AB-731: AI Transformation LeaderDemo

By Microsoft · Question Mode

QuestionQ42

Identify an implementation and adoption strategy for Microsoft’s AI apps and services

A generative AI solution used by your company has been fine-tuned on data that reflects the general population. You notice that some of the responses it generates contain inappropriate or exclusionary language stemming from ableist assumptions.

You need to stop these inappropriate responses from being generated, and your solution must minimize costs.

What should you do?

  • A Apply a newer version of the generative AI model.
  • B Apply a content-moderation filter.
  • C Create a new version of the solution that is trained on only inclusive and representative content.
  • D Create a new version of the solution that is trained on only exclusionary content.
Explanation

A content-moderation filter can be layered on top of the existing model to detect and block harmful, exclusionary, or ableist language in real time, without the need to retrain or replace the underlying model. This approach directly addresses the harmful output while avoiding the significant cost and time investment associated with retraining the solution (as would be required by creating a new fine-tuned version) or upgrading to a different model version, neither of which guarantees the specific issue is resolved and both of which are more costly. Content moderation/filtering is a standard responsible-AI mitigation for controlling generated output content cost-effectively.

Learn more

Community Discussion

No comments yet. Be the first to start the discussion!