思想与社会2025 · 分享于 Hacker News
Emergent Misalignment: Narrow finetuning can produce broadly misaligned LLMs
探索《Emergent Misalignment: Narrow finetuning can produce broadly misaligned LLMs》的原文与 Hacker News 社区讨论，查看来源并加入阅读清单。
从论文，走向讨论
看看社区如何讨论这个想法。
来自 Hacker News 的原始讨论。不同的视角、疑问，以及值得继续思考的线索。
最多展示 100 条相关讨论。论文可能出现在评论中，并非原帖投稿内容。
在讨论中发现的引用
按 HN 条目顺序展示最近的 50 条引用来源。
Yspeckx ·
It's insulting to read AI-generated blog posts
1,300 赞同540 评论
YCynddl ·
Training language models to be warm and empathetic makes them less reliable
358 赞同375 评论
Yotterley ·
DeepSeek writes less secure code for groups China disfavors?
265 赞同174 评论
Ymartythemaniak ·
Emergent Misalignment: Narrow finetuning can produce broadly misaligned LLMs
181 赞同48 评论
Ynsoonhui ·
The new science of “emergent misalignment”
126 赞同67 评论
YGodotX ·
LLM's Illusion of Alignment
56 赞同38 评论
YPhilpax ·
Anthropic Cofounder Chris Olah's Remarks on Pope Leo XIV's "Magnifica Humanitas"
26 赞同12 评论
Ycgorlla ·
Show HN: Distilling DeepSeek into GPT-OSS doesn't transfer censorship. Try it
7 赞同0 评论
Ykeeda ·
DeepSeek writes insecure code if prompt mentions topics restricted in China
5 赞同1 评论
Ymektrik ·
Where every major LLM stands politically
3 赞同5 评论