Answer: Negative prompting and content filtering

## Explanation **Correct Answer: B (Negative prompting and content filtering)** **Why this is correct:** 1. **Negative prompting** is a technique where you explicitly tell the AI model what NOT to include in the generated content. This is particularly useful for avoiding specific terms like competitor brand names. 2. **Content filtering** involves implementing filters or guardrails to detect and block offensive language, inappropriate content, or specific prohibited terms from appearing in the AI-generated output. 3. **Bedrock's capabilities**: AWS Bedrock provides built-in content filtering and safety features that can be configured to block specific types of content, including offensive language and competitor mentions. **Why other options are incorrect:** - **A) Zero-shot prompting**: This refers to providing a prompt without any examples, but it doesn't specifically address content filtering or preventing certain outputs. - **C) Reinforcement learning**: This is a training methodology where models learn through rewards/penalties, not a content filtering strategy for production use. - **D) Hyperparameter tuning**: This involves adjusting model training parameters to improve performance, not controlling specific content outputs. **Best practice approach:** The marketing agency should implement a combination of: - Negative prompting in their requests (e.g., "Do not mention [competitor brand names]") - Bedrock's built-in content filtering settings - Additional post-generation validation checks - Custom guardrails for brand-specific compliance

Author: Jin H