当前位置:首页 > 报告详情

03 AICon2025_MarkCollier final.pdf

上传人: Fl****zo 编号:724325 2025-07-01 48页 3.04MB

1、Training,Inference,Agents:Beyond Apps in the AI-Native WorldMark Collier,GM of AI&Infrastructure,Linux FoundationCo-founder of OpenStack&OpenInfra FoundatiomAbout me Co-founder,OpenStack&OpenInfra Foundation,raised$150m to create multi billion-dollar market Now GM,AI&Infrastructure,Linux Foundation

2、Mission:keep the intelligence layer as open and interoperable as the cloudMark Collier15 years of OpenInfra in ChinaFirst OpenStack Trip to ChinaAll over China,OpenStack is managing millions of cores of compute for public&private clouds,and Kata Containers secures critical infrastructure like AliPay

3、2010Today-2025Linux Foundation+OpenInfra FoundationEnsuring every computing era is openOPENINFRA.ORG/BLUEPRINTTHE WORLD RUNS ON OPENINFRATHE WORLD RUNS ON OPENINFRAYOU CANT SEPARATEAI FROMINFRASTRUCTUREAI is putting tremendous demands on infrastructureGoogles Pichai told I/O that the company now pro

4、cesses 480 trillion tokens a month 50more than a year agoSome of the biggest challenges in AI are infrastructureLets talk open source AIJan 27,2025Nvidia drops nearly 17%as Chinas cheaper AI model DeepSeek sparks global tech sell-offJan 27,2025DeepSeeks R1 Launch Shows There Are No Moats Among Large

5、 Language ModelsOpen Source is the right side of historyneeds a new open-source strategySam Altman,CEO of OpenAIJuly 12,2024“In the face of disruptive technology,the moat formed by closed-source systems is short-lived.”Liang Wenfeng,DeepSeekOpen drives innovation,adoption,accessApplicationsFine-Tune

6、d Specialized ModelsFoundationModelsProprietary+Public DataCloud PlatformClosed Source/APIModel HubTooling/LLMOpsSoftwareServicesAI SafetyHardwareOSS Quickly PushesCost DownValue and Innovation Go UpLets talk AI-Native ComputingAI-Native Computing:a simple definitionAI-Native Computing is infrastruc

word格式文档无特别注明外均可编辑修改,预览文件经过压缩,下载原文更清晰!
三个皮匠报告文库所有资源均是客户上传分享,仅供网友学习交流,未经上传用户书面授权,请勿作商用。
本文主要讨论了AI原生计算的三支柱:训练、推理和代理,强调了开放源代码和标准化在AI基础设施中的重要性。关键点如下: 1. AI对基础设施提出了巨大挑战,处理能力需求激增。 2. 开放源代码推动创新和成本降低,是AI发展的正确方向。 3. AI原生计算是为模型而非人类设计的计算基础设施,需支持连续学习、处理大量令牌流和优化大规模推理。 4. 推理是AI商业化的关键路径,对可靠性的要求高于训练。 5. 代理是新型软件范式,通过开放协议使模型输出可编程,实现工作流程自动化。 6. 开放平台需要集体投资,类似OpenStack和Kubernetes在云时代的投资。 引用的核心数据: - Google每月处理480万亿个令牌,比一年前多50倍。 - 推理是历史上最大的工作负载,比训练大50倍。 文章呼吁行业协作,推动AI原生计算栈的发展,确保AI原生时代是开放、创新和繁荣的。
"AI-Native时代如何开启?" 未来趋势还是乌托邦?" "AI计算三大支柱是什么?"
客服
商务合作
小程序
服务号
折叠