向量数据库的兴起使生成性人工智能成为可能.pptx

1、Kevin PetrieVP ResearchMay 8,2024Rise of the Vector DatabaseEnabling Generative AIGenerative AI Adoption StagesGenerative AI Adoption StagesMost GenAI adopters are integrating language models into multi-faceted workflows1.Language Model Platforms1.Language Model PlatformsEmployees use LLM tools as s

2、tandalone platformsOpenAI ChatGPTGoogle BardHugging Face BLOOM2.LM Assistants within Tools2.LM Assistants within ToolsEmployees use LLM functions within commercial toolsSalesforce EinsteinGitHub CopilotSAP Joule3.LM-Driven Workflows3.LM-Driven WorkflowsCompanies build LM functions into multi-faceted

3、 workflowsCustom tools,applications,and integrationsPARAMETER COUNTSPARAMETER COUNTSDOMAIN-SPECIFIC DATADOMAIN-SPECIFIC DATATODAYS FOCUSTODAYS FOCUSBUILD FROM SCRATCHFINE TUNE EXISTING LMENRICH LM PROMPTS(RAG)DETAILSCollect and prepare corpusDesign and train LMFine tune a pre-trained LM(ChatGPT,Bard

4、,LLaMA,etc.)on domain-specific dataUse retrieval augmented generation(RAG)Inject domain-specific content into LM promptsPROSAccuracyDomain-specific languageAccuracyDomain-specific languageLow data volumesAccuracyLow data volumeLower compute needsCONSData science expertiseHigh data volumesMany iterat

5、ionsExpensive computeData science expertiseExpensive computeReliance on data pipelinesVector databases play a critical role in all three architectural approaches+-Architectural Approaches to Domain-Specific Language Models Architectural Approaches to Domain-Specific Language Models MOST COMMONMOST C

6、OMMONBUILD FROM SCRATCHFINE TUNE EXISTING LMENRICH LM PROMPTS(RAG)INDUSTRIESTechnologyAI nativeCloud nativeHealthcareLegal servicesProfessional servicesRetailManufacturingTECH MATURITYAI-and cloud-nativeData science-drivenData-drivenDOMAIN SPECIFIC FOCUSLanguage and factsFactsUSE CASESCustomer servi