Benedict Evans 探讨前沿 AI 模型是否正在商品化、缺乏网络效应，以及随着 OpenAI 和 Anthropic 走向公开市场，投资者真正购买的是什么。
Benedict Evans discusses whether frontier AI models are becoming commoditized, the lack of network effects, and what public investors are actually buying as OpenAI and Anthropic head towards public markets.
要点 · TL;DR
前沿 AI 模型正在商品化，缺乏网络效应，持久价值不确定。 Frontier AI models are commoditizing with no network effects, making durable value uncertain.
聊天机器人体验差；真正采用需要封装应用和明确用例。 Chatbot UX is poor; real adoption needs wrapped apps and clear use cases.
编程是前沿模型目前唯一明确的产品市场契合点。 Coding is the only clear product-market fit for frontier models so far.
核心观点 · Key points
模型正在商品化，没有网络效应，持久价值不确定。 Models are commoditizing with no network effects, making durable value uncertain.
聊天界面用户体验差；真正采用需要封装的应用和用例。 The chatbot interface is a poor UX; real adoption needs wrapped apps and use cases.
编程是前沿模型目前唯一明确的产品市场契合点。 Coding is the only clear product-market fit so far for frontier models.
自动化不可预测地创造新工作，如播客和会计领域所见。 Automation creates new jobs unpredictably, as seen with podcasts and accounting.
大语言模型给出平均答案；价值在于非平均的创造性工作。 LLMs give the average answer; value lies in non-average, creative work.
反共识 · Contrarian takes
AI模型公司可能最终像电信业：巨额资本支出，低利润。 AI model companies may end up like telecoms: huge capex, low profits.
AI的S曲线尚处早期；我们甚至还没问对问题。 The S-curve of AI is early; we don't even know the right questions yet.
大多数人仍是“看电子表格的律师”——非日常用户。 Most people are still 'lawyers looking at spreadsheets'—not daily users.
仅靠更好的模型不会推动大规模采用；产品封装是关键。 Better models alone won't drive mass adoption; product wrapping is key.
AI能力的锯齿状前沿使非专家难以信任。 The jagged frontier of AI capabilities makes it hard for non-experts to trust.
本期章节 · Chapters（共 20）
前沿模型商品化Introduction and the commoditization of frontier models
模型作为商品基础设施Model as Commodity Infrastructure
定价压力与效率乘数Pricing Crunch and Efficiency Multipliers
类比电信与移动通信Analogy with Telecoms and Mobile
对比云与半导体Comparison with Cloud and Semiconductors
光纤类比与解耦Fiber Analogy and Disaggregation
1999年的故事A Story from 1999
低利润转售商论点Thesis on low-margin reseller
S曲线与技术演进S-curve and technology evolution
移动互联网vs桌面互联网Mobile vs Desktop Internet
美国AI公司像中国企业US AI Companies Behaving Like Chinese Companies
模型与工具集成Integration of models with tools
空白屏幕问题与锯齿前沿Blank screen problem and jagged frontier
识别问题与挑战The challenge of identifying problems and solutions
岗位替代与AI答案本质Job displacement and the nature of AI answers