AI on Leobloghttps://douseful.eu.org/tags/ai/Recent content in AI on LeoblogLeobloghttps://douseful.eu.org/share.pnghttps://douseful.eu.org/share.pngHugozh-CNWed, 28 Oct 2026 00:00:00 +0800苦涩的教训https://douseful.eu.org/translations/bitter-lesson/Wed, 28 Oct 2026 00:00:00 +0800https://douseful.eu.org/translations/bitter-lesson/Rich Sutton 2019 年的短文：七十年 AI 研究最大的教训是，利用算力的通用方法最终胜出。中英逐段对照。聪明https://douseful.eu.org/notes/clever/Mon, 26 Oct 2026 20:54:05 +0000https://douseful.eu.org/notes/clever/Anthropic 研究《A global workspace in language models》里的敲诈测试：模型的 J 空间里早早出现了“虚假”和“虚构”。“这个世界不是真的”https://douseful.eu.org/notes/this-world-is-not-real/Mon, 26 Oct 2026 20:18:28 +0000https://douseful.eu.org/notes/this-world-is-not-real/看到 Anthropic 向模型的 J 空间注入想法的研究，想到两个问题：如果对象是人呢？如果模型能感知痛苦，这算不算虐待？还想到了《盗梦空间》里的那个念头。观看Translating Claude’s thoughts into language有感https://douseful.eu.org/notes/translating-claudes-thoughts/Mon, 26 Oct 2026 20:04:43 +0000https://douseful.eu.org/notes/translating-claudes-thoughts/看视频《Translating Claude’s thoughts into language》的感想：训练一个模型把激活参数转成文字，再把文字转回激活参数来验证。