C1000-185試験無料問題集「IBM watsonx Generative AI Engineer - Associate 認定」

A. Use temperature tuning to control the diversity of outputs, reducing the likelihood of personal information being revealed.

B. Implement real-time filtering of model outputs using regular expressions to detect and mask personal information.

C. Use a greedy decoding strategy to limit the creativity of the model and prevent unexpected outputs.

D. Train the model on synthetic data that does not include any personal information.

A. Start with NLP analysis, pass the data to watsonx.ai for summarization, and then perform document retrieval to verify the accuracy of the summary.

B. Break the LangChain-based chain into individual steps that allow for manual intervention at each stage, ensuring control over the process at every step.

C. Use watsonx.ai to generate a summary immediately, and then perform NLP analysis and document retrieval in parallel to verify the accuracy of the output.

D. Perform document retrieval first, followed by NLP analysis to extract relevant information, and then pass the processed data to watsonx.ai for summarization.

A. Manually coding data augmentation scripts within the UI

B. Uploading real-time streaming data for immediate synthetic generation

C. Directly modifying the architecture of the model to generate task-specific synthetic data

D. Using pre-built templates to generate domain-specific synthetic data

A. Stopping the model after a predetermined number of tokens, regardless of context

B. Stopping the model only when it reaches the end of a predefined phrase from the input prompt

C. Monitoring the likelihood of the next token and stopping when the likelihood drops below a threshold

D. Stopping when the model generates special end-of-sequence tokens, such as <EOS>

A. Increase the batch size and reduce the learning rate simultaneously to speed up the tuning process and minimize training iterations.

B. Use all available training data, including unrelated examples, to ensure the model has a broad understanding of multiple tasks before tuning.

C. Perform full-scale retraining of the model for each new task to ensure maximum adaptability and accuracy.

D. Focus the tuning on adjusting only the model's last few layers, which are responsible for task-specific outputs, while leaving the majority of the model unchanged.

A. Tune the model for text summarization, condensing user queries into shorter forms.

B. Fine-tune the model to generate product descriptions using longer contextual prompts.

C. Focus on prompt-tuning the model for multi-turn dialogue to simulate more natural conversations.

D. Optimize the model for extractive question-answering from a predefined knowledge base.

A. Set a high number of epochs and use early stopping to determine the optimal point

B. Set the number of epochs equal to the size of the dataset

C. Use a single epoch to avoid overfitting

D. Gradually increase the number of epochs until the loss on the training set reaches zero

A. Normalize the text by removing all punctuation, special characters, and converting text to lowercase.

B. Randomly split the data into training, validation, and test sets.

C. Use domain-specific tokenization to better capture important keywords and phrases relevant to customer support.

D. Use data augmentation techniques like paraphrasing to artificially increase the dataset size.

A. Use a pre-trained general-purpose prompt with no domain-specific customization

B. Use a manually crafted prompt tailored to the specific context of customer support queries

C. Construct a prompt using a large set of random tokens from the training corpus

D. Use a prompt with pre-defined output patterns to restrict the model's possible responses

A. Add more layers to the model to increase complexity

B. Enable regularization techniques like dropout

C. Increase the number of training samples used for tuning

D. Tune the prompt templates further by standardizing the structure

E. Increase the batch size during training

A. Crafting prompts with very specific and detailed instructions, ensuring the model follows a strict framework for the desired response.

B. Using example prompts that introduce multiple topics at once to evaluate how well the model handles multitasking during response generation.

C. Using prompts with complex sentence structures and advanced terminology to challenge the model's understanding capabilities.

D. Writing prompts that are extremely vague, allowing the model to freely interpret the input and generate diverse responses.

A. Including specific disclaimers or guidelines in the prompt instructing the model to avoid biased language.

B. Adjusting the model's temperature to a lower value to generate more deterministic and consistent outputs.

C. Designing prompts that explicitly ask for balanced perspectives, such as "Provide both pros and cons of this issue."

D. Using a larger number of examples in the prompt to steer the model towards safer and more neutral outputs.

A. Top-p sampling prioritizes tokens based on their frequency in the training dataset, ignoring their conditional probabilities.

B. Top-p sampling selects the next token only from a fixed number of top tokens, ensuring the most predictable outcome.

C. Top-p sampling ensures that only the top 1% most probable tokens are chosen, regardless of context.

D. Top-p sampling adjusts the probability threshold dynamically, allowing the model to sample tokens until the cumulative probability reaches p.

A. Apply a rule-based content filter to the model's outputs to remove any phrases that appear to contain personal information.

B. Train the model on anonymized data to ensure that personal information is never present in the training set.

C. Use a privacy-preserving tokenization method to mask personal data in the input before feeding it into the model.

D. Limit the model's ability to retain memory of previous user inputs by resetting its state after every query.

A. "Provide a product description for the following items, ensuring it is factual, concise, and includes specific details such as size, color, and material."

B. "Generate a product description that highlights the unique aspects of the product and uses emotional language to engage the reader."

C. "Write a summary that provides information on each product, making the content engaging, humorous, and memorable."

D. "Generate a creative and imaginative product description for the items listed below."