. You have been asked to design a process to evaluate the model's inferences during deployment and mitigate any potential bias.
Which method would be most effective for evaluating the model's inferences for bias in this deployment scenario?
A. Manually review all loan decisions generated by the model for signs of bias before releasing them to customers.
B. Implement a fairness audit, where a sample of the model's inferences is checked for disparate impact across protected groups such as gender and race.
C. Periodically retrain the model with updated datasets that exclude sensitive attributes such as gender and race.
D. Use greedy decoding in the inference phase to ensure deterministic outputs, avoiding potential bias from probabilistic sampling methods.
正解：B

質問 5：
You are optimizing a large language model (LLM) by prompt-tuning it for specific enterprise-level tasks. The goal is to initialize the prompt in such a way that it helps the model generalize well across various enterprise domains, such as finance, healthcare, and retail.
What is the most effective method to initialize the prompt for such a use case?
A. Use a short prompt that provides no guidance and allow the model to self-optimize
B. Use a single, highly specific prompt tailored to only one domain, such as finance
C. Initialize the prompt with an ensemble of prompts covering multiple domains
D. Start with a general prompt and gradually specialize it during fine-tuning
正解：C

質問 6：
When tuning model parameters for a generative AI prompt, which of the following adjustments would most likely increase the model's tendency to generate coherent but less creative responses?
A. Reducing the beam size in beam search from 5 to 1
B. Decreasing the value of the temperature parameter to 0.2
C. Increasing the temperature parameter to 1.5
D. Using Top-k Sampling with a k value of 100
正解：B

質問 7：
You are building a customer support chatbot using IBM watsonx.ai and Watson Assistant. The chatbot must use watsonx.ai's large language model (LLM) to generate dynamic responses and Watson Assistant to manage dialog and interaction flow.
What is the most efficient way to integrate these two services to deliver an optimal solution?
A. Build a separate microservice for each service, allowing Watson Assistant and watsonx.ai's LLM to operate independently, with no communication between them.
B. Deploy watsonx.ai's LLM within Watson Assistant by embedding the LLM directly into the Watson Assistant environment.
C. Use Watson Assistant as the primary interface and call watsonx.ai's LLM through an API for generating dynamic responses in specific intents.
D. Use Watson Assistant to directly generate all responses, bypassing watsonx.ai's LLM.
正解：C

質問 8：
A data scientist is choosing between using hard prompts and soft prompts in a generative AI project.
Which of the following best explains why hard prompts might be more suitable for scenarios where explainability is crucial?
A. Hard prompts reduce the model's flexibility by making the output deterministic, which enhances explainability.
B. Hard prompts allow for a clear, human-readable set of instructions that directly guide the model's behavior.
C. Hard prompts dynamically adjust the model's internal representations, providing more clarity in complex situations.
D. Hard prompts are based on learned embeddings, which offer better model understanding due to their complexity.
正解：B

質問 9：
You are tasked with designing a prompt template to assist a chatbot in generating professional email responses for customer service inquiries. The system should prioritize politeness, clarity, and conciseness.
What elements should be included in the prompt template to achieve the best results, considering optimal behavior of a large language model (LLM)? (Select two)
A. Provide the customer's emotional context for better alignment with the tone
B. Ask the model to generate multiple versions of the response and rank them
C. Instruct the model to limit responses to a specific character count
D. Specify the output tone as polite and professional
E. Include examples of informal customer service responses for variability
正解：C,D