亚历山大·王分享他从数学奥赛到创办 Scale AI 的经历，核心洞察在于数据是训练模型的关键瓶颈。
Alexander Wang shares how he went from math competitions to building Scale AI, driven by the insight that data was the key bottleneck for training models.
要点 · TL;DR
智能将变得充裕，远见与雄心成为稀缺资源。 Intelligence will be abundant; vision and ambition become the scarce resources.
拥有严格评估的小型智能体团队，可以胜过大型工程团队。 Small agent teams with rigorous evals can outcompete large engineering organizations.
创业者需要系统思维与哲学罗盘，以驾驭指数级 AI 浪潮。 Builders need systems thinking and a philosophical compass to navigate exponential AI waves.
核心观点 · Key points
AI 进步处于陡峭的指数曲线上；智力与能动性将变得充裕，因而愿景与野心成为稀缺资源。 AI progress is on a steep exponential; intelligence and agency become abundant, so vision and ambition become scarce.
具备清晰评测与指标的智能体式循环，能让小型智能体集群胜过大型工程团队。 Agentic loops with clear evals and metrics let small agent swarms outperform large engineering teams.
前沿 AI 是研究工作；实验室需要人才密度和能与指数增长复利的运营模式。 Frontier AI is research; labs need talent density and operating models that compound with exponential growth.
未来十年对人类的变化将超过过去一百年；建设者必须同时应对风险与机遇。 The next decade will transform humanity more than the past hundred years; builders must address both risk and opportunity.
对非共识信念的坚持和内在罗盘，是穿越创业噪音与混乱的关键。 Conviction in non-consensus beliefs and an internal compass are essential to navigate startup noise and chaos.
Meta 的愿景是数十亿个人超级智能通过去中心化生态扩展人类能动性。 Meta's vision is billions of personal superintelligences expanding human agency through a decentralized ecosystem.
反共识 · Contrarian takes
数据曾被当作不值得做的生意，如今怀疑者却称其至关重要；当初错过的投资人如今反过来吹捧它。 Data was a dismissed business that skeptics now call critical; the same investors who passed now celebrate it.
如今创业公司是歌利亚对歌利亚；AI 智能体让它们无需巧妙的“大卫”切入角度便能胜过在位者。 Startups are now Goliath vs Goliath; AI agents let them outcompete incumbents without needing a clever David angle.
即使模型进步停滞，仅凭技术扩散就会带来几十年的巨变；能力本身不是瓶颈。 Even if model progress froze, diffusion alone would drive decades of upheaval; capabilities are not the bottleneck.
围绕超级智能何时到来的争论是短视的；不可避免的指数趋势比预测日期更重要。 Timing debates about superintelligence are shortsighted; the inevitable exponential is more important than forecast dates.
全力押注“文字脑”是错误的；系统思维和哲学罗盘仍然必不可少。 Going all-in on 'word cell' is a mistake; systems thinking and a philosophical compass are still essential.
最好的 AI 产品尚未被创造；每一波浪潮都比上一波大十倍，而智能体式机制是平淡无奇的，并非魔法。 Best AI products are unbuilt; each wave is ten times bigger, and agentic mechanics are mundane, not magical.
本期章节 · Chapters（共 16）
早期生活与Scale之路Early life and path to Scale
从AI代理转向数据The pivot from AI agents to data
数据瓶颈与投资者教训Data bottleneck and investor lessons
创办公司Building a Company
AI时代的创业Startups in the AI Age
Meta的超级智能Superintelligence at Meta
Meta Spark与前沿实验室Meta Spark and the Frontier Lab
在Meta建前沿实验室Building a Frontier Lab at Meta
模型成本与AI浪潮Model Cost and the AI Wave
使用Muse Spark与新HarnessUsing Muse Spark and the Upcoming Harness