2025-06-04

35 篇热帖

1. FFmpeg merges WebRTC support (git.ffmpeg.org)

Anubis 防护系统摘要

目的与背景 Anubis 是一种用于保护网站服务器免受AI公司大规模数据抓取(scraping)的系统。其设计目的是缓解因自动化抓取导致的服务器负载过高和网站服务不可用的问题。

工作机制

  1. 基于工作量证明(Proof-of-Work):系统采用类似Hashcash(一种用于减少垃圾邮件的工作量证明方案)的机制。
  2. 设计理念:对单个合法用户而言,完成证明所需的额外计算负担可以忽略不计;但对于大规模抓取者,累积的成本会变得非常高昂,从而有效增加其抓取成本。

当前状态与未来计划

  • 临时解决方案:Anubis 目前被描述为一个“占位符”或临时性解决方案。
  • 长期目标:未来计划转向更先进的指纹识别和无头浏览器(headless browsers)检测技术(例如通过字体渲染方式识别),以期能够直接区分合法用户与抓取程序,从而可能无需向大多数正常用户显示证明页面。

技术要求与提示

  • 该系统依赖现代JavaScript功能运行。
  • 用户需要为相关域名禁用像 JShelter 这类可能干扰正常脚本执行的浏览器插件。
3. Why I wrote the BEAM book (happihacking.com)

The story behind writing The BEAM Book-ten years, three publishers, and an open‑source sprint.

4. The Right to Repair Is Law in Washington State (www.eff.org)

Thanks in part to your support, the right to repair is now law in Washington.Gov. Bob Ferguson signed two bills guaranteeing Washingtonians' right to access tools, parts, and information so they can fix personal electronics, appliances, and wheelchairs. This is the epitome of common-sense...

5. Merlin Bird ID (merlin.allaboutbirds.org)

Identify Bird Songs and Calls

Sound ID listens to the birds around you and shows real-time suggestions for who’s singing. Compare your recording to the songs and calls in Merlin to confirm what you heard. Sound ID works completely offline, so you can identify birds you hear no matter where you ar

6. Deep learning gets the glory, deep fact checking gets ignored (rachel.fast.ai)

When impressive AI biology results are full of errors

7. (On | No) Syntactic Support for Error Handling (go.dev)

Go team plans around error handling support

9. Precious Plastic is in trouble (www.preciousplastic.com)

News update: Precious Plastic is in problems. You can decide the future of the project. Die or grow

12. Swift at Apple: Migrating the Password Monitoring Service from Java (www.swift.org)

Swift is heavily used in production for building cloud services at Apple, with incredible results. Last year, the Password Monitoring service was rewritten in Swift, handling multiple billions of requests per day from devices all over the world. In comparison with the previous Java service, the updated backend delivers a 40% increase in performance, along with improved scalability, security, and availability.

13. The Prompt Engineering Playbook for Programmers (addyo.substack.com)

Turn AI coding assistants into more reliable development partners

14. "AI Will Replace All the Jobs " Is Just Tech Execs Doing Marketing (sparktoro.com)

Over the weekend, I went digging for evidence that AI can, will, or has replaced a large percent of jobs. It doesn't exist. Worse than that, actually,

15. A manager is not your best friend (staysaasy.com)

管理者不应成为你最好的朋友

核心观点

管理者与下属之间的共情(抱怨、附和)是一种组织毒药,会侵蚀信任、形成小团体、阻碍问题解决,即使抱怨内容是虚构的。管理者的职责是探求真相、建设性地解决问题,而非无条件地提供情感支持。

共情的危害

  • 破坏组织结构:在团队与其他部门之间制造对立,阻碍问题反馈与改进。
  • 营造低信任环境:即使抱怨基于虚构事实,也会助长团队优越感和受害者心态。
  • 形成恶性循环:例如,若管理者与下属共情抱怨HR,团队可能轻视HR,而HR不知原因也无法改进,最终互相指责。

管理者的正确角色

管理者的共情必须是有条件的,职责是解决问题而非单纯安抚情绪。与私人友谊中无条件支持不同,管理者需要:

  • 像科学家一样:通过提问探究问题的灰色地带(90%的情况),而非直接站队。
  • 像好治疗师一样:允许倾诉,但通过提问(“这确实是事实吗?”)澄清事实,验证感受而非必然认同事实
  • 创造健康视角:帮助团队理解其他部门的动机(如销售激励的必要性),降低冲突情绪。
  • 用客观框架替代主观抱怨:例如,不说“市场部无能”,而说“我们的职责是提供产品观点,他们的职责是确保发布产生足够管线;如果做不到,我们可以共同沟通。”

应对共情请求的策略

  1. 追问细节:多数情况是灰色地带,需厘清事实。
  2. 避免贬低性语言:不使用“我不知道他们为什么这么……”,转而为他方行为提供合理推测以培养同理心。
  3. 注意非语言暗示:眼神、点头等细微动作可能无声传达共情,管理者需警惕。
  4. 将抱怨转化为行动:当被要求共情时,重申决策框架并聚焦客观事实。
  5. 处理对上级的抱怨:避免共情,转为讨论“如何更有效与上级合作”。
  6. 在极端情况下及时止损:若他人确有重大失误,表态“我会跟进”而非认同抱怨,关闭共情通道。

管理者自身的共情需求

管理者也可能需要情绪宣泄,但应选择:

  • 与同级或非直接下属共情,避免在团队内发出错误信号。
  • 最短时间内抽离情绪,避免负面情绪累积影响判断。

核心原则

管理者的沟通应减少情绪、聚焦行动,致力于构建解决问题的信任基础,而非通过共情强化对立。

16. Machine Code Isn't Scary (jimmyhmiller.com)

机器码并不可怕

核心观点

作者从学习高级Web语言起步,曾认为底层语言(尤其是机器码)难以理解。通过实践学习后,他发现机器码并非想象中那般可怕。其核心观点是:如果能确保JSON符合模式,就能编写机器码。 关键在于理解其基本构造原理。

机器码基础概念

理解机器码需要三个核心概念:

  1. 指令:机器执行的操作码,以二进制数字(例如在AArch64中为32位数字)表示,编码了操作类型(如加、减、跳转)及操作数(可以是常数、寄存器或内存地址)。
  2. 寄存器:用于临时存储数据的高速小型存储单元。不同指令集(如AArch64和x86-64)的寄存器数量、大小、命名和用途(调用约定)各不相同。
  3. 内存:可视为一个大型数组,用于存储数据。机器码中有专门的指令(如STR)来读写内存。

以AArch64(ARM 64位)为例

文章以AArch64指令集为例进行讲解。

  • 指令编码:每条指令是一个32位的二进制数,不同位段(bit field)代表不同参数(如操作类型、立即数、源寄存器、目标寄存器)。例如,一个ADD指令的二进制格式被解析为包含是否使用64位寄存器(sf)、是否移位(sh)、12位立即数(imm12)、源寄存器(Rn)和目标寄存器(Rd)等字段的结构体。
  • 汇编语言:为避免直接操作繁琐的二进制或十六进制,通常使用汇编语言(如add x1, x0, #0x2a)作为机器码的可读表示。一条汇编指令可能对应多种机器码编码。
  • 内存操作:以STR(存储)指令为例,展示了如何将寄存器的值存储到指定的内存地址(基址寄存器 + 偏移量)中。

对比X86-64架构

文章简要对比了主流的x86-64架构:

  • 主要区别:x86-64不是固定长度指令集,指令由可变部分组装而成,编码更复杂。
  • 关键组件:介绍了x86指令编码的三个部分:
    • REX前缀:用于支持64位操作和扩展寄存器访问(如访问r8-r15)。
    • ModR/M:用于指定操作模式(寄存器或内存指针)和操作数。
    • 操作码:指定具体操作的数字。
  • 示例:展示了如何通过组合REX、操作码和ModR/M等部分,将立即数42移动到RBX寄存器的x86-64指令编码过程。

总结与建议

作者指出,学习机器码虽然不会在本文内穷尽所有细节(如标志位、栈、跳转等),但掌握这些基础知识已能提供强大的入门指引。他强调了从底层实践学习的巨大价值:

  • 消除知识鸿沟:能填补依赖高级库时产生的理解空白。
  • 打破心理障碍:底层细节本身并不难,关键在于缺乏清晰易懂的文档和解释。
  • 实践建议:鼓励读者利用在线工具(如Compiler Explorer)从最基础层面开始学习,以真正克服对底层技术的恐惧。
18. A deep dive into self-improving AI and the Darwin-Gödel Machine (richardcsuwandi.github.io)

A deep dive into self-improving AI and the Darwin-Gödel Machine

19. Binary Wordle (wordle.chengeric.com)

Binary Wordle 是一个二进制版本的 Wordle 游戏,玩家可以使用键盘(0、1、Enter、Backspace 键)或屏幕上的按钮进行游戏操作。该游戏聚焦于二进制数字(0 和 1),是传统 Wordle 游戏的变体。此外,内容包含赞助商 altodial.com 的广告,宣传其服务以避免等待,并附带相关链接和图标。

20. Show HN: AirAP AirPlay server - AirPlay to an iOS Device (github.com)

airplay to an ios device. Contribute to neon443/AirAP development by creating an account on GitHub.

22. Meta pauses mobile port tracking tech on Android after researchers cry foul (www.theregister.com)

: Zuckercorp and Yandex used localhost loophole to tie browser data to app users, say boffins

23. Go is a good fit for agents (docs.hatchet.run)

Golang might just be the best language for building AI agents.

24. How we reduced the impact of zombie clients (letsencrypt.org)

Every night, right around midnight (mainly UTC), a horde of zombies wakes up and clamors for … digital certificates! The zombies in question are abandoned or misconfigured Internet servers and ACME clients that have been set to request certificates from Let’s Encrypt. As our certificates last for at most 90 days, these zombie clients’ software knows that their certificates are out-of-date and need to be replaced. What they don’t realize is that their quest for new certificates is doomed! These devices are cursed to seek certificates again and again, never receiving them.

25. Cloud Run GPUs, now GA, makes running AI workloads easier for everyone (cloud.google.com)

Support for GPUs in Cloud Run makes running GPU-accelerated applications simpler, faster, and more cost-effective, as well as enabling new use cases.

26. Show HN: Ephe – A minimalist open-source Markdown paper for today (github.com)

An Ephemeral Markdown Paper. Less but handy features for plain Markdown lovers. - unvalley/ephe

27. Changing Directions (jacobian.org)

I have two important announcements:

  1. I’m leaving the tech industry. Hopefully “for good”; if not, at least “for now”.

  2. As such, the content on this blog is going to shift, perhaps dramatically. I’m going to be writing about a broader range of topics that interest me (projects around my hobby farm, wilderness trips, emergency medicine) – more writing for me, less writing for some imagined audience. (I’ll probably still end up writing about some of the same topics as I’ve been covering since 2020, just less often.)

I’m writing this post mostly to give myself permission to make that change, and to give readers the opportunity to unsubscribe/unfollow if they’re not interested.

If you’re interested in more details about why I’m leaving the industry and what’s next for me and this blog, read on.

28. The Sky's the limit: AI automation on Mac (taoofmac.com)

I’ve been sitting on this draft for a few days now, partly because I thought it would turn down the bitterness, and partly because I kept as...

29. Mapping latitude and longitude to country, state, or city (austinhenley.com)

I made a library that takes GPS coordinates and gives you the state.

30. AGI is not multimodal (thegradient.pub)

"In projecting language back as the model for thought, we lose sight of the tacit embodied understanding that undergirds our intelligence." –Terry Winograd

The recent successes of generative AI models have convinced some that AGI is imminent. While these models appear to capture the essence of human intelligence, they defy

32. Doubling Down on Open Source (langfuse.com)

Open Sourcing all Product Features in Langfuse under the MIT license

33. A critical look at NetBSD’s installer (eerielinux.wordpress.com)

[New to Gemini? Have a look at my Gemini FAQ.] This article was bi-posted to Gemini and the Web; Gemini version is here: gemini://gemini.circumlunar.space/users/kraileth/neunix/2025/installing_bsd_pt3.gmi The first part of this series was about FreeBSD's installer and the second covered OpenBSD's installation program. This is a longer article and it has the most images that I ever…