Databricks Generative AI Engineer Associate exam dumps

Databricks Generative AI Engineer Associate practice question 250 of 306

Databricks Certified Generative AI Engineer Associate. Free level, Databricks. Free question with the correct answer and a full explanation.

Databricks Generative AI Engineer Associate Question 250

Single answer

You are training a large language model on sensitive company data using Databricks. To meet a specific performance objective, you decide to mask sensitive information during preprocessing to prevent it from influencing the model's training. Which masking technique would be appropriate to ensure the model can still generalize well while meeting the performance objective?

  1. A

    Replace sensitive information with random strings.

  2. B

    Use consistent token masking by replacing sensitive information with a placeholder token such as ''.

  3. C

    Remove all rows containing sensitive information from the dataset.

  4. D

    Replace sensitive information with realistic synthetic data that follows the same distribution as the original data.

Show answer and explanation

Correct answer: B

Explanation

Using consistent token masking with a placeholder token like '' ensures that sensitive information is protected while still allowing the model to learn underlying patterns in the data. This approach maintains the context needed for generalization and is often used as a guardrail to balance data privacy and model performance.

  • A. Incorrect.

    Replacing sensitive information with random strings can disrupt the model's ability to generalize because random strings do not follow meaningful patterns or context.

  • B. Correct.

    Using consistent token masking with a placeholder token like '' allows the model to learn general patterns without being exposed to sensitive information, and it ensures the model can still generalize.

  • C. Incorrect.

    Removing all rows containing sensitive information may lead to a loss of important context or patterns in the data, reducing the quality and diversity of the training dataset.

  • D. Incorrect.

    Replacing sensitive information with synthetic data could work in some cases, but it may introduce biases or fail to accurately represent the distribution of the original data, which could hinder model performance.

Timed practice exam

Take a Databricks Generative AI Engineer Associate practice test under exam conditions

45 questions in 90 minutes, drawn from this bank, with a score report and a per-question review when you finish.

Start timed exam