What types of manual review are available in AI Guardrails? How can I use them?
AI Guardrails supports the following types of manual review:
- Human review: You can review specific machine-assisted moderation results. In this case, human review results and machine-assisted moderation results are returned.
- Manual review: You can review all the content. In this case, only manual review results are returned.
For more information, see Enable manual review services.
For more details about the quality and solutions for manual review, contact the AI Guardrails helpdesk.
What are the service level agreement (SLA) metrics of manual review in AI Guardrails?
Manual review in AI Guardrails involves two important SLA metrics: time performance and quality. You can set the manual review duration to 5 or 30 minutes as needed to deliver different time performance. To ensure review quality, a manual review process consists of the review, double-check, and quality check phases. This way, content can be reviewed in a short time period with an accuracy up to 99%. In addition, you can flexibly expand the capabilities of manual review to respond to emergency scenarios in which manual review services must be deployed, managed, and offered in an efficient way. In major events or grand festivals, corresponding control standards are provided to help effectively reduce the risk of content violations.
How can I specify a callback URL for manual review in AI Guardrails?
You can specify a callback URL for human review in the AI Guardrails console. Human review and machine-assisted moderation share the same business scenario. The callback notifications for both human review and machine-assisted moderation are sent to the callback URL of this business scenario. Machine-assisted moderation and human review can share the same callback URL. To specify a callback URL for manual review, you must set the callback parameter in the review request.
For more information, see Manual image review.