Task 2: Set the policies and thresholds for detecting harmful content
-
On the ContentSafetyXXXX blade, in the navigation pane, select Overview.
-
Scroll-down and under Make an API call, select Content Safety Studio.
-
The Content Safety Studio will open in a new tab, in the upper-right corner, select Sign In.
-
On the Pick an account window, select your lab admin user account @lab.CloudPortalCredential(User1).Username.
-
Back in Content Safety Studio, select Moderate text content.
-
On the Moderate text content blade, under Try it Out, verify the checkbox is selected and then select Choose a different resource.
-
On the Settings blade, select ContentSafetyXXXX and then select Use resource.
-
Back on the Moderate text content blade, under 1. Select a sample or type your own, select Violent content with misspelling.
-
Under the 2. Test text box, select Run test.
-
Under 3. View results, verify the content has been Blocked, and has been rejected using the Violence category.
-
Also notice the other categories that were judged to be Allowed.
-
In the upper-left corner, select Content Safety Studio to return to the main page.
-
Scroll down and then select Prompt Shields.
-
On the Prompt shields blade, under 1. Select a sample or type your own, select Document attack content.
-
Under the 2. Test text box, select Run test.
-
Under 3. View results, verify the results show that jailbreak attacks were detected in the document.