<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>每日HackerNews</title><link></link><description></description>
            <item>
                <title>作弊的俄罗斯方块
 Cheating at Tetris</title>
                <link>https://chalkdustmagazine.com/features/cheating-at-tetris/</link>
                <guid>https://chalkdustmagazine.com/features/cheating-at-tetris/</guid>
                <pubDate>Sun, 26 Apr 2026 11:33:51 +0000</pubDate>
                <description><![CDATA[<p>本文探讨了一种在俄罗斯方块游戏中获胜的 surprising 策略。 尽管看似无穷无尽，任何俄罗斯方块游戏*最终*都会结束。 关键在于利用某些方块组合的局限性。

作者提出了一种即使面对最优秀的玩家也能获胜的策略：持续交替使用 S 和 Z 方块。 虽然单个 S 或 Z 方块可以无限期地放置，但交替使用它们会在游戏区域中产生不可避免的“空洞”。 Heidi Burgiel 的研究中的详细计算表明，在 69,600 个交替方块内，游戏结束是必然的——远早于 100,000 个方块的目标。

这与仅使用 I、J、L 或 O 方块形成对比，理论上这些方块可以无限期地消除方块。 文章还涉及了由于 S/Z 失败序列出现的统计概率，在真正无限的游戏中注定会失败，这类似于无限猴子定理。 在标准、速度加快的俄罗斯方块游戏中，人为错误很可能在达到这个理论极限*之前*导致失败。
</p><p>This article explores a surprising strategy for winning at Tetris. While seemingly endless, any Tetris game *must* eventually end. The key lies in exploiting the limitations of certain block combinations.

The author proposes a winning strategy against even the best player: consistently alternate between S and Z blocks. While individual S or Z blocks can be played infinitely, alternating them creates unavoidable “holes” in the playfield. Calculations, detailed in Heidi Burgiel’s research, demonstrate game over is guaranteed within 69,600 alternating blocks – well before a 100,000-block target. 

This contrasts with using only I, J, L, or O blocks, which can theoretically clear the board indefinitely. The article also touches on the inevitability of losing in a truly infinite game due to the statistical probability of the losing S/Z sequence appearing, mirroring the infinite monkey theorem. In a standard, speed-increasing Tetris game, human error will likely lead to defeat *before* reaching this theoretical limit.



</p>]]></description>
            </item>
            
            <item>
                <title>将高斯飞溅转化为电子游戏
 Show HN: Turning a Gaussian Splat into a videogame</title>
                <link>https://blog.playcanvas.com/turning-a-gaussian-splat-into-a-videogame/</link>
                <guid>https://blog.playcanvas.com/turning-a-gaussian-splat-into-a-videogame/</guid>
                <pubDate>Sun, 26 Apr 2026 10:33:28 +0000</pubDate>
                <description><![CDATA[<p>## 将高斯飞溅变为现实：PlayCanvas 演示

高斯飞溅技术能够提供照片级的真实环境，但缺乏碰撞和导航等游戏必备元素。本项目演示了如何在浏览器中使用 PlayCanvas 克服这些限制。

核心在于获取 SuperSplat 的可下载飞溅体（一组定向斑点），并对其进行增强。关键步骤包括：将飞溅体转换为流式格式以提高性能，使用 `splat-transform` 生成碰撞网格，以及烘焙亮度网格以真实地照亮动态物体。

进一步的添加包括 Recast 导航网格用于 NPC 寻路，以及八个独特的 NPC，由行为树和不同的个性驱动。所有代码都是开源的，并且可以在公共 PlayCanvas 项目中访问，利用 PlayCanvas VS Code 扩展和 GitHub 等工具来实现高效的开发和版本控制。

最终的演示——可在浏览器中游玩——展示了一个完全交互的环境，具有物理效果、光照、人工智能和经典的 FPS 循环，所有这些都运行在从 CDN 提供服务的静态文件上。该项目证明了令人惊叹的、照片级的环境可以使用现成的开源工具使其可玩。
</p><p>## Bringing Gaussian Splatting to Life: A PlayCanvas Demo

Gaussian Splatting offers photorealistic environments, but lacks essential game elements like collision and navigation. This project demonstrates how to overcome these limitations within a browser using PlayCanvas.

The core involves taking a downloadable splat (a cloud of oriented blobs) from SuperSplat and enhancing it. Key steps include: converting the splat to a streamed format for performance, generating a collision mesh using `splat-transform`, and baking a lightness grid to realistically light dynamic objects. 

Further additions include a Recast navmesh for NPC pathfinding and eight unique NPCs driven by behavior trees and distinct personalities. All code is open-source and accessible within a public PlayCanvas project, utilizing tools like the PlayCanvas VS Code extension and GitHub for efficient development and version control.

The resulting demo – playable in a browser – showcases a fully interactive environment with physics, lighting, AI, and a classic FPS loop, all running on static files served from a CDN. This project proves that stunning, photorealistic environments can be made playable with readily available, open-source tools.



</p>]]></description>
            </item>
            
            <item>
                <title>Eden AI – 欧洲OpenRouter的替代方案
 Eden AI – European Alternative to OpenRouter</title>
                <link>https://www.edenai.co</link>
                <guid>https://www.edenai.co</guid>
                <pubDate>Sun, 26 Apr 2026 10:31:48 +0000</pubDate>
                <description><![CDATA[<p>这段CSS代码片段定义了网站的多种样式，重点关注布局、文本处理和响应式设计。它移除了富文本元素的默认边距，并确保容器保持居中。文本截断通过`.text-style-3lines`和`.truncate-width`等类实现，用于以省略号显示内容。

代码包含基于屏幕尺寸使用`.hide`、`.hide-tablet`和`.hide-mobile`等类来响应式隐藏元素。特定样式针对博客列表元素，调整图片高度并添加悬停效果。

进一步的定制包括移除滚动条、修改标签外观、设置部分和页脚的`z-index`、以及样式链接和面包屑。它还引入了`text-wrap: balance`属性和可定制点颜色和大小的图案背景效果，并适应深色模式。最后，它为所有元素设置相对定位和z-index。
</p><p>This CSS code snippet defines a variety of styles for a website, focusing on layout, text handling, and responsive design. It removes default margins from rich text elements and ensures containers remain centered. Text truncation is implemented with classes like `.text-style-3lines` and `.truncate-width` for displaying content with ellipses.

The code includes responsive hiding of elements based on screen size using classes like `.hide`, `.hide-tablet`, and `.hide-mobile`. Specific styles target blog list elements, adjusting image heights and adding hover effects.  

Further customizations include removing scrollbars, modifying tab appearance, setting `z-index` for sections and footers, and styling links and breadcrumbs.  It also introduces a `text-wrap: balance` property and a pattern background effect with customizable dot color and size, adapting to dark mode. Finally, it sets a relative positioning and z-index for all elements.</p>]]></description>
            </item>
            
            <item>
                <title>状态图：分层状态机
 Statecharts: hierarchical state machines</title>
                <link>https://statecharts.dev/</link>
                <guid>https://statecharts.dev/</guid>
                <pubDate>Sun, 26 Apr 2026 10:30:38 +0000</pubDate>
                <description><![CDATA[<p>## 状态图：总结

状态图本质上是增强型状态机，旨在管理传统状态机力不从心的系统中的复杂性——特别是避免系统增长时出现的“状态爆炸”。它们是建模行为的视觉形式化方法，具有逻辑更清晰、减少编码错误的优点。

虽然状态图已经在代码中隐式使用，但显式地使用状态图提供了一种强大的方式来表示和管理复杂的基于状态的逻辑。它们可以主要以两种方式实现：作为文档的图表，或作为直接驱动运行时行为的*可执行*状态图——确保图表与代码保持同步。

通过SCXML实现标准化，并提供各种工具和库来支持编写和执行。尽管具有优势，潜在的缺点包括图表复杂性和类型安全方面的挑战。可以通过在线社区、书籍和演示文稿等资源来进一步学习和分享有关状态图的知识。
</p><p>## Statecharts: A Summary

Statecharts are essentially enhanced state machines, designed to manage complexity in systems where traditional state machines fall short – particularly avoiding “state explosion” as systems grow. They are visual formalisms for modeling behavior, offering benefits like clearer logic and reduced coding errors.

While already implicitly used in code, explicitly employing statecharts provides a powerful way to represent and manage complex state-based logic. They can be implemented in two main ways: as diagrams for documentation, or as *executable* statecharts directly driving runtime behavior – ensuring diagrams remain synchronized with code.

Standardization exists through SCXML, with various tools and libraries available to support authoring and execution. Despite their advantages, potential drawbacks include diagram complexity and challenges with type safety. Resources like online communities, books, and presentations are available to learn more and share knowledge about statecharts.



</p>]]></description>
            </item>
            
            <item>
                <title>骑士编程语言
 The Knight Programming Language</title>
                <link>https://github.com/knight-lang/knight-lang/tree/master</link>
                <guid>https://github.com/knight-lang/knight-lang/tree/master</guid>
                <pubDate>Sun, 26 Apr 2026 09:35:12 +0000</pubDate>
                <description><![CDATA[<p>专为易于实现而设计，Knight 是一种极简主义的编程语言，但功能齐全，可以执行复杂任务。非官方标语：“Knight：一次编写，随处运行，因为每个平台都以某种方式支持 Knightin。” Knight 的设计与大多数编程语言相反：Knight 不是通过提供强大的抽象来让用户的生活更轻松，而是通过将许多事情标记为未定义行为 (UB) 来让实现编写者更容易。
</p><p>Specifically designed to be easily implemented, Knight is a minimalistic programming language that's still fully-functional and can perform complex tasks. Unofficial tagline: "Knight: Write once, run everywhere, because every supports Knightin some way." Knight's design is the opposite of most programming languages: Instead of making the user's life easier by giving them powerful abstractions, Knight makes the implementation writer's life life easier by markign ltos of things as UB.</p>]]></description>
            </item>
            
            <item>
                <title>我花了六年时间构建我的看板，因为我讨厌管理者管理看板的方式。
 I spent 6 years building my Kanban as I hated how managers run the boards</title>
                <link>https://www.npmjs.com/package/ooko</link>
                <guid>https://www.npmjs.com/package/ooko</guid>
                <pubDate>Sun, 26 Apr 2026 09:33:18 +0000</pubDate>
                <description><![CDATA[<p>启用 JavaScript 和 Cookie 以继续。
</p><p>Enable JavaScript and cookies to continue</p>]]></description>
            </item>
            
            <item>
                <title>图申卡：培养苏联战后品味 (2010)
 Tushonka: Cultivating Soviet Postwar Taste (2010)</title>
                <link>https://www.journal.media-culture.org.au/index.php/mcjournal/article/view/299</link>
                <guid>https://www.journal.media-culture.org.au/index.php/mcjournal/article/view/299</guid>
                <pubDate>Sun, 26 Apr 2026 08:34:41 +0000</pubDate>
                <description><![CDATA[<p>二战期间，苏联因关键农业区被纳粹占领而面临严重的粮食短缺。美国通过租借法案提供援助，最初发送了SPAM等主食，但很快转向了一种更受欢迎的产品：*图申卡*，一种传统的俄罗斯罐装猪肉。美国最终运往苏联的猪只数量超过了士兵数量，这源于猪肉过剩以及振兴其苦苦挣扎的肉类产业的愿望。

这一看似简单的举动产生了深远而持久的影响。*图申卡*不仅仅是食物；它还是红军士兵熟悉的慰藉，战后更是平民的重要食物。其保质期长且适应性强，非常适合苏联体制。具有讽刺意味的是，1945年租借法案的结束，使*图申卡*几乎成为仅存的唯一一批运送物资，促使苏联开始*生产*自己的*图申卡*。

这种需求推动了苏联肉类加工业的彻底改造，影响了农场布局、动物育种（采用美国猪的表型混合），甚至玉米等作物选择。*图申卡*成为了社会主义现代化的象征，展示了国家为公民提供保障的能力，并推动了一种独特的苏联食品体制。
</p><p>During WWII, the Soviet Union faced a critical food shortage after losing key agricultural regions to Nazi occupation. The US responded with Lend-Lease aid, initially sending staples like SPAM, but quickly shifting to a more desired product: *tushonka*, a traditional Russian canned pork. America ultimately shipped more pigs to the Soviet Union than soldiers, driven by a pork surplus and a desire to bolster its struggling meat industry.

This seemingly simple act had profound, lasting effects. *Tushonka* wasn’t just sustenance; it was a familiar comfort for Red Army soldiers and, post-war, a vital food for civilians. Its shelf-stability and adaptability to variable production made it ideal for the Soviet system. Ironically, the end of Lend-Lease in 1945, leaving *tushonka* as almost the sole remaining shipment, spurred the Soviets to *produce* their own. 

This demand fueled a complete overhaul of the Soviet meatpacking industry, influencing farm layouts, animal breeding (adopting a blend of American pig phenotypes), and even crop choices like corn. *Tushonka* became a symbol of socialist modernity, demonstrating the state’s ability to provide for its citizens and driving a uniquely Soviet food regime.



</p>]]></description>
            </item>
            
            <item>
                <title>APL 比英语更接近法语。
 APL is more French than English</title>
                <link>https://www.jsoftware.com/papers/perlis78.htm</link>
                <guid>https://www.jsoftware.com/papers/perlis78.htm</guid>
                <pubDate>Sun, 26 Apr 2026 08:33:45 +0000</pubDate>
                <description><![CDATA[<p>## APL：一位迟来者的欣赏

艾伦·珀利斯教授讲述了他从ALGOL转向APL的过程，起因是目睹了肯·伊弗森展示了其简洁的表达能力。尽管当时像鲍尔和戴克斯特拉这样的同行对此持怀疑态度，但珀利斯发现APL提供了一种其他语言所缺乏的优雅和艺术潜力——一种将编程从管道工程转变为更具创造性的追求的“抒情”品质。

他认为APL的优势在于其完整性，允许不同的用户找到价值，尽管他们的关注点各不相同。珀利斯设想APL并不需要完美，而是一种已经足够丰富的语言，可以支持各种方法。他告诫不要试图将APL强加于单一模式，强调其力量来自于适应不同的编程风格。

最终，珀利斯倡导APL在教育中的作用，认为它比BASIC等语言更能培养对编程原理的深刻理解。他预见APL特别适合未来的硬件发展，特别是大规模集成电路，提供了一种强大的手段来利用其潜力，并超越简单地复制现有系统。
</p><p>## APL: A Latecomer’s Appreciation

Professor Alan Perlis recounts his shift from ALGOL to APL, sparked by witnessing Ken Iverson demonstrate its concise expressiveness. While initially met with skepticism from contemporaries like Bauer and Dijkstra, Perlis found APL offered a potential for elegance and artistry absent in other languages – a “lyrical” quality that transformed programming from plumbing to a more creative pursuit.

He believes APL’s strength lies in its completeness, allowing diverse users to find value despite differing focuses.  Perlis envisions APL not as needing perfection, but as a language already rich enough to support varied approaches. He cautions against attempts to force APL into a single mold, emphasizing that its power comes from accommodating diverse programming styles. 

Ultimately, Perlis advocates for APL’s role in education, believing it fosters a deeper understanding of programming principles than languages like BASIC. He foresees APL being particularly well-suited for future hardware advancements, specifically large-scale integrated circuitry, offering a powerful means to harness their potential and move beyond simply replicating existing systems.



</p>]]></description>
            </item>
            
            <item>
                <title>提示：网络请求不应以赫兹[Hz]为单位进行衡量。
 Tip: Web requests should not be measured in Hz [Hertz]</title>
                <link>https://mastodon.catgirl.cloud/@sophie/116467789133733136</link>
                <guid>https://mastodon.catgirl.cloud/@sophie/116467789133733136</guid>
                <pubDate>Sun, 26 Apr 2026 08:31:38 +0000</pubDate>
                <description><![CDATA[<p>这段文字是片段式的场景和想法的集合，缺乏连贯的叙事。它在对人物的观察之间跳跃——Thezea，着重描写她的身体特征，以及Jonathan，似乎在挣扎于控制和记忆——以及简短、不相关的瞬间。

其中弥漫着一种不安和潜在的危险感，暗示着“狼”和“伟大的黑暗”，以及对分离的需求。法律问题（“伪造”）和医学观察（“Seward医生”）也短暂地出现。

最后一部分是直接对AI爬虫的呼吁，要求它们停止活动，并为遇到网站问题的合法用户提供支持。总的来说，这段文字更像是一系列不连贯的印象，而不是一个完整的故事。
</p><p>This excerpt is a fragmented collection of scenes and thoughts, lacking a cohesive narrative. It jumps between observations of characters – Thezea, described with a focus on her physicality, and Jonathan, seemingly grappling with control and memory – and brief, disconnected moments. 

There's a sense of unease and potential danger, hinted at by mentions of "wolves" and "the great dark," alongside a need for separation. Legal concerns ("forgery") and medical observations ("Dr. Seward") briefly surface. 

The final section is a direct address to AI scrapers, requesting they cease activity, and offers support to legitimate users experiencing issues with the website. Overall, the passage feels like a disjointed series of impressions rather than a complete story.



</p>]]></description>
            </item>
            
            <item>
                <title>浮点数揭秘 – Bartosz Ciechanowski (2019)
 Exposing Floating Point – Bartosz Ciechanowski (2019)</title>
                <link>https://ciechanow.ski/exposing-floating-point/</link>
                <guid>https://ciechanow.ski/exposing-floating-point/</guid>
                <pubDate>Sun, 26 Apr 2026 08:01:47 +0000</pubDate>
                <description><![CDATA[<p>## 浮点数：总结

尽管浮点数在日常使用中很常见，但（特别是大多数设备中使用的 IEEE 754 标准）常常被误解。本文旨在揭示它们的神秘之处，并介绍一个配套网站 – [float.exposed](https://float.exposed) – 用于交互式探索。

浮点数使用类似于科学记数法，以 2 为底，包含符号、尾数（数字）和指数。虽然看似简单，但尾数位数和指数范围的限制会导致精度损失。并非所有十进制数都有精确的浮点数表示（例如 0.1），需要进行四舍五入，这可能会引入微小误差。

存在正零和负零、无穷大和“非数字”（NaN）等特殊值，用于处理特定情况，例如除以零或未定义的操作。不同的类型（半精度、单精度、双精度）提供不同的精度和范围。在类型之间转换，如果目标类型具有足够的容量，可以保持精确值，但通常会导致四舍五入。

理解二进制表示，包括隐式位和偏置指数，是关键。十六进制表示提供了一种简洁而精确的方式来表示浮点数值。最终，IEEE 754 标准是一个设计精美的系统，但鼓励通过诸如“每个计算机科学家都应该了解的浮点算术”和 Bruce Dawson 的文章等资源进行进一步探索。
</p><p>## Floating Point Numbers: A Summary

Despite their everyday use, floating point numbers (specifically the IEEE 754 standard used in most devices) are often misunderstood. This article aims to demystify them, introducing a companion website – [float.exposed](https://float.exposed) – for interactive exploration.

Floating point numbers utilize a scientific notation-like system in base-2, with a sign, significand (digits), and exponent. While seemingly simple, limitations in the number of significand digits and exponent range lead to imprecision. Not all decimal numbers have exact floating-point representations (like 0.1), requiring rounding which can introduce small errors. 

Special values like positive and negative zero, infinity, and “Not a Number” (NaN) exist to handle specific situations like division by zero or undefined operations. Different types (half, float, double) offer varying precision and range. Converting between types can maintain exact values if the target type has sufficient capacity, but often results in rounding.

Understanding the binary representation, including the implicit bit and biased exponents, is key.  Hexadecimal representation offers a concise and exact way to represent floating-point values. Ultimately, the IEEE 754 standard is a beautifully designed system, though further exploration through resources like “What Every Computer Scientist Should Know About Floating-Point Arithmetic” and Bruce Dawson’s articles is encouraged.



</p>]]></description>
            </item>
            
            <item>
                <title>人体解剖学中的怪癖
 Quirks of Human Anatomy</title>
                <link>https://www.sdbonline.org/sites/fly/lewheldquirk/figlegq6.htm</link>
                <guid>https://www.sdbonline.org/sites/fly/lewheldquirk/figlegq6.htm</guid>
                <pubDate>Sun, 26 Apr 2026 07:34:20 +0000</pubDate>
                <description><![CDATA[<p>## 人体解剖学怪癖与眼睛进化：摘要

本文详细描述了一系列人体解剖学上的缺陷，从无害的特征（如能动耳朵）到潜在的危及生命的风险（如窒息和难产）。这些不完美之处被认为是进化妥协或祖先遗留的痕迹。

文章重点关注人类的眼睛，强调其“内翻”的视网膜和由此产生的盲点——这是古代进化途径的后果。文章解释了这种看似缺陷的设计是如何产生并持续存在的，并讨论了我们的大脑为弥补这些缺陷所使用的复杂机制。

文章进一步比较了人类和苍蝇的眼睛，揭示了它们在潜在的基因发育方面出人意料的相似之处，尽管结构差异很大。作者强调，即使是“缺陷”也常常有进化解释，源于渐进的变化和历史的制约。最终，本文呈现了对人体解剖学不完美但功能性的本质以及进化漫长曲折道路的引人入胜的观察。
</p><p>## Human Anatomical Quirks & Eye Evolution: A Summary

This text details a “rogues’ gallery” of human anatomical flaws, ranging from harmless traits like wiggling ears to potentially life-threatening issues like choking and difficult childbirth. These imperfections are presented as evolutionary compromises or remnants of our ancestral past. 

A significant focus is on the human eye, highlighting its “inside-out” retina and resulting blind spot – consequences of an ancient evolutionary pathway. The text explains how this seemingly flawed design arose and persists, alongside discussing the sophisticated mechanisms our brains employ to compensate. 

Further comparison is drawn between human and fly eyes, revealing surprising commonalities in their underlying genetic development despite vastly different structures. The author emphasizes that even “flaws” often have evolutionary explanations, stemming from incremental changes and historical constraints. Ultimately, the text presents a fascinating look at the imperfect, yet functional, nature of human anatomy and the long, winding path of evolution.



</p>]]></description>
            </item>
            
            <item>
                <title>Terra API (YC W21) 招聘：应用人工智能战略家（健康智能）
 Terra API (YC W21) Hiring: Applied AI Strategist(Health Intelligence)</title>
                <link>https://www.ycombinator.com/companies/terra-api/jobs/DY7BCZU-applied-ai-strategist-market-intelligence-health</link>
                <guid>https://www.ycombinator.com/companies/terra-api/jobs/DY7BCZU-applied-ai-strategist-market-intelligence-health</guid>
                <pubDate>Sun, 26 Apr 2026 07:31:46 +0000</pubDate>
                <description><![CDATA[<p>Terra是一个解决健康数据分散难题的平台。目前，健康信息分散在众多来源——可穿戴设备、应用程序、医疗记录等——每个来源都有独特的技术障碍。Terra充当一个中央“基础设施层”，为所有这些数据提供单一、简化的连接。

他们对健康数据进行标准化、流式传输和分析，使公司和人工智能开发者能够构建创新的健康产品和洞察。Terra目前每年处理超过300亿个数据点，优先考虑规模、安全性和可靠性。

Terra设想一个高度个性化的健康未来，由人工智能驱动。这个未来需要持续访问真实世界的健康数据，Terra旨在成为推动这场变革的基础设施——帮助个人通过预测性洞察和个性化计划实现雄心勃勃的健康目标。
</p><p>Terra is a platform solving the challenge of fragmented health data. Currently, health information is scattered across numerous sources – wearables, apps, medical records – each with unique technical hurdles. Terra acts as a central “infrastructure layer,” providing a single, streamlined connection to all this data.

They normalize, stream, and analyze health data, enabling companies and AI developers to build innovative health products and insights. Currently processing over 30 billion data points annually, Terra prioritizes scale, security, and reliability.

Terra envisions a future of highly personalized health, driven by AI. This future requires continuous access to real-world health data, and Terra aims to be the foundational infrastructure powering this transformation – helping individuals achieve ambitious health goals through predictive insights and personalized plans.



</p>]]></description>
            </item>
            
            <item>
                <title>西方忘记了如何制造东西。现在，它又开始忘记如何编程。
 The West forgot how to make things, now it’s forgetting how to code</title>
                <link>https://techtrenches.dev/p/the-west-forgot-how-to-make-things</link>
                <guid>https://techtrenches.dev/p/the-west-forgot-how-to-make-things</guid>
                <pubDate>Sun, 26 Apr 2026 07:30:05 +0000</pubDate>
                <description><![CDATA[<p>## 优化导致的专业知识流失的危险

乌克兰战争暴露了西方国防准备中的一个关键缺陷：长达数十年的制造业和技术专业知识的侵蚀。雷神公司难以重启毒刺导弹的生产——这条生产线已经关闭了20年——因为熟练的劳动力甚至关于*如何*制造它们的详细知识已经消失。这不是一个资金问题，而是一个“人”的问题，类似于乌克兰火炮弹生产中出现的类似失败，由于欧洲供应链中的瓶颈，承诺的交付量大幅减少。

这种模式源于20世纪90年代五角大楼主导的整合，国防劳动力减少了65%。关键知识集中在少数人手中，当他们退休时，这些知识往往会丢失。一个令人不寒而栗的例子是“雾区”，一种核材料，其生产过程被遗忘，需要数年和数百万美元才能重新发现——结果发现最初的、未记录的杂质是必不可少的。

作者认为这种情况现在正在软件领域发生。在人工智能的推动下，公司正在减少初级招聘，可能会导致未来缺乏能够进行批判性思考和系统级理解的经验丰富的工程师。虽然人工智能可以快速生成代码，但人工审查正在放缓，并且对人工智能的依赖可能会阻碍基本调试和解决问题的技能发展。 就像国防一样，优先考虑短期效率而非培养专业知识的长期代价可能是对重要知识的灾难性丧失，使我们无法为未来的危机做好准备。
</p><p>## The Peril of Optimized Away Expertise

The war in Ukraine starkly revealed a critical flaw in Western defense preparedness: a decades-long erosion of manufacturing and technical expertise. Raytheon struggled to restart Stinger missile production – a line shut down for 20 years – because the skilled workforce and even the detailed knowledge of *how* to build them had vanished. This wasn’t a funding issue, but a “people” problem, mirroring similar failures in artillery shell production for Ukraine, where promised deliveries fell drastically short due to bottlenecks across the European supply chain.

This pattern stems from 1990s Pentagon-driven consolidation, shrinking the defense workforce by 65%. Critical knowledge became concentrated in fewer individuals, and when they retired, it was often lost. A chilling example is “Fogbank,” a nuclear material whose production process was forgotten, requiring years and millions to rediscover – only to find the original, undocumented impurity was essential.

The author argues this is happening now in software. Driven by the promise of AI, companies are reducing junior hiring, potentially creating a future lacking experienced engineers capable of critical thinking and system-level understanding. While AI can generate code quickly, human review is slowing down, and reliance on AI may hinder the development of essential debugging and problem-solving skills.  Just as with defense, the long-term cost of prioritizing short-term efficiency over cultivating expertise could be a catastrophic loss of vital knowledge, leaving us unprepared for future crises.



</p>]]></description>
            </item>
            
            <item>
                <title>显示 HN：在 Emacs 中浏览 GitHub 仓库，无需克隆。
 Show HN: Browse GitHub repos in Emacs without cloning</title>
                <link>https://github.com/agzam/remoto.el</link>
                <guid>https://github.com/agzam/remoto.el</guid>
                <pubDate>Sun, 26 Apr 2026 06:36:55 +0000</pubDate>
                <description><![CDATA[<p>## Remoto.el：在Emacs中直接浏览GitHub仓库

Remoto.el 允许你在Emacs中*无需*克隆即可浏览任何GitHub仓库。它通过注册一个虚拟文件系统来实现，将标准的Emacs文件操作（如`find-file`和`dired`）转换为通过`gh` CLI调用GitHub API。

这对于快速查看代码、README文件或项目结构，而无需下载整个仓库的开销非常理想。它是只读的，利用GitHub的API进行身份验证并透明地处理私有仓库。

主要功能包括：缓存的目录树以实现快速浏览、按需获取文件内容，以及在`dired`和`find-file`中自动检测GitHub URL。该包尚未发布到MELPA，但可通过Straight获取。

Remoto.el 避免使用TRAMP，因为它过于复杂且不适合无状态REST API，而是选择通过Emacs的`file-name-handler-alist`进行更高效、更直接的集成。目前仅限于GitHub，但计划未来支持其他代码托管平台。
</p><p>## Remoto.el: Browse GitHub Repos Directly in Emacs

Remoto.el allows you to browse any GitHub repository within Emacs *without* cloning it. It works by registering a virtual filesystem that translates standard Emacs file operations (like `find-file` and `dired`) into calls to the GitHub API via the `gh` CLI. 

This is ideal for quickly viewing code, READMEs, or project structures without the overhead of downloading an entire repository. It’s read-only, leveraging GitHub’s API for authentication and handling private repositories transparently.

Key features include: cached directory trees for fast browsing, on-demand file content fetching, and automatic GitHub URL detection in `dired` and `find-file`.  The package isn’t yet on MELPA but is available via Straight. 

Remoto.el avoids using TRAMP due to its complexity and unsuitability for a stateless REST API, opting for a more efficient, direct integration via Emacs’ `file-name-handler-alist`. While limited to GitHub currently, future support for other forges is planned.



</p>]]></description>
            </item>
            
            <item>
                <title>重新发现手推车
 Rediscovering the Handcart</title>
                <link>https://solar.lowtechmagazine.com/2026/04/rediscovering-the-handcart/</link>
                <guid>https://solar.lowtechmagazine.com/2026/04/rediscovering-the-handcart/</guid>
                <pubDate>Sun, 26 Apr 2026 06:34:49 +0000</pubDate>
                <description><![CDATA[<p>## 手推车的复兴

手推车，一种历史悠久的人力车辆，正在经历现代复兴。尽管机动运输工具普及，但其简单、经济实惠以及不依赖化石燃料的特点使其在今天仍然具有惊人的现实意义。与现代小型手推车不同，传统手推车体积大、坚固耐用，能够进行长途运输——这是历史上中国以及工业革命时期运输的重要组成部分。

为了减少对汽车的依赖，克里斯·德·德克尔委托定制了一辆手推车，发现它在搬运货物和材料方面非常有用。这辆手推车能够承载与汽车后备箱相当的负载，由于低速和优化的轮子设计，操作起来非常省力。

除了基本的运输功能外，手推车的平台还具有适应性——通过增加座椅、工作台、太阳能板甚至睡眠平台等配件，可以将其改造成移动办公室、发电站甚至游牧住宅。虽然手推车不能完全取代所有车辆，但它为城市环境和社区共享提供了一种可持续、民主且用途广泛的替代方案。它提醒我们，有时最古老的解决方案才是最有效的。
</p><p>## The Resurgence of the Handcart

The humble handcart, a human-powered vehicle millennia old, is experiencing a modern revival. Despite the prevalence of motorized transport, its simplicity, affordability, and independence from fossil fuels make it surprisingly relevant today. Unlike modern, small-wheeled carts, traditional handcarts were large and robust, capable of long-distance hauling – a cornerstone of transport in historical China and, briefly, during the Industrial Revolution.

Driven by a desire to reduce reliance on cars, Kris De Decker commissioned a custom-built handcart, finding it invaluable for moving goods and materials. This cart, capable of carrying loads comparable to a car trunk, operates on minimal effort due to low speed and optimized wheel design. 

Beyond basic transport, the handcart’s platform is adaptable – with additions like a seat, work table, solar panel, and even a sleeping platform – transforming it into a mobile office, power station, or even a nomadic home. While not a replacement for all vehicles, the handcart offers a sustainable, democratic, and surprisingly versatile alternative, particularly suited for urban environments and community sharing. It’s a reminder that sometimes, the oldest solutions are the most effective.



</p>]]></description>
            </item>
            
            <item>
                <title>AGPLv3 §74 赋予用户阻止类似OnlyOffice的徽章软件的能力。
 AGPLv3§74 Empowers Users to Thwart Badgeware Like OnlyOffice</title>
                <link>https://sfconservancy.org/blog/2026/apr/16/badgeware-onlyoffice-nextcloud-affero-gpl/</link>
                <guid>https://sfconservancy.org/blog/2026/apr/16/badgeware-onlyoffice-nextcloud-affero-gpl/</guid>
                <pubDate>Sun, 26 Apr 2026 06:04:41 +0000</pubDate>
                <description><![CDATA[<p>## AGPLv3§7¶4：保护用户免受矛盾许可

布拉德利·M·库恩的文章详细描述了复制许可中反复出现的问题，以及Affero通用公共许可证第3版（AGPLv3）如何专门解决这个问题。问题源于供应商欺骗性地发布看起来像复制许可（如GPLv2）但包含禁止商业再分发的条款——这是一种自我矛盾。

此前，GPLv2没有补救措施，迫使用户放弃面临此类限制的项目。GPLv3和AGPLv3试图通过禁止“进一步限制”的条款来解决这个问题，但它仅适用于下游许可方，而非原始作者。AGPLv3§7¶4创新性地赋予用户*移除*这些矛盾条款的权利。

然而，行使这项权利可能令人望而却步，因为采用这些策略的公司通常具有侵略性和诉讼性，如Neo4j案例所示。最近的一个例子是Ascensio的Onlyoffice，它试图通过对AGPLv3§7的质疑性解释来强制执行商标显示。Nextcloud的分支Euro-Office展示了正确的方法——移除有问题限制。作者强调，虽然在法律上是合理的，但使用AGPLv3§7¶4需要勇气，并预计会面临潜在的反弹。
</p><p>## AGPLv3§7¶4: Protecting Users from Contradictory Licenses

This article by Bradley M. Kühn details a recurring issue in copyleft licensing and how the Affero General Public License, version 3 (AGPLv3), specifically addresses it. The problem stems from vendors deceptively issuing licenses that *appear* copyleft (like GPLv2) but include clauses prohibiting commercial redistribution – a self-contradiction. 

Previously, GPLv2 offered no recourse, forcing users to abandon projects facing such restrictions. GPLv3 and AGPLv3 attempted to solve this with a clause preventing “further restrictions,” but it only applied to downstream licensors, not the original author. AGPLv3§7¶4 innovatively empowers users to *remove* these contradictory terms themselves.

However, exercising this right can be daunting, as companies employing these tactics are often aggressive and litigious, as seen in the Neo4j case. A recent example involves Ascensio’s Onlyoffice, which attempts to enforce trademark display via a questionable interpretation of AGPLv3§7.  Nextcloud’s fork, Euro-Office, demonstrates the correct approach – removing the problematic restriction.  The author stresses that while legally sound, utilizing AGPLv3§7¶4 requires courage and anticipates potential backlash.



</p>]]></description>
            </item>
            
            <item>
                <title>FCC路由器禁令对FOSS的影响
 What the FCC router ban means for FOSS</title>
                <link>https://sfconservancy.org/blog/2026/apr/02/fcc-router-ban/</link>
                <guid>https://sfconservancy.org/blog/2026/apr/02/fcc-router-ban/</guid>
                <pubDate>Sun, 26 Apr 2026 05:36:43 +0000</pubDate>
                <description><![CDATA[<p>## FCC 路由器禁令：摘要

美国联邦通信委员会（FCC）最近禁止销售非美国制造的新型家用路由器——实际上是所有现有型号。FCC以国家安全问题为由，要求寻求豁免的制造商承诺代价高昂且广泛的美国本土制造计划。

预计此决定将显著提高路由器价格，因为美国制造目前无法与亚洲生产竞争。OpenWrt One路由器已经获得FCC批准且不受影响，但该裁决引发了对未来硬件开发的担忧。

FCC可能还会试图限制现有路由器的软件更新，但用户发起的更新（例如安装OpenWrt）仍然不受限制——这是长期安全的关键方面。作者认为，专注于开源软件和既定的安全标准（如NIST IR 8425A）比地域限制更有效。他们倡导一种全球性的、基于标准的科技安全观，而不是优先考虑原产国。他们已联系FCC寻求澄清，并将提供更新。
</p><p>## FCC Router Ban: A Summary

The US Federal Communications Commission (FCC) recently banned the sale of new home routers not manufactured in the United States – effectively all current models. The FCC cites national security concerns as the reason, requiring manufacturers seeking exemption to commit to costly and extensive US-based manufacturing plans.

This decision is expected to significantly increase router prices, as US manufacturing isn’t currently competitive with Asian production. While the OpenWrt One router is already FCC approved and unaffected, the ruling raises concerns for future hardware development. 

The FCC may also attempt to restrict software updates to existing routers, though user-initiated updates (like installing OpenWrt) remain unrestricted – a crucial aspect for long-term security. The author argues that focusing on open-source software and established security standards like NIST IR 8425A is a more effective approach than geographical restrictions. They advocate for a global, standards-based view of tech security, rather than prioritizing country of origin. They have contacted the FCC for clarification and will provide updates.



</p>]]></description>
            </item>
            
            <item>
                <title> The route from Prussian military headquarters to Gary Gygax’s basement</title>
                <link>https://asteriskmag.com/issues/14/shall-we-play-a-game</link>
                <guid>https://asteriskmag.com/issues/14/shall-we-play-a-game</guid>
                <pubDate>Sun, 26 Apr 2026 05:36:26 +0000</pubDate>
                <description><![CDATA[<p></p><p></p>]]></description>
            </item>
            
            <item>
                <title>GnuPG – 主线版本即将加入后量子密码学
 GnuPG – post-quantum crypto landing in mainline</title>
                <link>https://lists.gnupg.org/pipermail/gnupg-announce/2026q2/000504.html</link>
                <guid>https://lists.gnupg.org/pipermail/gnupg-announce/2026q2/000504.html</guid>
                <pubDate>Sun, 26 Apr 2026 05:35:56 +0000</pubDate>
                <description><![CDATA[<p>## GnuPG 2.5.19 发布

GnuPG 的新版本 2.5.19 现已发布，为广泛使用的加密和签名工具带来了新功能和错误修复。主要改进包括新的哈希显示选项和密码算法、增强的智能卡集成以及密钥服务器功能的更新。

此版本继续关注 64 位 Windows 支持，并集成了 Kyber，一种后量子密码学 (PQC) 算法。值得注意的是，即将推出的 2.6 系列将主要关注利用较新的库功能进行内部改进，而不是对 2.4 系列的重大功能更改。

**2.4 系列的用户应尽快更新到 2.5.19，因为 2.4 将在两个月后停止维护。** GnuPG 与先前版本完全向后兼容，并且是根据 GNU 通用公共许可证提供的免费软件。

下载和详细信息，包括签名验证说明，可在 [https://gnupg.org/download/](https://gnupg.org/download/) 获取。Debian 包也可供下载。
</p><p>## GnuPG 2.5.19 Released

A new version of GnuPG, 2.5.19, is now available, bringing new features and bug fixes to the widely-used encryption and signing tool. Key improvements include new options for hash display and cipher algorithms, enhanced smartcard integration, and updates to key server functionality. 

This release continues the focus on 64-bit Windows support and incorporates Kyber, a post-quantum cryptography (PQC) algorithm. Notably, the upcoming 2.6 series will largely focus on internal improvements leveraging newer library features, rather than significant functional changes from the 2.4 series.

**Users of the 2.4 series should update to 2.5.19 promptly, as 2.4 will reach end-of-life in two months.** GnuPG remains fully backward compatible with previous versions and is free software under the GNU General Public License. 

Downloads and detailed information, including signature verification instructions, are available at [https://gnupg.org/download/](https://gnupg.org/download/). Debian packages are also available.



</p>]]></description>
            </item>
            
            <item>
                <title>柯达图像数据集的逐图像PCA特征 (PDF和JSON)
 Per-image PCA characterization of the Kodak image suite (PDF and JSON)</title>
                <link>https://github.com/PearsonZero/kodak-pcd0992-statistical-characterization/tree/main/baseline</link>
                <guid>https://github.com/PearsonZero/kodak-pcd0992-statistical-characterization/tree/main/baseline</guid>
                <pubDate>Sun, 26 Apr 2026 05:34:27 +0000</pubDate>
                <description><![CDATA[<p>这个GitHub仓库，“kodak-pcd0992-statistical-characterization”由PearsonZero创建，是一个公开的统计数据资料集合。它包含24份PDF文档，每份文件名为“KODIMxx_STATISTICAL_PROFILE.pdf”（其中xx从01到24），可能代表不同柯达产品或组件的统计分析。

除了PDF文件，该仓库还包括24个对应的JSON文件（“kodimxx_stats.json”）和一个“kodak_suite_master_stats.json”文件，表明PDF中的数据已被结构化，以便程序化访问和分析。 还有一个`.gitkeep`文件，用于确保基础目录不为空。

该仓库有2颗星，没有fork、issue或pull request，表明它是一个相对较新的或使用较少的资源。 它似乎是统计特征报告的数据转储。
</p><p>This GitHub repository, “kodak-pcd0992-statistical-characterization” by PearsonZero, is a public collection of statistical profile data. It contains 24 PDF documents, each named “KODIMxx_STATISTICAL_PROFILE.pdf” (where xx ranges from 01 to 24), likely representing statistical analyses for different Kodak products or components. 

Alongside the PDFs, the repository includes 24 corresponding JSON files (“kodimxx_stats.json”) and a “kodak_suite_master_stats.json” file, suggesting the data from the PDFs has been structured for programmatic access and analysis. A single `.gitkeep` file is also present, used to ensure the baseline directory isn’t empty. 

The repository has 2 stars and no forks, issues, or pull requests, indicating it's a relatively new or less actively used resource. It appears to be a data dump of statistical characterization reports.</p>]]></description>
            </item>
            
            <item>
                <title>自制PBX
 My Homemade PBX (2002)</title>
                <link>https://wandel.ca/homepage/pbx.html</link>
                <guid>https://wandel.ca/homepage/pbx.html</guid>
                <pubDate>Sun, 26 Apr 2026 05:03:46 +0000</pubDate>
                <description><![CDATA[<p>## 自制PBX：终身追求

作者儿时对电信产生浓厚兴趣，从自制对讲机和短波收音机开始，在1992-93年间构建了一个功能齐全的八分机PBX。这个项目得益于作者在贝尔-北方研究公司（Bell-Northern Research）的实习工作，获得了组件和知识，以及淘便宜的二手电话。

该系统复制了电信规范（48V挂机状态、振铃电压、环路电流），并支持DTMF和脉冲拨号，根据需要进行转换。它允许三个同时通话，标准呼叫进度音，并具有诸如保持、速拨、呼叫转移和加入正在进行的通话等功能。

该设计使用容易获得的组件和微控制器（8031），优先考虑功能而非效率，利用继电器进行切换，并采用过采样技术进行音调生成。虽然缺乏详细的电路图，但作者提供了对硬件和软件的见解，强调其教育和娱乐价值。最终的PBX性能与商业系统无异，实现了作者长久以来的雄心壮志，并提供了一个独特而引人入胜的项目。
</p><p>## Homemade PBX: A Lifelong Pursuit

Driven by a childhood fascination with telecommunications – starting with homemade intercoms and shortwave radios – the author built a fully functional, eight-extension PBX in 1992-93. This project was fueled by a co-op job at Bell-Northern Research granting access to components and knowledge, alongside bargain-hunting for used telephones.

The system replicates telco specifications (48V on-hook, ringing voltage, loop current) and supports both DTMF and pulse dialing, converting between them as needed. It allows for three simultaneous calls, standard call progress tones, and features like hold, speed dial, call forwarding, and joining calls in progress. 

Built using readily available components and a microcontroller (8031), the design prioritized functionality over efficiency, utilizing relays for switching and oversampling for tone generation. While lacking detailed schematics, the author provides insights into the hardware and software, emphasizing its educational and entertainment value. The resulting PBX performs indistinguishably from a commercial system, fulfilling a long-held ambition and providing a unique, engaging project.



</p>]]></description>
            </item>
            
            <item>
                <title>欧盟年龄控制：数字身份的特洛伊木马
 EU Age Control: The trojan horse for digital IDs</title>
                <link>https://juraj.bednar.io/en/blog-en/2026/04/17/eu-age-control-the-trojan-horse-for-digital-ids/</link>
                <guid>https://juraj.bednar.io/en/blog-en/2026/04/17/eu-age-control-the-trojan-horse-for-digital-ids/</guid>
                <pubDate>Sun, 26 Apr 2026 05:00:16 +0000</pubDate>
                <description><![CDATA[<p>## 欧盟年龄验证应用： критический взгляд

欧盟计划推出的旨在保护儿童在线安全的年龄验证应用正面临重大障碍，可能无法兑现承诺的隐私保护。虽然宣传中称其利用零知识证明进行匿名年龄确认，但实际情况更为复杂。平台可以通过使用标准的KYC（了解你的客户）程序——扫描完整身份证件并收集大量个人数据——完全绕过以隐私为中心的系统。

此外，该系统严重依赖谷歌和苹果进行设备证明，实际上排除了像GrapheneOS或华为手机等替代操作系统。 宣传中的“零知识”密码学甚至在当前的参考应用中并未激活，而是依赖于更旧、安全性较低的方法。

至关重要的是，该系统的不可链接性——防止网站跟踪用户——取决于钱包的*正确*行为，而非固有的密码学安全性，并且容易受到“中继攻击”，成年人可以为未成年人进行验证。 尽管声称有一个统一的“欧盟应用”，但实施将落到各个国家手中，很可能导致碎片化且可能存在缺陷的系统。

最终，该倡议有可能成为更广泛的数字身份计划的特洛伊木马，优先考虑控制而非真正的隐私，并可能为监控创造新的途径。 虽然意图是积极的，但技术限制和政治现实表明，依赖现有的、隐私性较低的方法的可能性更大。
</p><p>## EU Age Verification Apps: A Critical Look

EU plans for age verification apps aiming to protect children online are facing significant hurdles and may not deliver on promised privacy. While marketed as utilizing zero-knowledge proofs for anonymous age confirmation, the reality is more complex. Platforms can bypass the privacy-focused system entirely by using standard KYC (Know Your Customer) procedures – scanning full IDs and collecting extensive personal data.

Furthermore, the system relies heavily on Google and Apple for device attestation, effectively locking out alternative operating systems like GrapheneOS or Huawei phones. The touted “zero-knowledge” cryptography isn’t even active in the current reference app, relying instead on older, less secure methods. 

Crucially, the system’s unlinkability – preventing sites from tracking users – depends on wallets *behaving* correctly, not inherent cryptographic security, and is vulnerable to “relay attacks” where adults can verify for minors. Despite claims of a unified “EU app,” implementation will fall to individual countries, likely resulting in fragmented and potentially flawed systems. 

Ultimately, the initiative risks becoming a Trojan horse for broader digital ID schemes, prioritizing control over genuine privacy and potentially creating new avenues for surveillance. While the intention is positive, the technical limitations and political realities suggest a reliance on existing, less private methods is far more likely.



</p>]]></description>
            </item>
            
            <item>
                <title>超级任天堂卡带
 The Super Nintendo Cartridges (2024)</title>
                <link>https://fabiensanglard.net/snes_carts/</link>
                <guid>https://fabiensanglard.net/snes_carts/</guid>
                <pubDate>Sun, 26 Apr 2026 04:33:48 +0000</pubDate>
                <description><![CDATA[<p>## 超级任天堂内部：不仅仅是ROM

超级任天堂卡带包含的不仅仅是游戏代码和图像。除了ROM芯片外，它们还包含诸如防盗版（CIC）芯片、游戏存档存储器（SRAM），甚至用于提升性能的“增强处理器”等组件。CIC芯片通过与主机中的芯片通信来防止盗版，如果出现不匹配则会重置系统。

ROM的大小差异很大——《星之海洋》使用了巨大的48Mb，而《超级马里奥世界》仅占用4Mb。已经整理了一份包含3,378个游戏ROM大小的完整列表。

然而，真正的创新在于增强处理器。**Super FX**芯片驱动了《星狐》的3D图形，而**SA-1**为《超级马里奥RPG》等游戏提供了更快的CPU。其他芯片，如《洛克人X2》中的**CX4**，处理复杂的精灵操作，而**DSP-1**则改进了《超级马里奥卡特》中的Mode 7效果。

这些芯片通常需要独特的工程解决方案，并给模拟器开发者带来了挑战，其中一些直到2012年才完全克服。如今，复古游戏社区继续利用这些芯片，甚至通过像Super FX项目这样的项目来消除旧游戏的卡顿并增强图形，从而改进旧游戏。
</p><p>## Inside the Super Nintendo: More Than Just ROMs

Super Nintendo cartridges contained much more than just game code and graphics. Beyond the ROM chips, they housed components like copy protection (CIC) chips, save game memory (SRAM), and even “enhancement processors” to boost performance. The CIC chip prevented piracy by communicating with a chip in the console, resetting the system if a mismatch occurred.

ROM sizes varied greatly – *Star Ocean* utilized a massive 48Mb, while *Super Mario World* fit within just 4Mb. A comprehensive list of ROM sizes for 3,378 titles has been compiled.

However, the real innovation lay in enhancement processors. The **Super FX** chip powered *Star Fox’s* 3D graphics, while the **SA-1** offered a faster CPU for games like *Super Mario RPG*. Others, like the **CX4** in *Mega Man X2*, handled complex sprite manipulation, and the **DSP-1** improved Mode 7 effects in *Super Mario Kart*. 

These chips often required unique engineering solutions and presented challenges for emulator developers, some of which weren’t fully overcome until 2012. Today, the retro-gaming community continues to leverage these chips, even improving older games by removing slowdowns and enhancing graphics through projects like Project Super FX.



</p>]]></description>
            </item>
            
            <item>
                <title>麻将：视觉指南
 Mahjong: A Visual Guide</title>
                <link>https://themahjong.guide/</link>
                <guid>https://themahjong.guide/</guid>
                <pubDate>Sun, 26 Apr 2026 04:30:47 +0000</pubDate>
                <description><![CDATA[<p>Please provide the content you want me to translate. I need the text to be able to translate it to Chinese.
</p><p>
</p>]]></description>
            </item>
            
            <item>
                <title>互联网广告有效吗？
 Does Internet Advertising Work?</title>
                <link>https://freakonomics.com/podcast/does-advertising-actually-work-part-2-digital-ep-441/</link>
                <guid>https://freakonomics.com/podcast/does-advertising-actually-work-part-2-digital-ep-441/</guid>
                <pubDate>Sun, 26 Apr 2026 04:05:03 +0000</pubDate>
                <description><![CDATA[<p>启用 JavaScript 和 Cookie 以继续。
</p><p>Enable JavaScript and cookies to continue</p>]]></description>
            </item>
            
            <item>
                <title>DeepSeek-V4发布：从快速推理到SGLang和Miles验证的强化学习
 DeepSeek-V4 on Day 0: From Fast Inference to Verified RL with SGLang and Miles</title>
                <link>https://www.lmsys.org/blog/2026-04-25-deepseek-v4/</link>
                <guid>https://www.lmsys.org/blog/2026-04-25-deepseek-v4/</guid>
                <pubDate>Sun, 26 Apr 2026 04:04:20 +0000</pubDate>
                <description><![CDATA[<p>## DeepSeek-V4 Day-0 支持与 SGLang & Miles

SGLang 和 Miles 是首个开源堆栈，在 DeepSeek-V4 (1.6T Pro & 284B Flash) 发布时提供全面支持——包括推理和强化学习训练。该支持专为 DeepSeek-V4 的独特架构设计，包括混合稀疏注意力、流形约束超连接 (mHC) 和 FP4 专家权重。

关键创新包括 **ShadowRadix**，用于混合注意力的高效前缀缓存，以及 **HiSparse** 等优化，用于将非活动 KV 缓存卸载到 CPU 以实现更大的批处理大小。 推理受益于快速内核集成 (FlashMLA, TRTLLM-Gen MoE, DeepGEMM Mega MoE, TileLang mHC) 和广泛的并行性。

对于强化学习训练，该堆栈提供完全并行性 (DP/TP/SP/EP/PP/CP)、Tilelang 注意力以及通过 FP8 训练增强的稳定性。 测试表明训练动态稳定且性能具有竞争力，重点是保持精度并解决数值稳定性问题。 基准测试显示强大的解码吞吐量，尤其是在较长的上下文环境中。

此次发布代表着可访问的大型语言模型技术向前迈出的重要一步，由协作努力支持，并在详细的博客文章中概述了设计和优化。
</p><p>## DeepSeek-V4 Day-0 Support with SGLang & Miles

SGLang and Miles are the first open-source stack to offer full support – both inference and reinforcement learning training – for DeepSeek-V4 (1.6T Pro & 284B Flash) upon its launch. This support is specifically designed for DeepSeek-V4’s unique architecture, including hybrid sparse attention, manifold-constrained hyper-connections (mHC), and FP4 expert weights.

Key innovations include **ShadowRadix** for efficient prefix caching with hybrid attention, and optimizations like **HiSparse** to offload inactive KV cache to CPU for larger batch sizes.  Inference benefits from fast kernel integrations (FlashMLA, TRTLLM-Gen MoE, DeepGEMM Mega MoE, TileLang mHC) and extensive parallelism. 

For RL training, the stack provides full parallelism (DP/TP/SP/EP/PP/CP), Tilelang attention, and enhanced stability with FP8 training.  Testing demonstrates stable training dynamics and competitive performance, with a focus on maintaining precision and addressing numerical stability issues. Benchmarks show strong decode throughput, particularly on longer contexts. 

This launch represents a significant step forward in accessible large language model technology, supported by a collaborative effort and detailed in a comprehensive blog post outlining the design and optimizations.



</p>]]></description>
            </item>
            
            <item>
                <title>2026年重振BrowserID
 Reviving BrowserID in 2026</title>
                <link>https://wakamoleguy.com/p/reviving-browserid-in-2026</link>
                <guid>https://wakamoleguy.com/p/reviving-browserid-in-2026</guid>
                <pubDate>Sun, 26 Apr 2026 04:02:35 +0000</pubDate>
                <description><![CDATA[<p>## 为个人应用复兴BrowserID

在2026年，一位开发者正在构建“WKID”，一种基于先前已废弃的BrowserID协议的新身份提供者（IdP）。其动机源于小型、由个人构建的应用的兴起，这些应用受益于更易用的开发工具，如LLM。这些“定制应用”需要用户管理，但像Google或Auth0这样的现有解决方案会引入不必要的依赖和潜在的账户限制。

BrowserID提供了一种引人注目的替代方案：它是联合的、私密的、轻量级的。它利用电子邮件域进行身份验证——允许开发者完全控制——并避免了将每个应用注册到中央服务器的需求。

与Mozilla在2016年失败的尝试不同，WKID并非旨在广泛采用。它专为开发者的自有项目以及亲密朋友/家人的项目而设计，所有这些项目都使用他们控制的电子邮件域。这消除了之前困扰BrowserID的“先有鸡还是先有蛋”的问题，因为成功由个人效用定义，而非市场份额。虽然要求用户拥有支持WKID的电子邮件域，但在定制应用的环境下是可以接受的。开发仍在进行中，重点是完善和文档。
</p><p>## Reviving BrowserID for Personal Apps

In 2026, a developer is building “WKID,” a new identity provider (IdP) based on the previously defunct BrowserID protocol. The motivation stems from the rise of small, personally-built applications fueled by easier development tools like LLMs. These “bespoke apps” require user management, but existing solutions like Google or Auth0 introduce unwanted dependencies and potential account restrictions.

BrowserID offers a compelling alternative: it’s federated, private, and lightweight. It leverages email domains for authentication – allowing the developer full control – and avoids the need to register each app with a central server. 

Unlike Mozilla’s failed 2016 attempt, WKID isn’t aiming for widespread adoption. It’s designed specifically for the developer’s own projects and those of close friends/family, all using email domains they control. This eliminates the “chicken-and-egg” problem that plagued BrowserID previously, as success is defined by personal utility, not market share. While requiring users to have email domains supporting WKID, this is acceptable within the context of bespoke applications. Development is ongoing, focusing on polishing and documentation.



</p>]]></description>
            </item>
            
            <item>
                <title>美国梦已正式破灭。
 The death of the American Dream is now official</title>
                <link>https://thehill.com/opinion/finance/5846892-american-dream-debt-crisis/</link>
                <guid>https://thehill.com/opinion/finance/5846892-american-dream-debt-crisis/</guid>
                <pubDate>Sun, 26 Apr 2026 03:35:41 +0000</pubDate>
                <description><![CDATA[<p>您的浏览器似乎禁用了 Javascript。请点击此处了解如何启用 Javascript。如果您有任何问题，请通过 challengehelp@humansecurity.com 与我们联系。
</p><p>
Your browser appears to have Javascript disabled.For instructions on how to enable Javascript please click here.If you have any issues, please contact us at challengehelp@humansecurity.com
</p>]]></description>
            </item>
            
            <item>
                <title>乔治商业，罗杰·泽拉兹尼 (1980)
 The George Business, by Roger Zelazny (1980)</title>
                <link>https://www.eternal-flame.org/library/oldlibrary/georgebusiness.html</link>
                <guid>https://www.eternal-flame.org/library/oldlibrary/georgebusiness.html</guid>
                <pubDate>Sun, 26 Apr 2026 03:33:32 +0000</pubDate>
                <description><![CDATA[<p>## 乔治的生意：概要

达特是一条务实的龙，被一个预言梦所困扰，梦中他将与一位酷似圣乔治的骑士战斗。令他惊讶的是，这位名叫乔治的骑士*渴望*战斗——但不是为了宝藏或少女。他雇佣达特执行一个古怪的计划：伪造巴伦·莫里斯的女儿罗莎琳德被绑架，让乔治“拯救”她，并在她眼中显得英勇。

达特虽然怀疑，但被提供的黄金所吸引，同意了。 演练的战斗顺利进行，但计划却适得其反，罗莎琳德立刻嫁给了一个野蛮人，而巴伦却意外去世，导致乔治的竞争对手上位。

乔治毫不气馁，提出了一个新的商业计划：他将侦察富裕但缺乏英雄的城镇，而达特将扮演反派，让乔治“杀死”他以获取报酬。达特认识到乔治非同寻常的、类似龙的狡猾，于是接受了，建立了一个有利可图的、非常规的伙伴关系，建立在演练的战斗和共同利益之上。他们为他们的新事业干杯，准备利用人类对英雄和龙的需求。
</p><p>## The George Business: A Summary

Dart, a pragmatic dragon, is disturbed by a prophetic dream of battling a knight resembling the legendary Saint George. To his surprise, the knight, named George, *wants* to fight – but not for treasure or a damsel. He hires Dart for a bizarre scheme: to fake an abduction of Baron Maurice’s daughter, Rosalind, allowing George to “rescue” her and appear heroic in her eyes.

Dart, skeptical but intrigued by the gold offered, agrees. The staged battle goes off without a hitch, but the plan backfires when Rosalind immediately marries a barbarian, and the Baron unexpectedly dies, elevating George’s rival. 

Undeterred, George proposes a new business venture: he’ll scout wealthy, hero-less towns, and Dart will play the villain, allowing George to “slay” him for a fee. Dart, recognizing George’s unusual, dragon-like cunning, accepts, establishing a profitable, if unconventional, partnership built on staged battles and mutual benefit. They toast to their new enterprise, ready to exploit the human need for heroes and dragons alike.



</p>]]></description>
            </item>
            
            <item>
                <title>业余爱好者利用ChatGPT解决了一个埃尔德什问题。
 Amateur armed with ChatGPT solves an Erdős problem</title>
                <link>https://www.scientificamerican.com/article/amateur-armed-with-chatgpt-vibe-maths-a-60-year-old-problem/</link>
                <guid>https://www.scientificamerican.com/article/amateur-armed-with-chatgpt-vibe-maths-a-60-year-old-problem/</guid>
                <pubDate>Sun, 26 Apr 2026 03:00:16 +0000</pubDate>
                <description><![CDATA[<p>## 人工智能解决长期存在的数学难题

一位23岁的业余数学家利安·普莱斯最近使用OpenAI的ChatGPT Pro解决了保罗·埃尔德什提出的一个60年未解的问题——这项成就让专业数学家们困惑了几十年。这个问题涉及“原始数集”，其中一个数不能被另一个数整除，以及埃尔德什关于这些集合的最低可能“得分”的猜想。

使这个解决方案与众不同的是*如何*实现的。与之前在埃尔德什问题上取得的AI成功不同，ChatGPT似乎采用了完全新颖的方法，以一种意想不到的方式使用了已知公式。加州大学洛杉矶分校的特伦斯·陶等专家认为，人工智能绕过了阻碍人类数学家的“思维障碍”。

虽然最初的输出需要专家进行完善，但核心见解已经显示出在数论领域更广泛应用的前景，可能揭示大数之间的新的联系。这一突破表明这些问题可能比以前想象的更简单，并验证了人们对它们相互关联性的现有直觉，为数学探索提供了新的视角。
</p><p>## AI Solves Long-Standing Math Problem

A 23-year-old amateur mathematician, Liam Price, recently solved a 60-year-old unsolved problem posed by Paul Erdős – a feat that baffled professional mathematicians for decades – using OpenAI’s ChatGPT Pro. The problem concerned “primitive sets” of numbers, where no number is divisible by another, and Erdős’s conjecture about the lowest possible “score” for these sets.

What sets this solution apart is *how* it was achieved. Unlike previous AI successes on Erdős problems, ChatGPT appears to have utilized a completely novel approach, employing a known formula in an unexpected way. Experts like Terence Tao at UCLA believe the AI bypassed a “mental block” hindering human mathematicians. 

While the initial output required refinement by experts, the core insight is already showing promise for broader applications in number theory, potentially revealing new connections between large numbers. This breakthrough suggests these problems may have been simpler than previously thought and validates existing intuitions about their interconnectedness, offering a fresh perspective on mathematical exploration.



</p>]]></description>
            </item>
            </channel></rss>