2025-04-19

30 篇热帖

1. Synology Lost the Plot with Hard Drive Locking Move (www.servethehome.com)

According to reports, Synology plans to restrict features to its own branded hard drives with the new 2025 Plus models. This is bad

4. Claude Code: Best practices for agentic coding (www.anthropic.com)

Tips and patterns for getting the most out of Claude Code, from configuring your environment to scaling across parallel sessions.

5. Show HN: I made a Doom-like game fit inside a QR code (github.com)

A self-contained game that fits inside a QR code inspired by DOOM 1993 and The Backrooms - Kuberwastaken/backdooms

6. Judge Rules Blanket Search of Cell Tower Data Unconstitutional (www.404media.co)

Judge says tower dumps violate the 4th amendment, but will let the cops do it this one time, as a treat.

7. Android phones will soon reboot themselves after sitting unused for three days (arstechnica.com)

The latest Google update will make your phone more secure if you don't touch it...

8. 15,000 lines of verified cryptography now in Python (jonathan.protzenko.fr)

In November 2022, I opened issue 99108 on Python’s GitHub repository, arguing that after a recent CVE in its implementation of SHA3, Python should embrace verified code for all of its hash-related infrastructure.

10. Solidjs: Simple and performant reactivity for building user interfaces (www.solidjs.com)

SolidJS 概述

SolidJS 是一个用于构建用户界面的JavaScript库。它以其简单且高性能的响应式特性为核心,提供声明式、高效和灵活的编程方式。该库专注于优化用户界面开发,通过响应式系统实现高效的状态管理,并支持声明式代码风格,以提升开发效率和应用性能。

11. Restoring the Galaxian3 Theatre 6, 1992 six player arcade machine (philwip.com)

Introduction During the late 1980s, Namco set out to create the world's largest arcade game. The result was Galaxian3: Project Dragoon; a 28 player behemoth that debuted in April 1990 at the International Garden and Greenery Exposition in Osaka, Japan. Codenamed GH-28, it featured real-time, flat-shaded 3D graphics, powered by hardware derived from the System…

12. Hypertext TV (hypertext.tv)

Daily programming for the handmade web. Airing now.

13. Full Text Search of US Court records (www.judyrecords.com)

Instantly search 760 million+ United States court cases.

16. Making a smart bike dumb so it works again (francisco.io)

My smart bike company went bankrupt, so I added a physical button to turn the light on/off

17. A Map of British Dialects (2023) (starkeycomics.com)

This map took me a long time to make, and is very detailed, but will always be incomplete and inaccurate due to the nature of language. Why this map is so detailed The diversity of English dialects in the United Kingdom is enormous. It’s common for people from either side of a river, mountain, or […]

18. JavaScript Views, the Hard Way – A Pattern for Writing UI (github.com)

JavaScript views; done as painfully as possible. Contribute to matthewp/views-the-hard-way development by creating an account on GitHub.

19. Hands-On Large Language Models (github.com)

Official code repo for the O'Reilly Book - "Hands-On Large Language Models" - HandsOnLLM/Hands-On-Large-Language-Models

20. There's Life Inside Earth's Crust (www.noemamag.com)

Revelations about microbes living deep inside Earth’s crust are challenging scientists’ conceptions of life and how it evolves.

21. Show HN: Undercutf1 – F1 Live Timing TUI with Driver Tracker, Variable Delay (github.com)

F1 Live Timing TUI for all F1 sessions with variable delay to sync to your TV. Supports replaying previously recorded sessions. - JustAman62/undercut-f1

23. OpenAI's new reasoning AI models hallucinate more (techcrunch.com)

OpenAI's reasoning AI models are getting better, but their hallucinating isn't, according to benchmark results.

24. UML diagram for the DDD example in Evans' book (github.com)

UML diagram for the DDD example—a cargo shipping system—from Eric Evans' book - takaakit/uml-diagram-for-ddd-example-in-evans-book

25. How to Write a Fast Matrix Multiplication from Scratch with Tensor Cores (2024) (alexarmbr.github.io)

This is my blog

27. I passionately hate hype, especially the AI hype (unixdigest.com)

对技术炒作(尤其是AI炒作)的批判

核心观点:作者强烈反对技术领域的炒作行为,认为其本质是商业利益驱动的夸大宣传,最终损害投资者、企业和个人利益,并可能导致隐私与自由的丧失。AI是当前最突出的炒作案例。

炒作的运作机制与危害

  • 起源:公司或行业为盈利,发起大规模宣传。
  • 传播:媒体为利润放弃客观性,盲目跟风;大众(被比作“盲从者”)不加思考地接受并传播。
  • 后果
    • 投资者:因虚假承诺蒙受经济损失。
    • 企业(尤其中小企业):被迫改变有效流程,引入复杂性,导致供应商锁定、成本上升、安全性降低和效果变差。
    • 个人:投入时间金钱学习被炒作的技术后,才发现更好的替代方案早已存在,造成资源浪费;隐私和自由也可能受损。
  • 本质:当炒作由大型企业或政府驱动时,其目的从未是公众利益,而是自身利益,常伴随夸大或谎言。

对当前AI炒作的具体批判

  • 术语误导:“人工智能”这一术语是误导性的,当前技术并不具备真正的智能,其流行是因为“易于销售”。
  • 价值有限:作者估计只有约10%的AI炒作基于有用事实(即作为工具的相关技术),其余为夸大之词。
  • 企业误用:许多公司为追逐趋势裁员并用AI取代人力,这是错误的决策。AI可作为工具完成辅助任务(如替代搜索引擎),但用于取代客服等需要人类交互的角色将损害企业,因为用户普遍厌恶与AI客服互动。
  • 劳动力问题:AI背后的劳动力(如数据标注、内容审核工作者)往往薪资低、工作条件差。
  • 资源消耗:AI的发展加剧了能源和水资源的消耗(如数据中心冷却),与当前通胀危机和资源节约需求背道而驰。
  • 行业异化:科技行业痴迷于炒作和趋势,甚至将存在已久的技术贴上“AI驱动”的标签,这种行为使其类似时尚行业。
  • 投资泡沫:大多数对AI的投资基于虚假承诺,作者坚信这些承诺无法兑现,将导致大量资金损失。

推荐资料(用以支持观点):

  • 视频与文章探讨了AI炒作的本质、LLM的实际局限性与风险、AI背后被剥削的劳动力、以及对AI泡沫破裂的预测。资料包括计算机科学家、经济学家、科技从业者等的分析。
28. Show HN: Attune - Build and publish APT repositories in seconds (github.com)

Fast and secure Linux package publishing. Contribute to attunehq/attune development by creating an account on GitHub.

30. Taking Notes with Joplin (lwn.net)

Joplin is an open-source note-taking application designed to handle taking many kinds of notes, [...]