geohot:AI 没有护城河,模型折旧比想象中快,中国没有 AI 狂热
Source: geohot | 2026-04-22 URL: https://geohot.github.io/blog/jekyll/update/2026/04/22/ai-has-no-moat.html
Summary (Chinese)
geohot 撰文称「AI 没有护城河」。核心论点:Cursor 60B 估值收购是因为 FOMO,实际没人真正用;opencode 确实是最好的 coding agent 但写一个类似的并不难;真正的壁垒在模型,但模型折旧极快,开源已有详细训练指南。顶级闭源模型(GPT-5.4、Opus 4.7)成本是次级模型(Kimi K2.6、GLM 5.1)的 10 倍,但能力差距只有约 6 个月。判断「AGI singularity」是硅谷特有狂热,中国 AI 是正常指数增长曲线,AI 泡沫会自行消退,希望科技圈「死得快一点而不是漫长而痛苦的死亡」。
Summary (English)
geohot argues AI has no moat. Core claims: Cursor's $60B acquisition reflects FOMO not real usage; opencode is best coding agent but not hard to replicate; real moat is models but they depreciate fast. Top closed models cost 10x more than second-tier (Kimi K2.6, GLM 5.1) but only ~6 months ahead. AGI singularity is a Silicon Valley peculiarity; China's AI follows normal exponential curves.
注:原文抓取失败(web_fetch 被封锁),此 content 文件基于 AK RSS Digest 摘要整理。原始内容请参考 AK RSS Digest 源文。