Doctorow 《Textured》：LLM 统计平滑隐藏了“意识参与”才是产出意外的唯一条件
- ID: 3156d329
- 原文链接: https://pluralistic.net/2026/09/18/surprise/
- 作者: Cory Doctorow
- 日期: 2026-09-18
- 分类: learning
- 来源类型: article
- 标签: llm, ai-hype, epistemics, doctorow, essays
- 质量评分: 4/5
- 抓取时间: 2026-09-19T22:35:00Z
中文导读
Doctorow 9 月 18 日的 daily links 长篇论述以一个具体反例为刃入：“你能预测妻子下一句说什么” 与 “你了解妻子” 是两件不同的事，在它说 “我要离婚” 时立刻失语。他把全部 AI hype 叙事回溯到同一根因：大家低估了自然语句里的统计规律有多强、高估了“意识”的稀缺性，模型只在 “平整度”这一项上超越；但平整之下的纹理 —— 惊喜、价值、生产力——才是产出下一个意外的唯一条件，训练数据本身把意外剔除。他进一步指出：“AI 越接近 AGI”这个资本叙事所买的是 “理解力叙事”，实际产出里这部分为零。作者是 AI 工具论支持者中最不可能“不行”口头禄的人，论述质量高于常见 hype 反对者。
为什么值得关注
Doctorow：LLM 只能平整，意外在纹理里，验证意识参与才是产出下一个意外的唯一条件
关键信息
- 标题：Doctorow 《Textured》：LLM 统计平滑隐藏了“意识参与”才是产出意外的唯一条件
- URL：https://pluralistic.net/2026/09/18/surprise/
- 发布时间：2026-09-18
- 分类：learning
- 标签：llm, ai-hype, epistemics, doctorow, essays
English Summary
Cory Doctorow's 18 Sep 2026 daily links opens with a concrete counter-example: predicting your spouse's next sentence is not the same as knowing them — it breaks the moment she says 'I want a divorce.' He traces every AI hype narrative to one root cause: we underestimate how much statistical regularity lives in natural language and overestimate how scarce 'consciousness' is. Models win on flatness; everything that produces surprise, value, and productivity lives below the texture, which the training data has already stripped out. He closes that 'AI is approaching AGI' is a story the AI capital narrative is buying itself — the same story with zero presence in actual output. Doctorow is an AI tooling supporter; the essay reads as a coolant for people battered by weekly 'passes Turing test' headlines.