AI HN来自 Hacker News 的 AI 新闻
EN
今天
3天
7天
30天
全部
7 · "chain of thought"
每页
1
Reflections on AI at the End of 2025(antirez.com)
50 ·danielfalbo·6 天前·45 评论
AI SafetyInference Optimization
本文反思了2025年底的AI发展状况,包括“LLM是随机鹦鹉”争论的终结、思维链(CoT)成为LLM的关键改进手段,以及LLM在ARC测试上的进展。文章还讨论了程序员对AI辅助编程的抵触情绪降低、对Transformer替代架构的持续研究、关于AGI路径的不同观点,以及避免AI相关灭绝风险的长期挑战。
2
What I don’t like about chains of thoughts (2023)(samsja.github.io)
56 ·jxmorris12·25 天前·33 评论
本文批评了思维链(CoT)作为增强LLM推理能力的方法。作者认为即使使用CoT,LLM也可能缺乏真正的高级思维能力——以两个任务(挑选普通数字与质数)为例,尽管复杂度不同,但LLM执行这两个任务所需的计算量相同,这表明LLM要么过度思考,要么缺乏真正的推理能力。
3
Why We Think(lilianweng.github.io)
147 ·gmays·3 个月前·30 评论
AI Reasoning
这篇人工智能相关的新闻通过研究人类思维过程和人工智能系统推理能力背后的认知机制,探讨了‘我们为何思考’这一问题。
4
Chain of thought monitorability: A new and fragile opportunity for AI safety(arxiv.org)
134 ·mfiguiere·5 个月前·64 评论
AI Safety
思维链可监控性是一个新发现但脆弱的机会,有望为推进AI安全举措提供支持。
5
Chain of thought empowers transformers to solve inherently serial problems(twitter.com)
73 ·alexmolas·超过 1 年前·2 评论
LLM Research
💡 The story focuses on Chain of Thought (CoT), a reasoning technique for transformers, which falls under the research category as it involves AI reasoning research.
6
Chain of Thought empowers transformers to solve inherently serial problems(arxiv.org)
261 ·krackers·超过 1 年前·184 评论
LLM Research
💡 The story is an ArXiv paper about Chain of Thought, which falls under reasoning research—a subcategory of the research category.
7
OpenAI threatens to revoke o1 access for asking it about its chain of thought(twitter.com)
538 ·jsheard·超过 1 年前·305 评论
AI SafetyOpenAI Ecosystem
💡 The story involves OpenAI threatening to revoke access to its o1 model for users asking about its chain of thought, which relates to mechanistic interpretability—a subtopic of AI safety.
📅周报
Hacker News|Powered by Doubao