KM知识管理需要 AI：为什么？.pdf

《KM知识管理需要 AI：为什么？.pdf》由会员分享，可在线阅读，更多相关《KM知识管理需要 AI：为什么？.pdf（19页珍藏版）》请在三个皮匠报告上搜索。

1、KM for AIDaniel W.RasmusDaniel W.RasmusPrincipal AnalystSerious Insights LLChttps:/ on X and InstagramAgenda3 3GuardrailsThe Impact of Unguarded Guardrails5 5Guardrails are often(almost always)opaque too end usersToo cautiousCause errorsCan censor informationGuardrails can be easily bypassedExternal

2、“mandates”implemented inconsistently across productsOpen guardrails arent guardrailsOperationalizing guardrails can be complex.Integrating guardrails into existing workflows and systems requires collaboration across different teams and ensuring that all stakeholders understand and adhere to them It

3、can be difficult to keep up with legal and regulatory changesFine-tuning can compromise safety.LLM DiscoverabilityLLM Discoverability7 7Nonstandard metadata and model descriptionsThe models themselves cant be indexedInconsistent naming conventionsA vast number of models with varying capabilitiesDiff

4、iculty in assessing model quality without extensive testing and the potential for misleading or outdated information about model performanceContext Model ManagementContext Model Management9 9Model Focus/DomainsConfigurationDepthEditorial PoliciesPrompt Collaboration and SharingPrompt Collaboration a

5、nd Sharing1111Prompt sharing and reusePrompt improvementsPrompt variations by LLMLLM Model Management and RetirementLLM Model Management and Retirement1313Model Transparency:What model am I using?Which version of the model am I using?What is the“knowledge cutoff date”of my model?After the discontinu

6、ation date,will the new model require different interactions?Provide different answers?How do costs change with a new model?New Knowledge or Lack ThereofNew Knowledge or Lack Thereof1515What does my model know?How does my model know it?Knowledge GraphRAGAPI to