An AI moderation system is missing more toxic comments after a policy update. Diagnose it.

Focus on identifying potential issues stemming from the recent policy update that could affect the AI moderation system's performance. Consider the specifics of the changes made in the policy—did they alter the criteria for what constitutes a toxic comment? Analyze how these changes might impact the training data used for the AI model, including whether the model was retrained or if it still relies on outdated data. Discuss the importance of feedback loops and how the system's learning mechanism may need adjustment. Finally, think about the metrics that could help you evaluate the system's effectiveness post-update, and suggest ways to gather user feedback to improve the moderation process.