赞助方：蒙特卡洛 完成的幻觉：为什么人工智能的真正工作始于生产.pdf

1、 2025 Monte Carlo Data,Inc.The Illusion of DoneWhy the Real Work for AI Starts in ProductionShane Murray&Bryce Heltzel,Monte Carlo 2025 Monte Carlo Data,Inc.Shane MurrayMonte CarloBryce HeltzelMonte CarloPRESENTERS 2025 Monte Carlo Data,Inc.2025 Monte Carlo Data,Inc.2025 Monte Carlo Data,Inc.3.Opera

2、tionalizing Reliability&TrustHidden Hurdles1.The Illusion of Done2.Lessons in Building Impactful AI 2025 Monte Carlo Data,Inc.Launch Day:model deployed,now what?2025 Monte Carlo Data,Inc.A pragmatic approach6Focus on deep business pains-May or may not be a chatbot Double down on GenAI strengths-Summ

3、arization-Structuring the unstructured-Code generation-Process automation-Autonomous agents 2025 Monte Carlo Data,Inc.Starting from the user problem7 Problem:Complex data estates make troubleshooting issues feel like finding a needle in a haystack High pain:Costly incidents,long TTR,SME resource-dep

4、endency We had:Existing access&collection of metadata,logs,lineage,&data sampling Rich historical incident data Clear workflows(and pain points)from data teams Repeated,codifiable questionsripe for LLMs 2025 Monte Carlo Data,Inc.Getting to value8 Push quickly towards a prototype What can we build in

5、 a month?What can we add this week?Validate with real customer data/scenarios Walls of text arent the best UI,even harder to validate when its mock data.Push out production problems-focus on what can get you customer feedback Sharing model outputs with shared google docs Punted infrastructure until

6、we got clear customer validation 2025 Monte Carlo Data,Inc.How it worksLive Demo 2025 Monte Carlo Data,Inc.How we are evaluating quality&reliability:Expert validation Would this explanation help me find the real issue faster?Is this causal