4278 - 云原生成熟度与人工智能应用之间的联系.pdf

1、Orlando,FLOctober 69IBM TechXchange 2025Explore.Build.Launch.Session code 2478John DoyleIBM,Director of Product ManagementThe Links Between Cloud Native Maturity and AI AppsAgenda01020304New AI ChallengesRemediationsAI ArchitecturesCloud Native RelevanceIBM TechXchange|2025 IBM Corporation“AI will b

2、e in integral part of solving the worlds biggest problems,but it must be developed in a way that reflects human values”Satya NadellaCEO of MicrosoftIBM TechXchange|2025 IBM Corporation3New AI Quality ChallengesMassive expansion of the variety of acceptable inputs and outputs in comparison to the pas

3、t.Machine Learning/LLMs are nondeterministic by nature.Errors are to be expected.IBM TechXchange|2025 IBM Corporation4New risks to protect your enterprise fromErrors in fact.AI enabled applications should accurately reflect the facts about your business.Tone.Replies should always be professional and

4、 polite,regardless of the input.Off topic conversations.Your model may be able to write Python,your customer service agent should not.IP exposure or infringement.Never expose internal IP or violate the IP of others.IBM TechXchange|2025 IBM Corporation5Validating outputs with an LLM as a judge to man

5、aged the wide range of acceptable outputsTracing and Audit logging to identify where errors and drift originateEvaluation via scores rather than Pass/Fail testingToken based authentication and authorization to validate end user permission to sensitive data.Application guardrails that block harmful c

6、ontent and validate sensitive data.Drift monitoring of outputs and errors over time as models changeForms of Risk RemediationIBM TechXchange|2025 IBM Corporation6Moving to Evaluation in Production EnvironmentsTraditional forms of testing are insufficient for A