{"pageContext":{"index":321,"total":364,"name":"2025-06-28","items":[{"node":{"frontmatter":{"title":"Phoenix.new – Phoenix 的远程 AI 运行时","publish":true,"cover":"","showCover":null,"date":"2025-06-28","category":"other","url":"https://fly.io/blog/phoenix-new-the-remote-ai-runtime/","author":"Chris McCord","translator":"","tags":["other"],"priority":1},"html":""}},{"node":{"frontmatter":{"title":"Harper – Grammarly 的开源离线替代品 - 语法检查器","publish":true,"cover":"https://github.com/Automattic/harper/raw/master/logo.svg","showCover":null,"date":"2025-06-28","category":"javascript","url":"https://github.com/automattic/harper?tab=readme-ov-file","author":"Automattic","translator":"","tags":["other"],"priority":1},"html":"<p>目前只支持english，但其机制支持扩展到其他语言</p>"}},{"node":{"frontmatter":{"title":"Unregistry – 无需注册将Docker 镜像推送到远程服务器","publish":true,"cover":"https://github.com/psviderski/unregistry/raw/main/.github/images/logo-light.svg#gh-light-mode-only","showCover":null,"date":"2025-06-28","category":"devops","url":"https://github.com/psviderski/unregistry","author":"psviderski","translator":"","tags":["other"],"priority":1},"html":""}},{"node":{"frontmatter":{"title":"Warka - react电子墨水屏ui框架","publish":true,"cover":"https://github.com/k3nz0/Warka/raw/main/pictures/warka2.jpg","showCover":null,"date":"2025-06-28","category":"frontend","url":"https://github.com/k3nz0/Warka","author":"Moez Ezzeddine","translator":"","tags":["react"],"priority":2},"html":"<p>实际图像由后端(python)绘制</p>"}},{"node":{"frontmatter":{"title":"Workout.cool – 开源健身指导平台","publish":true,"cover":"https://github.com/Snouzy/workout-cool/raw/main/public/logo.png","showCover":null,"date":"2025-06-28","category":"nodejs","url":"https://github.com/Snouzy/workout-cool","author":"Snouzy","translator":"","tags":["other"],"priority":1},"html":""}},{"node":{"frontmatter":{"title":"无人驾驶出租车Waymo 的乘车费用高于 Uber 或 Lyft，但人们仍然愿意付费","publish":true,"cover":"https://techcrunch.com/wp-content/uploads/2025/03/GettyImages-2206294608.jpg?resize=2048,1365","showCover":null,"date":"2025-06-28","category":"AI","url":"https://techcrunch.com/2025/06/12/waymo-rides-cost-more-than-uber-or-lyft-and-people-are-paying-anyway/","author":"Sean OKane","translator":"","tags":["other"],"priority":1},"html":"<p>Obi（一个聚合实时定价和乘车服务的应用程序）的一份报告显示，Waymo的价格平均为20.43美元，而Lyft为14.44美元，Uber为15.58美元。\n尽管价格更高，Waymo仍然很受欢迎，每周提供25万次付费行程。\n调查显示，用户对无人驾驶出租车充满热情，但安全仍然是主要担忧。许多人希望有某种形式的远程人工监控。尽管多数受访者希望价格相同或更低，但也有相当一部分人愿意为Waymo的服务支付更高的价格，这可能是因为它提供了一种独特的、舒适的乘车体验。</p>"}},{"node":{"frontmatter":{"title":"Web 爬虫程序和爬虫程序检测技术简史","publish":true,"cover":"https://sinja.io/images/bot-or-not/captchas.png","showCover":null,"date":"2025-06-28","category":"javascript","url":"https://sinja.io/blog/bot-or-not","author":"Oleg Wock","translator":"","tags":["other"],"priority":1},"html":"<p>其中一项，对于破解验证码的工作外包给真实的人</p>"}},{"node":{"frontmatter":{"title":"nodejs序列化各种方式比较","publish":true,"cover":"","showCover":null,"date":"2025-06-28","category":"nodejs","url":"https://adamfaulkner.github.io/serialization_from_nodejs.html","author":"Adam Faulkner","translator":"","tags":["serialization"],"priority":1},"html":"<p>JSON和其他Protobuf、Msgpackr(纯js)、Avro，</p>"}},{"node":{"frontmatter":{"title":"soundz - react sound组件","publish":true,"cover":"https://raw.githubusercontent.com/kazewaze/assets-holder/main/soundz.svg","showCover":null,"date":"2025-06-28","category":"frontend","url":"https://github.com/kazewaze/soundz","author":"Kaycee Ingram","translator":"","tags":["react","sound"],"priority":1},"html":""}},{"node":{"frontmatter":{"title":"为什么 SSL 在 90 年代末更名为 TLS (2014)","publish":true,"cover":"","showCover":null,"date":"2025-06-28","category":"other","url":"https://tim.dierks.org/2014/05/security-standards-and-name-changes-in.html","author":"Tim Dierks","translator":"","tags":["other"],"priority":1},"html":"<p>在Netscape和微软的浏览器大战期间，Netscape开发了SSL协议，但最初版本存在缺陷。第一个生产版本是SSL 2，但它也有一些问题。微软为了参与竞争，基于SSL 2开发了PCT协议，仅在IE和IIS中支持。Netscape也想解决SSL 2的问题，但不想让微软主导标准，因此开发了SSL 3.0。为了避免协议分叉，作者所在的公司（Consensus Development）促成了Netscape和微软的会面，最终达成协议，由IETF接管协议并进行标准化。作为妥协，SSL 3.0需要进行一些修改（避免看起来像IETF直接认可Netscape的协议），并且需要更名。因此，TLS 1.0诞生了，它实际上是SSL 3.1。</p>"}},{"node":{"frontmatter":{"title":"三星在西亚和北非地区的手机上嵌入IronSource间谍软件应用程序","publish":true,"cover":"https://smex.org/wp-content/uploads/2025/05/SAMSUNG_AFP-1200x686.jpg","showCover":null,"date":"2025-06-28","category":"mobile","url":"https://smex.org/open-letter-to-samsung-end-forced-israeli-app-installations-in-the-wana-region/","author":"SMEX","translator":"","tags":["samsung"],"priority":1},"html":"<p>作者SMEX (黎巴嫩NGO) 指出在西亚和北非 (WANA) 地区销售的三星A和M系列智能手机预装了一款名为AppCloud的流氓软件，用户对此表示担忧。该软件由以色列公司ironSource（现已被美国Unity公司收购）开发，会在用户不知情或未经同意的情况下收集敏感个人数据，且无法在不影响设备安全性的情况下移除。</p>"}},{"node":{"frontmatter":{"title":"分享 - 日常开发使用Claude Code的经验","publish":true,"cover":"","showCover":null,"date":"2025-06-28","category":"AI","url":"https://kadekillary.work/blog/#2025-06-16-snorting-the-agi-with-claude-code","author":"Kade Killary","translator":"","tags":["claude"],"priority":1},"html":""}},{"node":{"frontmatter":{"title":"miniDiffusion - 纯 PyTorch 从头开始重新实现了 Stable Diffusion 3.5","publish":true,"cover":"https://github.com/yousef-rafat/miniDiffusion/raw/main/assets/display.png","showCover":null,"date":"2025-06-28","category":"AI","url":"https://github.com/yousef-rafat/miniDiffusion","author":"yousef-rafat","translator":"","tags":["Diffusion"],"priority":1},"html":"<p>该网页内容介绍了一个名为 miniDiffusion 的项目，它是一个使用纯 PyTorch 从头开始重新实现的 Stable Diffusion 3.5 模型。</p>\n<p><strong>核心内容：</strong></p>\n<ul>\n<li><strong>目的：</strong> 该项目旨在用于教育、实验和破解用途，目标是以最少的代码重现 Stable Diffusion 3.5。</li>\n<li>\n<p><strong>代码结构：</strong></p>\n<ul>\n<li><code class=\"language-text\">dit.py</code>, <code class=\"language-text\">dit_components.py</code>, <code class=\"language-text\">attention.py</code>：包含主要的 Stable Diffusion 模型代码。<code class=\"language-text\">dit.py</code>是主模型，<code class=\"language-text\">dit_components.py</code>包含 embedding, normalization, patch embedding 等辅助函数，<code class=\"language-text\">attention.py</code>包含 Joint Attention 的实现。</li>\n<li><code class=\"language-text\">noise.py</code>：包含用于求解 Rectified Flow ODE 的 Euler 调度器。</li>\n<li><code class=\"language-text\">t5_encoder.py</code>, <code class=\"language-text\">clip.py</code>, <code class=\"language-text\">tokenizer.py</code>：包含文本编码器（T5 和 CLIP）及其分词器。</li>\n<li><code class=\"language-text\">metrics.py</code>：实现了 Fréchet inception distance (FID) 评估指标。</li>\n<li><code class=\"language-text\">common.py</code>：包含训练的辅助函数。</li>\n<li><code class=\"language-text\">common_ds.py</code>：实现了将图像数据转换为 DiT 模型可训练数据的可迭代数据集。</li>\n<li><code class=\"language-text\">encoders</code> 文件夹：保存 VAE, CLIP 等模块的 checkpoint。</li>\n<li><code class=\"language-text\">model</code> 文件夹：训练后保存模型 checkpoint 和 logs。</li>\n</ul>\n</li>\n<li>\n<p><strong>组件：</strong></p>\n<ul>\n<li>VAE, CLIP 和 T5 文本编码器的实现。</li>\n<li>Byte-Pair &#x26; Unigram 分词器。</li>\n<li>SD3 组件，包括 Multi-Modal Diffusion Transformer Model，Flow-Matching Euler Scheduler，Logit-Normal Sampling，Joint Attention。</li>\n</ul>\n</li>\n<li>\n<p><strong>使用方法：</strong></p>\n<ul>\n<li>克隆仓库：<code class=\"language-text\">git clone &quot;https://github.com/yousef-rafat/miniDiffusion&quot;</code></li>\n<li>安装依赖：<code class=\"language-text\">pip install -r requirements.txt</code></li>\n<li>安装模型 Checkpoint：在 <code class=\"language-text\">get_checkpoints.py</code> 中添加 Hugging Face Token，然后运行 <code class=\"language-text\">python3 encoders/get_checkpoints.py</code></li>\n</ul>\n</li>\n<li><strong>许可协议：</strong> 该项目使用 MIT 许可证。</li>\n<li><strong>警告：</strong> 该项目仍处于实验阶段，需要更多测试。</li>\n</ul>"}},{"node":{"frontmatter":{"title":"本田成功发射和着陆实验性可重复使用火箭","publish":true,"cover":"https://global.honda/content/dam/site/global-en/topics-new/cq_img/2025/c_2025-06-17ceng/c_2025-06-17ceng_02.jpg","showCover":null,"date":"2025-06-28","category":"other","url":"https://global.honda/en/topics/2025/c_2025-06-17ceng.html","author":"honda","translator":"","tags":["other"],"priority":1},"html":"<p>2025年6月17日，本田研发有限公司成功进行了实验性可重复使用火箭的发射和着陆测试。该火箭长6.3米，直径85厘米，干重900公斤，湿重1312公斤，由本田自主研发。这是本田首次在火箭达到近300米的高度后成功着陆。</p>\n<p>本次测试旨在验证火箭重复使用所需的关键技术，例如上升和下降过程中的飞行稳定性以及着陆能力。测试中，火箭达到了271.4米的高度，着陆点距离目标地点37厘米，飞行时长为56.6秒，并收集了上升和下降过程中的数据。</p>\n<p><a href=\"https://global.honda/content/dam/site/global-en/topics-new/cq_img/2025/c_2025-06-17ceng/print/c_2025-06-17c_101H.mp4\">本田官方视频</a></p>"}},{"node":{"frontmatter":{"title":"Anthropic公司在构建LLM Agent方面的经验","publish":true,"cover":"https://www.anthropic.com/_next/image?url=https%3A%2F%2Fwww-cdn.anthropic.com%2Fimages%2F4zrzovbb%2Fwebsite%2F4b9a1f4eb63d5962a6e1746ac26bbc857cf3474f-2400x1666.png&w=3840&q=75","showCover":null,"date":"2025-06-28","category":"AI","url":"https://www.anthropic.com/engineering/building-effective-agents","author":"Erik Schluntz and Barry Zhang","translator":"","tags":["other"],"priority":1},"html":"<p>强调了简单、可组合的模式的重要性，并分享了构建高效Agent的实用建议。</p>\n<ul>\n<li>\n<p><strong>Agent的定义：</strong> Agent系统分为两种：</p>\n<ul>\n<li><strong>Workflow：</strong> LLM和工具通过预定义的代码路径进行编排。</li>\n<li><strong>Agent：</strong> LLM动态地指导其自身流程和工具使用，自主控制任务完成方式。</li>\n</ul>\n</li>\n<li><strong>何时使用Agent：</strong> 优先选择最简单的解决方案，仅在必要时增加复杂性。Workflow适用于定义明确的任务，而Agent更适合需要灵活性和模型驱动决策的场景。</li>\n<li><strong>关于框架：</strong> 框架可以简化初始开发，但也可能增加抽象层，使得调试更加困难。建议先直接使用LLM API，并在使用框架时理解底层代码。</li>\n</ul>\n<p>构建成功的LLM Agent的关键在于选择合适的系统，从简单的Prompt开始，进行全面的评估和优化，并在必要时才添加多步骤的Agent系统。应坚持设计简洁、过程透明、精心设计Agent与计算机的接口等原则。</p>"}},{"node":{"frontmatter":{"title":"将一台旧的iPhone 8改造成太阳能供电的视觉OCR服务器","publish":true,"cover":"https://terminalbytes.com/iphone-8-solar-powered-vision-ocr-server/iphone-8-vision-ocr-server-solar.jpg","showCover":null,"date":"2025-06-28","category":"mobile","url":"https://terminalbytes.com/iphone-8-solar-powered-vision-ocr-server/","author":"Hemant Kumar","translator":"","tags":["iphone"],"priority":1},"html":"<ul>\n<li>使用SwiftUI编写iOS应用，利用Apple Vision框架进行本地OCR处理。</li>\n<li>EcoFlow River 2 Pro提供电力，通过太阳能充电。</li>\n<li>\n<p>Mini PC负责web服务和API路由，并通过Tailscale网络连接所有设备。`</p>\n<blockquote>\n<p>评论区神吐槽：节约的费用不够交给apple的开发者年费</p>\n</blockquote>\n</li>\n</ul>"}}]}}