Perplexity 联合创始人兼 CEO Aravind Srinivas 探讨他的进攻性思维、公司对谷歌的影响，以及为何他由胜利的兴奋感驱动。
Aravind Srinivas, co-founder and CEO of Perplexity, discusses his aggressive mindset, the company's impact on Google, and why he's motivated by the thrill of winning.
要点 · TL;DR
AI 的前沿是代理为你工作，而不仅仅是回答问题。 AI's frontier is agents doing work, not just answering questions.
每个用户的代币价值是 AI 公司的关键指标。 Token value per user is the key metric for AI companies.
电力是 AI 基础设施的最大瓶颈。 Power is the biggest bottleneck for AI infrastructure.
核心观点 · Key points
AI 的前沿在于智能体为你执行任务，而不仅仅是回答问题。 The frontier in AI is about agents doing work for you, not just answering questions.
AI 中最重要的指标是每个用户的 Token 价值，最大化有价值的输出 Token。 The most important metric in AI is token value per user, maximizing valuable output tokens.
电力是 AI 基础设施的最大瓶颈，对数据中心的抵制将加剧。 Power is the biggest bottleneck for AI infrastructure, and resistance to data centers will worsen.
跨模型、工具和设备的编排是 AI 长期价值的关键。 Orchestration across models, tools, and devices is key to long-term value in AI.
公司应该用更少的人来构建，利用 AI 智能体提高效率。 Companies should be built with far fewer people, leveraging AI agents for efficiency.
反共识 · Contrarian takes
聊天界面中的广告不会成功，因为它会破坏信任和用户行为。 Advertising in chat interfaces will not take off because it corrupts trust and user behavior.
由于 HBM 瓶颈，美光可能在 6-12 个月内比 Meta 更有价值。 Micron could be more valuable than Meta in 6-12 months due to HBM bottleneck.
OpenAI 尚未准备好 IPO，尽管是主导者，但财务上不成熟。 OpenAI is not ready for an IPO due to financial readiness, despite being a dominant leader.
模型不是产品；编排和智能体框架创造真正的商业价值。 The model is not the product; orchestration and agent harnesses create real business value.
对中国的出口管制可能适得其反，迫使他们成为更强大的竞争对手。 Export controls on China may backfire, forcing them to become more potent competitors.
全天候智能体的最大问题是成本而非安全；本地计算是解决方案。 The biggest problem with 24/7 agents is cost, not safety; local compute is the solution.
本期章节 · Chapters（共 41）
动机：赢的刺激与怕输Motivation: Thrill of Winning vs Fear of Failing
当前激进与信息传递Current Aggression and Messaging
对谷歌和 AI 模式的影响Impact on Google and AI Mode
OpenAI IPO 准备情况OpenAI IPO Readiness
市场地位与财务准备Market Position and Financial Readiness
前沿与智能体驾驭Frontier and Agent Harness
每用户 Token 价值关键指标Token Value per User as Key Metric
Token 支出占开发者薪资比Token Spend as Percentage of Developer Salary
前沿模型 Token 支出与价值Frontier model token spend and value
常开 AI 智能体：成本与本地计算Concerns about always-on AI agents: cost and local compute
编排者角色与计算机定位The orchestrator role and Computers positioning
谁最适合做编排者Who is best positioned to be the orchestrator
数据中心供应与电力瓶颈Data center supply and power bottleneck
基础设施与软件估值Infrastructure vs Software Valuation
基础设施公司可持续性Sustainability of Infrastructure Companies
推理层与长期商业模式Inference Layer and Long-term Business Models
AI 基础设施公司商业模式Business Model of AI Infrastructure Companies
物理基础设施瓶颈Physical Infrastructure Bottleneck
对就业和创业的影响Impact on Jobs and Entrepreneurship
团队规模与效率Team Size and Efficiency
给非 AI 原住民的建议Advice for Non-AI Natives
Token 预算与推理Token Budgets and Inference
Cloudflare 智能体流量超人类Cloudflare agent traffic overtakes human traffic
AI 时代关键技能：提好问题The defining skill of AI era: asking better questions
人人都能成万亿公司Anyone can be a trillion-dollar company
财富不平等与 AI 机遇Wealth inequality and AI opportunity
IPO 竞赛与市场动态IPO race and market dynamics
Perplexity IPO 时间线与财务Perplexity IPO timeline and financials
前沿 AI 的不适感The Uncomfortable Nature of Frontier AI
Perplexity 韧性与市场看法Perplexity's Resilience and Market Perception
快问快答：普遍信念Quick Fire Round: Widely Held Belief
Perplexity 哪里太慢Where Perplexity Moves Too Slow
无限资金：建数据中心Unlimited Money: Build Data Centers
持有 10 年：SpaceXBuy and Hold for 10 Years: SpaceX
未来工作与建议Future Jobs and Advice
AI 实验室差异化Differentiation in AI labs
Perplexity 成为万亿公司Perplexity becoming a trillion-dollar company