资讯与观察
AI 行业动态、工程实践、产品方法和社区故事。
维基百科仍在坚持清理 AI 生成内容
AI 生成内容越来越难以识别，但并未阻止维基百科对此进行清理。 2023 年 12 月，几个维基编辑在 Discord 上聊天，发现彼此都见过同一类文章：满篇"作为……的见证"（stands as a testament）、"丰富的文化遗产"，结尾必定带个结论段，或者干脆来一节叫"这为什么重要"。这是 2023 到 2024 年初 ChatGPT 的招牌文风，在维基...
Attackers compromised 42 GitHub repositories by using files too large to be searched
A security research firm reported Monday that it had identified breaches of 42 third-party GitHub repositories in which attackers had deployed payload blobs large enough “to evade the GitHub search and API paths normall…
18 分钟 · InfoWorld
Introducing Fern
Fern is Mirendil’s first model. It is pretrained from scratch and supports a
1 分钟 · Hacker News 最新LATEST STORIES
最新内容
Compressing LLMs Without Compromise精选导读
by Aditya Kamath, Simon Peter on Jul 23, 2026 | Tags: Accelerators, Machine Learning
Show HN: Software Bloat Calculator精选导读
暂无摘要
Wikipedia: Signs of AI Writing精选导读
As of the April 2008 census, the population of Douera stood at approximately 56,998 inhabitants, creating a lively community within its borders. Situated in the central-north region of the country, Douera enjoys close p…
Tailscale Tailworm: SSH socket forwarding ran as root, fixed in 1.104.4精选导读
暂无摘要
After its Odyssey, IMAX hits the big time精选导读
暂无摘要
开源终端模拟器 Ghostty 发布 1.4.0，优化内存占用 & 性能精选导读
Ghostty 1.4.0 出来了，7 个月、172 位贡献者、2889 个提交。这个版本最值得说的是一个先让人意外、再让人拍大腿的发现：macOS 内核把每一次 pty 读取都卡在 1 KiB 左右，不管你给多大的缓冲。 这一条直接解释了以前为什么慢——Ghostty 来一次读就处理一次，每读 1 KiB 就要付一次解析开销，而写输出的程序还得在这期间...
Genie 4 来了，谷歌 DeepMind 的世界模型开始同步生成声音精选导读
谷歌旗下 AI 公司 DeepMind 宣布推出 Genie 4，世界模型这条线又往前走了一格。最实质的进步不是画质，是声音——Genie 4 是原生多模态的，画面、音乐、环境声、对话可以同时生成。你在模拟里推倒一块石头，它不只是画出来，还会响。 技术上它基于 Gemini 家族：Gemini Omni 负责让模拟锚定真实世界的样貌和物理行为，其...
Writing AI agents in YAML instead of Python精选导读
There are plenty of frameworks for building AI agents, such as LangChain or LlamaIndex. But as you add models, tools, and features, the agent quickly grows into a lot of code to maintain. More code means more places for…
Ask HN: Why aren't maintainers worried about the Copyright status of LLM output?精选导读
LLM output, according to court rulings IIUC, cannot be copyrighted. This should worry both proprietary software companies and open source projects alike (because you can't put a license on something you don't hold the r…