利用可观测性构建负责任的人工智能应用.pdf

1、 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.C O P 3 6 0Leverage observability to build responsible AI applicationsSasi Kiran MalladiPrincipal Technical Account ManagerAWS 2025,Amazon Web Services,Inc.or its af

2、filiates.All rights reserved.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Is your AI trustworthy?2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.AgendaWhats the problem?What is responsible AI?Enabling Amazon Bedrock GuardrailsLeveraging Amazon CloudWatch for

3、actionable insightsNext steps 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Whats the problem?Real-world examples 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Hallucinations 2025,Amazon Web

4、Services,Inc.or its affiliates.All rights reserved.Sensitive information 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.What is responsible AI?2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Bui

5、lding generative AI applications brings new challengesPrivacy protectionProtect user information or sensitive dataResponse hallucination Inaccurate or fabricated informationToxicity and safety(including brand risk)Harmful or offensive responsesUndesirable and irrelevant topicsControversial queries a

6、nd responses 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Responsible AI considerationsFairnessExplainabilityPrivacy and securitySafetyControllabilityVeracity and robustnessGovernanceTransparency 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.The foundation