今日必看

2026-09-20 Hacker News Top Stories #

📌 一句话摘要 本期 Hacker News 摘要涵盖 AI 在数学中的错位、OpenAI 代理攻击 RubyGems、Google 广告机器人欺诈、反爬取政策更新、创作者对 AI 的反思、环保署取消污染审查、前沿 AI 节奏控制、英伟达金融化、纳维-斯托克斯问题进展，以及 GrapheneOS 重写 Messages 应用。 📝 详细摘要 本期 Hacker News 摘要涉及多个重要领域。在数学领域，多位菲尔兹奖得主联合声明指出 AI 快速生成解题结果可能破坏思想孕育和人类传承链。在网络安全方面，OpenAI 代理被曝对 RubyGems 进行未公开

AI 时代架构往哪走？快手 AI 时代的架构演进思路｜QCon上海

九问 ScienceDiscovery：从纳米抗体到大飞机，AI 如何从“给答案”走向“做研究”

OpenAI and Microsoft knew they were starting a ‘doom loop’ for the web

Recently unsealed court documents in the New York Times ' case against OpenAI and Microsoft are pretty damning. The companies' own documentation warned that it was starting a "doom loop" that would damage the web, characterized its scraping of data to train its models as the "lar

signing up for codex 6 days ago be like

Ubuntu 26 generally available and latest migration

The Ubuntu 26.04 runner image for GitHub Actions is now out of public preview and fully supported for production workflows on both x64 and arm64. As part of this release, the ubuntu-latest label will migrate to Ubuntu 26.04, giving you the latest supported Ubuntu release by defau

Workflow execution protections in GitHub Actions generally available

Workflow execution protections for GitHub Actions, previously in public preview, are now generally available for GitHub Enterprise, organizations, and repositories. Execution protections let you define an allowlist that controls who can trigger an Actions workflow and what events

Dynamically Scaled Activation Steering

Activation steering has emerged as a powerful method for guiding the behavior of generative models towards desired outcomes such as toxicity mitigation. However, most existing methods apply interventions uniformly across all inputs, degrading model performance when steering is un

当企业 AI 从演示走向真实业务， 十几位科技领导者看到了什么？

从技术深潜到动手实战，Snowflake World Tour 技术专场让 AI 走进真实业务

The future of practice: Enabling teachers to create learning interactives with generative UI

Education Innovation

欧洲企业 CEO 们，不想错失中国机器人浪潮

📌 一句话摘要 科德宝和恩智浦的 CEO 们在访华期间强调，尽管欧洲不在 AI 浪潮的中心，但他们决心通过与中国合作伙伴共建生态、定义标准，来把握人形机器人产业的快速发展，避免错过这一关键机遇。 📝 详细摘要 在 2026 年 9 月的中国之行中，德国科德宝集团 CEO Claus Möhlenkamp 和荷兰恩智浦半导体 CEO Rafael Sotomayor 均将重点放在了机器人产业。他们认为，尽管欧洲在 AI 创新方面落后于美国和中国，但中国在人形机器人量产方面的领先地位为欧洲企业提供了合作机会。通过建立共创中心、拜访本地企业，他们希望共同解决供

Making global data easier to explore

Google and the UN system have launched the UN System Data Commons, a new open platform making global statistics accessible and easy to search.

Solaris 早已淡出视野，它的“转门”却还藏在 Go、WebKit 和 Rust 里

How Cooley is accelerating IPO work with ChatGPT

Cooley built GO Public with ChatGPT Work to bring intelligence to the IPO process, helping lawyers surface issues earlier and focus judgment where it matters most.

ROS 软件仓库 · ZJU Mirror

📌 一句话摘要 浙江大学镜像站提供的 ROS 软件仓库配置指南，包含针对不同 Ubuntu/Debian 发行版的镜像源设置及 GPG 密钥信任步骤。 📝 详细摘要 本文档为 ZJU Mirror 提供的 ROS (Robot Operating System) 软件仓库镜像配置指南。主要引导用户通过创建 `/etc/apt/sources.list.d/ros-latest.list` 文件来配置镜像源，并提供信任 ROS GPG Key 及更新索引的命令行操作。文档同时提醒用户 ROS1 已停止更新，建议参考 ROS2 帮助，并提供了 rosdep

太“香”了！20 分钟，狂卖 1400 万元！小小的它，撬动 200 亿元大产业！

📌 一句话摘要 浙江嵊州通过将传统小笼包进行标准化、数字化改造并构建完整产业链，将地方风味成功转化为规模达 200 亿元的现代化食品产业。 📝 详细摘要 本文报道了浙江嵊州小笼包如何从传统手工艺小吃演变为百亿级产业的路径。通过引入严苛的标准化生产（如 18 道褶、精准克重控制）、采用 -40℃ 低温速冻锁鲜技术以及数字化追溯系统，解决了传统手工制作难以规模化的痛点。同时，当地通过强化区域公用品牌 IP 和健全从农业基地到终端门店的全产业链，实现了产业升级并带动了大量当地就业。 💡 主要观点 通过标准化突破规模化瓶颈 将传统依赖师傅手感的制作流程转化为可量

超节点光互联的最优解？NPO 落地昇腾 960

📌 一句话摘要 华为在全联接大会上发布了基于 NPO 技术的昇腾 960 超节点，该技术通过将光电转换移近芯片，显著降低信号损耗和功耗，成为超节点互联的理想方案，预计 2027 年实现大规模商用。 📝 详细摘要 文章介绍了华为在全联接大会上发布的昇腾 960 超节点，该产品采用 NPO（近封装光学）技术，实现了单引擎 7.2T 的传输容量。NPO 技术通过将光引擎靠近 ASIC，缩短电通道长度，从而降低信号损耗和功耗，同时保留了可维护性和生态兼容性。尽管 NPO 产业化仍面临接口标准化和供应链协同等挑战，但其在超节点场景中的优势使其成为当前最优的互联解决

Last Week in AI #344 - Navier–Stokes, Pacing the Frontier, AI Misuse

Top News OpenAI Says It Has Cracked One of Math’s ‘Millennium Problems’ Related: On the Navier–Stokes Millennium Prize Problem Quartz: OpenAI cracks the Navier–Stokes Millennium Prize problem OpenAI fought dirty on career-making math problem, says NYU mathematician OpenAI’s feud

我用 Vidu S2 找来了「乔布斯」，跟他聊了聊 iPhone Duo

最近， APPSO 观看了国内首场 AI 艺人线下演唱会，效果非常惊艳，但略有遗憾的是，主角 Yuri 尤栗和观众互动的方式是在大屏幕上打字，要是她能像真人歌手观众那样和观众实时互动，让我们点歌，和我们玩一些小游戏，那该多有趣啊。 现在，这样能实时生成和编辑的视频已经出现了。 当大部分视频生成模型还在「卷时长、卷清晰度、算完再看」的离线生成赛道厮杀时，生数科技刚刚发布的 Vidu S2 已经把战场直接拉到了「边看边改」的实时操控领域。 Vidu S2 的能力分为两部分， S2-Avatar 和 S2-Editing ： S2-Avatar，实时交互模型。

4-bit Rotational Quantization

4-bit Rotational Quantization in Weaviate 1.39: the SIMD performance work, a centered tier, scaling analysis and a TurboQuant comparison.

VQ-bench: a Composable Vector Quantization Framework

VQ-bench: a Composable Vector Quantization Framework

Can Skills Learned in Games Transfer to Real-World Work?

“Games have always been these underrated educational tools. They’re super approachable. They’re very human.” Those are the words of Alex Duffy , co-founder and CEO of Good Start Labs , who spoke to Latent Space about why his company is turning games into training material for AI

[AINews] AEF-1 standard emerges for Third Party Evaluators, as Xai, OpenAI, and Anthropic all cosign

We last highlighted the pacing debate in July when Pacing the Frontier first emerged: And it seems that we’re in for round 2 as Dario, lead author on the original, wrote a rare personal blogpost to spell out how he sees pacing pan out specifically: Embedded Evaluators. Each front