We are excited to announce that Custom Categories is coming soon to Azure AI Content Safety. This new feature enables you to create your own customized classifier based on your specific needs for content filtering and AI safety whether you want to detect sensitive content, moderate user-generated content, or comply with local regulations. Use Custom Categories to train and deploy your own custom content filter with ease and flexibility.
The Azure AI Content Safety custom categories feature is powered by Azure AI Language, a service that provides advanced natural language processing capabilities for text analysis and generation. The custom categories feature is designed to provide a streamlined process for creating, training, and using custom content classification models.
Here's an overview of the underlying workflow:
We are offering two deployment options for our customers:
Both options serve to empower organizations with the capability to protect their AI applications and users more effectively against a wide array of harmful content and security risks, offering a balance between responsiveness and thoroughness based on the specific needs and circumstances.
By creating a custom category, you are telling the AI exactly which types of content you wish to detect and mitigate. You need to create a clear category name and a detailed definition that encapsulates the content's characteristics. The setup phase is crucial, as it lays the groundwork for the AI to understand your specific filtering needs.
Then, collect a balanced and small dataset with both positive and (optional) negative examples allows the AI to learn the nuances of the category. This data should be representative of the variety of content that the model will encounter in a real-world scenario.
Once you have your dataset ready, the Azure AI Content Safety service uses it to train a new model. During training, the AI analyzes the data, learning to distinguish between content that matches the custom category and content that does not. Built on top of the underlying technology of LLM-powered low-touch customization from Azure AI Language, we are tailoring the experience for Content Safety customer towards consistency and more focus on content moderation scenario.
After training, you need to evaluate the model to ensure it meets your accuracy requirements. This is done by testing the model with new content that it hasn't seen before. The evaluation phase helps you identify any potential adjustments needed before deploying the model into a production environment.
In the upcoming release of custom categories studio experience, we will introduce a feature that allows users to modify their definition and training samples using suggestions generated by GPT.
South Australia Department for Education
“The Custom Categories feature from Azure AI Content Safety is set to be a game-changer for the Department for Education in South Australia, and our pioneering AI chatbot, EdChat. This new feature allows us to tailor content moderation to our specific standards, ensuring a safer and more appropriate experience for users. It's a significant step towards prioritizing the safety and well-being of our students in the digital educational space.”
- Dan Hughes, Chief Information Officer, South Australia Department for Education
Learn more about how South Australia Department for Education is using Azure AI Content Safety
Thank you for your support as we continue to enhance our platform. We are excited for you to begin using custom categories. Stay tuned for more updates and announcements on our progress.
In the meantime, we encourage you to visit our Content Safety documentation or studio to explore the existing capabilities available to you. Custom categories is also coming soon to Azure AI Studio and Azure OpenAI Service.
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.