Anthropic 的超级碗广告：针对 OpenAI 的蓝龟壳攻击？

Anthropic's Super Bowl Ads: A Blue Shell Attack on OpenAI?

肖尔托·道格拉斯 Sholto Douglas · TBPN · 2026-02-05 · 约 196 分钟 · 原视频 ↗

打开互动全文版（中英对照 + 朗读 + 问答）→

本期速览 · Overview

TVPN 主持人讨论 Anthropic 有争议的超级碗广告，将其比作政治攻击广告和马里奥赛车的蓝龟壳，同时预告了对 Sam Altman 和 Dan Barkello 的采访。

TVPN hosts discuss Anthropic's controversial Super Bowl ads, comparing them to political attack ads and Mario Kart's blue shell, while previewing interviews with Sam Altman and Dan Barkello.

要点 · TL;DR

• AI 能力过剩，编排和工具比原始智能更重要。
  AI capability overhang is massive; orchestration and tools now matter more than raw intelligence.
• 算力是新石油，芯片供应是当前 AI 进展的瓶颈。
  Compute power is the new oil; chip supply is the current bottleneck for AI progress.
• 广告可以扩大消费者 AI 的普及，超越仅靠订阅的模式。
  Advertising can scale consumer AI access beyond subscriptions alone.

核心观点 · Key points

• AI 模型现在有巨大的能力过剩；更好的工具和编排比原始智能更重要。
  AI models now have massive capability overhang; better tools and orchestration matter more than raw intelligence for now.
• 未来的工作涉及管理 AI 智能体团队，在更高的抽象层次上操作。
  The future of work involves managing teams of AI agents, operating at higher levels of abstraction.
• 算力是新的石油；芯片供应是当前 AI 进展的瓶颈。
  Compute power is the new oil; chip supply is the current bottleneck for AI progress.
• 广告是规模化消费者 AI 访问的成熟模式；仅靠订阅会限制覆盖范围。
  Advertising is a proven model to scale consumer AI access; subscriptions alone limit reach.
• 软件没有消亡，而是在转型；由于智能体集成，每家公司现在都是 API 公司。
  Software is not dead but transforming; every company is now an API company due to agent integration.
• 像 METR 这样的长任务时间基准很关键，但智能体集群可能需要新的指标，如 GDP 影响。
  Long task horizon benchmarks like METR are key, but agent swarms may require new metrics like GDP impact.

反共识 · Contrarian takes

• Anthropic 的超级碗广告是欺骗性的政治攻击广告，而非诚实的竞争营销。
  Anthropic's Super Bowl ads are deceptive political attack ads, not truthful competitive marketing.
• AI 生成的视频对被动观看没有吸引力；其主要用途是群聊中的个性化表情包。
  AI-generated video is not compelling for passive viewing; its main use is personalized memes in group chats.
• 争夺 30 亿月活用户的竞赛已经结束；Anthropic 应专注于类似 iPhone 的高端利基市场。
  The race to 3 billion monthly active users is over; Anthropic should focus on a premium iPhone-like niche.
• 世界模型不在通往 AGI 的关键路径上；编程和 AI 研究是直接路线。
  World models are not on the critical path to AGI; coding and AI research are the direct route.
• 可能出现仅软件的奇点，数字任务快速进步而物理世界滞后。
  A software-only singularity is possible where digital tasks advance rapidly while physical world lags.
• 太空数据中心在未来 10 年内不会为 OpenAI 提供有意义的算力。
  Space data centers will not provide meaningful compute for OpenAI in the next 10 years.

本期章节 · Chapters（共 97）

• 开场与嘉宾介绍 Opening and Guest Introduction
• 对 Anthropic 广告活动的反应 Reactions to Anthropic's ad campaign
• 广告活动战略分析 Strategic analysis of the campaign
• LLM 的广告与订阅模式对比 Ads vs. subscription model for LLMs
• 比较 Anthropic 与 OpenAI Comparing Anthropic and OpenAI
• Anthropic 商业模式与广告 Anthropic's Business Model and Advertising
• CrowdStrike 赞助插播 CrowdStrike Sponsor Break
• Claude Opus 4.6 发布 Claude Opus 4.6 Announcement
• Claude 的 SVG 能力 Claude's SVG Capabilities
• Anthropic 推特账号问题 Anthropic's Twitter Handle Issue
• Paul 迁移至 Blue Sky Paul's migration to Blue Sky
• Claude 代码采用统计 Claude code adoption stats
• 下期嘉宾与 Doug 的瘾 Upcoming guest and Doug's addiction
• OpenAI 推出 Frontier AI 同事 OpenAI unveils Frontier for AI co-workers
• 赞助插播：Reream Sponsor break: Reream
• Frontier 详情与企业集成 Frontier details and enterprise integration
• Frontier 图形与生态系统 Frontier graphic and ecosystem
• GPT-5.2 基准与时间线 GPT-5.2 Benchmark and Time Horizon
• Nvidia GB200 与 AI 硬件发布 Nvidia GB200 and AI Hardware Rollout
• 谷歌财报与 AI 资本支出 Google Earnings and AI Capex
• Gemini 采用与用户数 Gemini adoption and user numbers
• 谷歌财务表现与投资者情绪 Google's financial performance and investor sentiment
• YouTube Shorts 与内容策略 YouTube shorts and content strategy
• 介绍与新闻概览 Introduction and News Overview
• Nvidia 游戏芯片延迟与内存短缺 Nvidia Gaming Chip Delay and Memory Shortage
• 内存芯片价格与市场瓶颈 Memory chip prices and market bottlenecks
• 当前限制与智能体集群 Current limitations and agent swarms
• 预测与未来指标 Predictions and future metrics
• 新研究实验室与行业研究 New research labs and industry research
• 数据是新石油与规模化 Data as the new oil and scaling
• AI 投资的权衡 Trade-offs in AI investment
• OpenClaw 与开源创新 OpenClaw and open source innovation
• 社交网络与 AI 智能体 Social networks and AI agents
• Sora 与视频生成 Sora and video generation
• 迪士尼合作与角色生成 Disney partnership and character generation
• 病毒式图像生成与发布策略 Viral Image Generation and Rollout Strategy
• Anthropic 超级碗广告与批评 Anthropic's Super Bowl Ads and Criticism
• 语音模式暂停与瓶颈 Voice Mode Pausing and Bottlenecks
• 模型 IQ 上限 Upper Bound on Model IQ
• 开场闲聊与太空数据中心 Initial banter and space data centers
• 广告发布与优先级排序 Ads rollout and prioritization
• 内部写作模型与优先级 Internal writing model and prioritization
• 媒体审视与内部平静 Media scrutiny and internal calm
• 结束语与介绍 Sholto Closing remarks and introduction of Sholto
• 模型发布与能力 Model Release and Capabilities
• 任务范围与 Meter 评估 Task Horizon and Meter Eval
• 编排与抽象 Orchestration and Abstraction
• 管理多个 AI 智能体 Managing multiple AI agents
• 游戏开发与规模化时代 Game development and Age of Scaling
• 纯软件奇点 Software-only singularity
• AI 语境中定义奇点 Defining Singularity in AI Context
• AI 进步的瓶颈 Bottlenecks to AI Progress
• 超级碗广告与计算限制 Super Bowl Ad and Compute Constraints
• 平衡增长与瓶颈 Balancing Growth and Bottlenecks
• 前向部署工程师的角色 Role of Forward Deployed Engineers
• 数据与专业知识 Data vs Expertise
• UI 与产品设计 UI and Product Design
• 设备端计算与智能民主化 On-device computation and intelligence democratization
• 介绍 Dan Barkella 与 T1 Energy Introduction of Dan Barkella and T1 Energy
• 中国太阳能增长与美国制造 China's solar growth and US manufacturing
• 规模化太阳能制造的瓶颈 Bottlenecks in scaling solar manufacturing
• 主要买家与需求驱动 Main buyers and demand drivers
• 未来扩展计划与垂直整合 Future scaling plans and vertical integration
• 太空太阳能板 Solar panels for space
• 做空者与压力 Short sellers and stress
• 商品计划 Merchandise plans
• 公司增长与招聘 Company growth and hiring
• 股票表现与结束 Stock performance and closing
• AI 超级用户与怀疑者 AI Super Users and Skeptics
• 学前应用与 AI 垃圾 Preschool applications and AI slop
• 建立直接面向消费者品牌 Building a Direct-to-Consumer Brand
• 思科合作与 AI 峰会 Cisco Partnership and AI Summit
• 每县马匹数与自动驾驶汽车 Horses per County and Autonomous Cars
• Meta 的鳄梨模型 Meta's Avocado Model
• 开场闲聊与市场讨论 Opening banter and market talk
• Ivan 的病毒式互动与介绍 Ivan's viral interaction and introduction
• 编排市场与智能体需求 Orchestration market and agent needs
• 产品演进与公司状态 Product evolution and company state
• 为何智能体需要计算机 Why agents need computers
• 智能体用例：代码、浏览器、强化学习 Agent Use Cases: Code, Browser, RL
• 介绍 Savvy Introduction to Savvy
• 日常生活与搬到洛杉矶 Day in the life and move to LA
• 商业哲学与风投文化 Business philosophy and VC culture
• 在 HQ 获得工作 Getting the job at HQ
• 媒体科技混合与竞争 Media-tech hybrid and competition
• 对 Savvy 的承诺与网络反弹 Commitment to Savvy and online backlash
• 社区与用户基础 Community and User Base
• 联合与营销 Syndication and Marketing
• 变现与订阅 Monetization and Subscriptions
• 大型节目与变现时刻 Big Shows and Monetization Moments
• 能源与计算瓶颈 Energy vs Compute Bottleneck
• 组建团队 Building the team
• 把关与策展 Gatekeeping and Curating
• 应用评论与反馈 App Review and Feedback
• 赞助与商品 Sponsorship and Merchandise
• 反思谦卑经历 Reflections on Humbling Experiences
• 结束语 Closing Remarks

阅读全文双语转录 →