3508 - 借助 NVIDIA 和 IBM 在企业中大规模启用智能代理 AI.pdf

《3508 - 借助 NVIDIA 和 IBM 在企业中大规模启用智能代理 AI.pdf》由会员分享，可在线阅读，更多相关《3508 - 借助 NVIDIA 和 IBM 在企业中大规模启用智能代理 AI.pdf（45页珍藏版）》请在三个皮匠报告上搜索。

1、Orlando,FLOctober 69IBM TechXchange 2025Session 3508Joseph DainIBM Storage CTO Office STSM,Master InvenrorJacob LibermanNVIDIA Director of Product ManagementEnable Agentic AI at scale in the Enterprise with NVIDIA and IBM Agenda0102030405Problem StatementIBM Content Aware StorageNVIDIA AI Data Platf

2、ormAgentic AI Use CasesIBM Fusion AI Data PlatformIBM TechXchange|2025 IBM Corporation3”Organizations Need to Unlock Value From DataWherever it Resides1 1IBM(IBM)Q2 2021 Earnings Call TranscriptIBM(IBM)Q2 2021 Earnings Call Transcript“We fundamentally believe that core to the competitiveness of ever

3、y company going forward will be their ability to use AI to unlock real-time value from their data wherever the data resides.Nearly all public datais represented in foundation modelsBut only 1%ofenterprise data!Data is the fuel for an effective AI strategyOnly small fraction of enterprise data is use

4、d in Gen AITypical RAG processUserQuestionSearch&RetrievalPrompt=Instructions+Search Results+QuestionLLMGenerated output with sourcesTop search resultsSignificantly elevates level of trust:Ensures that the model has access to themost current and reliable facts System becomes business-aware Sources a

5、re known,ensuring output can bechecked for accuracy Less likely to make-up a factually inaccurateresponses Current RAG implementations for enterprise today are complex,inefficient and costly51.High cost:too many copies of data and unnecessary data movement 2.Complex:too many technologies to integrat

6、e and data types to support3.Security exposure:too many replicas of data with Inconsistent ACL between raw data and embedding4.Stale embedding:long latency between data change and updated vectors5.Limited scalability Making Enterprise Data Ready for AIIBM TechXchange|2025 IBM Corporation6SwitchSwitc