Scale 联合创始人兼 CEO Alexander Wang 探讨为何数据是新代码、Scale 的创立故事，以及他对 AI 风险和加剧不平等问题的看法。
Alexander Wang, co-founder and CEO of Scale, discusses why data is the new code, the founding story of Scale, and his views on AI risks and inequality.
要点 · TL;DR
数据是新代码而非新石油，是 AI 的基本构建块。 Data is the new code, not oil, and is the fundamental building block for AI.
AI 将成为史上最大经济引擎，超越 PC、互联网和智能手机。 AI will be the greatest economic engine, bigger than PC, internet, and smartphone.
未来 2-3 年的 AI 发展将决定全球未来 2-3 十年的格局。 Next 2-3 years of AI development will define the next 2-3 decades globally.
核心观点 · Key points
数据是新的代码，而非新的石油；它是 AI 应用的基本构建块。 Data is the new code, not the new oil; it is the fundamental building block for AI applications.
AI 将比 PC、互联网和智能手机革命更大，可能是有史以来最伟大的经济引擎。 AI will be bigger than the PC, internet, and smartphone revolutions, potentially the greatest economic engine ever.
未来 2-3 年的 AI 发展将定义全球未来 2-3 十年，包括经济与地缘政治。 The next 2-3 years of AI development will define the next 2-3 decades globally, economically and geopolitically.
企业应利用专有数据微调基础模型，以获得独特能力。 Enterprises should leverage proprietary data to fine-tune base models for unique capabilities.
AI 风险分为三类：AI 本身、滥用以及劳动力替代等二阶效应。 AI risks fall into three buckets: AI itself, misuse, and second-order effects like labor displacement.
图灵陷阱警告不要将 AI 仅视为人类替代品；人机混合系统创造更多价值。 The Turing Trap warns against viewing AI solely as human replacement; hybrid human-AI systems create more value.
反共识 · Contrarian takes
数据不像石油那样是商品；它更丰富，需要深思熟虑的策略来整合不同来源。 Data is not a commodity like oil; it is richer and requires a thoughtful strategy to stitch together diverse sources.
AI 规模扩张的最大限制因素不是算力，而是数据和后训练所需的人类专家。 The most limiting factor for AI scaling is not compute but data and human expert availability for post-training.
AI 模型将成为历史上最大的投资之一，堪比粒子加速器。 AI models will become some of the largest investments in history, comparable to particle accelerators.
几十年内，人们将花费超过一半的时间与 AI 模型互动，而非人类。 People will spend more than half their time interacting with AI models rather than humans within decades.
开源模型对于充分的经济影响是必要的，但闭源模型也将占据主导。 Open source models are necessary for full economic impact, but closed source models will also dominate.
“古怪”是积极特质；正常人处于钟形曲线内，不太可能产生差异化影响。 Being 'weird' is a positive trait; normal people are in the bell curve and less likely to have differentiated impact.
本期章节 · Chapters（共 36）
引言：数据是新石油Introduction and Data as the New Oil
数据作为 AI 基础设施Data as Infrastructure for AI
企业专有数据机遇Enterprise Opportunity with Proprietary Data
给高管和创始人的建议Advice for Executives and Founders
与过去技术趋势对比Comparison to Past Tech Trends
AI 的经济潜力Economic potential of AI
AI 与核：遏制与威慑AI vs Nuclear: Containment and Deterrence
AI 趋势：巨额投资与人机交互Inevitable AI Trends: Massive Investment and Human-Model Interaction
AI 优势与数据可用性AI strengths and data availability
限制因素：算力与数据Limiting factors: compute and data
平台风险与供应链Plateau risk and supply chain
地缘政治风险：台湾Geopolitical risk: Taiwan
开源与闭源Open source vs closed source
开源模型与经济影响Open source models and economic impact
图灵陷阱The Turing Trap
对 AI 改进的误解Misconceptions about AI improvement
AI 地缘政治博弈Geopolitical battle in AI
全球 AI 竞争与地缘政治赌注Global AI Competition and Geopolitical Stakes
社会对 AI 准备程度的共识Societal agreement on AI readiness
Scale 的创立洞见Founding insight of Scale
早期职业生涯与麻省理工Early career and MIT
来自亚马逊的启发Inspiration from Amazon
不可预测的转型：亚马逊与英伟达Unpredictable Reinvention: Amazon and Nvidia
面试问题：最艰难的工作Interview Question: Hardest Work
顶级招聘特质Top Hiring Traits
大品牌声望与影响力Big Brand Prestige vs. Impact
非共识创业洞见Non-Consensus Startup Insight
创业文化与人才Startup culture and talent
Scale 作为基础设施提供商Scale's role as infrastructure provider