{"pageContext":{"index":297,"total":364,"name":"2025-01-11","items":[{"node":{"frontmatter":{"title":"AI 辅助编码将如何改变软件工程：残酷的事实","publish":true,"cover":"https://substackcdn.com/image/fetch/w_1456,c_limit,f_webp,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F9d618b4e-8fff-447f-b9e6-b46b62deafd4_1456x819.png","showCover":null,"date":"2025-01-11","category":"AI","url":"https://newsletter.pragmaticengineer.com/p/how-ai-will-change-software-engineering","author":"Gergely Orosz, Addy Osmani","translator":"","tags":["research"],"priority":1},"html":"
\n

经验丰富的软件工程师在未来可能比现在更受欢迎。 我们在 AI 工具中看到的共同主题是高级及以上工程师如何更有效地使用这些工具，因为他们可以更好地“瞄准”这些工具。

\n

作为一名软件工程师，掌握 AI 辅助开发将使您的工作效率更高，也更有价值。

"}},{"node":{"frontmatter":{"title":"Apple Photos 在 iOS 18 和 macOS 15隐私问题","publish":true,"cover":"https://lapcatsoftware.com/articles/2024/12/3-iPhone.png","showCover":null,"date":"2025-01-11","category":"mobile","url":"https://lapcatsoftware.com/articles/2024/12/3.html","author":"Jeff Johnson","translator":"","tags":["ios"],"priority":2},"html":"
\n

\"相片增强视觉搜索\"默认开启，建议关闭

"}},{"node":{"frontmatter":{"title":"Nodejs的10 亿行挑战","publish":true,"cover":"","showCover":null,"date":"2025-01-11","category":"nodejs","url":"https://jackyef.com/posts/1brc-nodejs-learnings","author":"jackyef","translator":"","tags":["performance"],"priority":1},"html":"
\n

分析有助于避免猜测\n使用字节和整数往往比使用字符串和浮点数更快\nNode.js 中的大多数内置函数已经足够快了，但根据实际情况有机会针对特定情况进行优化\n如果我们以后决定利用多个内核，则以一种易于并行化的方式编写代码

"}},{"node":{"frontmatter":{"title":"我们在 2024 年学到的关于 LLM 的事情","publish":true,"cover":"","showCover":null,"date":"2025-01-11","category":"AI","url":"https://simonwillison.net/2024/Dec/31/llms-in-2024/","author":"Simon Willison","translator":"","tags":["LLM"],"priority":1},"html":"
\n

GPT-4 壁垒被全面打破\n其中一些 GPT-4 模型可以在我的笔记本电脑上运行\n由于竞争和效率的提高，LLM 价格暴跌\n多模态视觉很常见，音频和视频开始出现\n语音和实时摄像机模式让科幻小说栩栩如生\n提示驱动的应用程序生成已经是一种商品\n对最佳模型的普遍访问仅持续了短短几个月\n“Agent”还没有真正出现\n评估真的很重要\nApple Intelligence 很糟糕，Apple 的 MLX 库很棒\n推理扩展“推理”模型的兴起\n环境影响有所改善\n环境影响变得更糟\nslop的一年\n合成训练数据效果很好\n不知何故，LLM 变得更加难用\n知识分布非常不均匀\nLLM 需要更好的批评\n2024 年我博客上的所有内容都标有“llms”

\n

DeepSeek v3 真正令人印象深刻的是训练成本。该模型在 2788000 个 H800 GPU 小时上进行了训练，估计成本为 5576000 USD

\n

Slop 描述了 AI 生成的内容，这些内容既未经请求又未经审核.

"}},{"node":{"frontmatter":{"title":"AI机器人成了新的DDos攻击","publish":true,"cover":"","showCover":null,"date":"2025-01-11","category":"devops","url":"https://pod.geraspora.de/posts/17342163","author":"Dennis Schubert","translator":"","tags":["AI"],"priority":1},"html":"
"}},{"node":{"frontmatter":{"title":"Node 23即将发布默认支持typescript","publish":true,"cover":"","showCover":null,"date":"2025-01-11","category":"nodejs","url":"https://www.totaltypescript.com/typescript-is-coming-to-node-23","author":"Matt Pocock","translator":"","tags":["typescript"],"priority":1},"html":"
"}},{"node":{"frontmatter":{"title":"react 全栈 - TanStack Start(beta) 简介","publish":true,"cover":"https://github.com/tanstack/router/raw/main/media/header.png","showCover":null,"date":"2025-01-11","category":"nodejs","url":"https://frontendmasters.com/blog/introducing-tanstack-start/","author":"Adam Rackis","translator":"","tags":["react"],"priority":2},"html":"
\n

TanStack Start目前处于beta阶段，提供服务端函数,支持SSR

"}},{"node":{"frontmatter":{"title":"zasper - 强大的数学科学IDE","publish":true,"cover":"https://raw.githubusercontent.com/zasper-io/assets/refs/heads/main//editor.png","showCover":null,"date":"2025-01-11","category":"python","url":"https://github.com/zasper-io/zasper","author":"Zasper","translator":"","tags":["ide"],"priority":1},"html":"
"}},{"node":{"frontmatter":{"title":"我一直在将我的 Google Sheets变成适合手机的网络应用程序","publish":true,"cover":"","showCover":null,"date":"2025-01-11","category":"database","url":"https://arstechnica.com/gadgets/2024/12/making-tiny-no-code-webapps-out-of-spreadsheets-is-a-weirdly-fulfilling-hobby/","author":"Ars Technica","translator":"","tags":["google sheet"],"priority":1},"html":"
\n

从技术上讲，Glide 是一种针对企业的无代码工具，但您可以免费获得一个基于用户的已发布应用程序，如果您真的将其留给您的家庭或朋友群，您可以拥有更多“私人”应用程序。每个成熟的应用程序可以有 10 个用户和最多 25000 行，这对于大多数用途来说可能已经足够了。

"}},{"node":{"frontmatter":{"title":"贝塞尔曲线的 CSS动画生成器","publish":true,"cover":"","showCover":null,"date":"2025-01-11","category":"frontend","url":"https://www.easing.dev/","author":"Alex Vanderzon, Lochie Axon","translator":"","tags":["css"],"priority":1},"html":"
"}},{"node":{"frontmatter":{"title":"我如何使用AI自动化我的工作申请流程","publish":true,"cover":"","showCover":null,"date":"2025-01-11","category":"AI","url":"https://blog.daviddodda.com/how-i-automated-my-job-application-process-part-1","author":"David Dodda","translator":"","tags":["LLM"],"priority":1},"html":"
\n

Look, I'll be honest - job hunting sucks.

"}},{"node":{"frontmatter":{"title":"css平衡文本换行","publish":true,"cover":"https://ishadeed.com/assets/text-wrap/figures/coffee-4.jpeg","showCover":null,"date":"2025-01-11","category":"frontend","url":"https://ishadeed.com/article/balancing-text-css/","author":"Ahmad Shadeed","translator":"","tags":["css"],"priority":2},"html":"
\n

使用text-wrap: balance 和 text-wrap: pretty

"}},{"node":{"frontmatter":{"title":"静态搜索树：比二叉搜索快 40 倍","publish":true,"cover":"https://curiouscoding.nl/ox-hugo/28-threads.svg","showCover":null,"date":"2025-01-11","category":"math","url":"https://curiouscoding.nl/posts/static-search-tree/","author":"Ragnar Groot Koerkamp","translator":"","tags":["algorithm"],"priority":2},"html":"
\n

总而言之，我们在 4GB 输入上进行二进制搜索的 1150ns/query 变为具有交错查询的优化 S-tree 的 27ns，速度提高了 40x以上！这种改进的很大一部分是由于批处理查询和预取即将到来的节点。为了获得更高的吞吐量，不同级别的交错查询有助于平衡计算的 CPU 绑定部分与内存绑定部分，从而获得更高的整体吞吐量。

"}},{"node":{"frontmatter":{"title":"我很富有，不知道该如何度过我的生活","publish":true,"cover":"","showCover":null,"date":"2025-01-11","category":"other","url":"https://vinay.sh/i-am-rich-and-have-no-idea-what-to-do-with-my-life/","author":"Vinay Hiremath","translator":"","tags":["life"],"priority":1},"html":"
\n

作者是loom联合创始人，loom被9亿美元收购。有了大把的时间和金钱后作者却感到孤独，女朋友都分手了，目前他在夏威夷

"}}]}}