The Claude Delusion
- ID: 53398a68
- 原文链接: https://pluralistic.net/2026/09/21/sunsetting
- 作者: Cory Doctorow
- 日期: 2026-09-21
- 分类: models
- 来源类型: article
- 标签: anthropic, claude, ai-hallucination, doctorow, critique, writing-with-llm
- 质量评分: 4/5
- 抓取时间: 2026-09-21T23:30Z
中文摘要
Doctorow 9 月 21 日 daily links 里发The Claude Delusion核心论点：人类倾向于把意图塞进 LLM 的输出，这本身就是幻觉引用 Mark Fisher那里应该什么都没有，却出现了东西形容这种感觉；模型是从统计里挤出来的文字，背后没有真作者判断：一旦把模型输出当成某人在说话就已经被骗，写作和编辑工作里这条线要划清引用了自己在 Locus 谈过的故事就是强行给无生命物体套上作者意图的折中方案
English Abstract
Cory Doctorow's September 21 daily-links post argues the real hallucination belongs to us: humans reflexively project intention onto LLM outputs. He cites Mark Fisher's 'something appearing where there should be nothing' to characterize the sensation; LLMs are statistical text with no author waiting at the other end. He links back to his own Locus essay that stories are a compromise forcing authorial intent onto inanimate things chatting with an LLM is like talking to a novel character, except novels have a real author and LLMs only have matrix math. The takeaway he draws for writers and editors is to keep this line hard: once you start treating model output as 'someone speaking,' you've already been fooled.
为什么值得关注
Doctorow: 真正幻觉的不是 LLM 是人类自己写作编辑必须划清这条线
Obsidian 证据摘要
来源: OpenClaw定时任务/AK-RSS-Digest（89源精选）/2026-09-21-AK-RSS-Digest（89源精选）.md + 对应 evidence-2026-09-21 文件