维护现有 .name 域名注册及电子邮件重定向的斗争已以重大胜利告终。 Verisign 已正式撤回其备受争议的计划，该计划本打算终止 22,000 个域名。尽管该公司将不再接受新的 3rd-level .name 域名注册，但这一决定确保所有现有用户——包括本网站的所有者——都能无限期地保留自己的数字存在。

这一成果来自一支专注团队的共同努力，也得到了众多支持者的推动。行动起初由 Doytchin 发起，并在 Jim 前往 Bali 直接向 ICANN 陈述理由后迅速获得强劲进展。法律辩护由律师 Jan 、 Flip 和 Diégo 主导，他们对 ICANN 采取了强硬措施，以保护域名持有者的利益。

在核心组织者和法律团队之外，Patrik 、 Troy 、 Huyen 、 Keith 、 Judy 、 Erik 和 Ryan 提出的正式重新考虑请求也进一步增强了这项事业。其他数十人则提供了关键的支持信，汇聚成一股难以忽视的合力。

这项努力同样受益于来自内部的支持，因为 ICANN 内部有些人敢于挑战该组织的官僚体系。回顾这场行动时，作者强调参与者的高水平，认为这段经历之所以意义非凡，正源于所有为此付出的人展现出的卓越奉献精神与正直。 The battle to preserve existing .name domain registrations and email redirects has ended in a major victory. Verisign has officially retracted its controversial plan that threatened to terminate 22,000 domains. While the company will no longer accept new registrations for 3rd-level .name domains, this decision ensures that all current users, including the owner of this website, can maintain their digital presence indefinitely.

This success was the result of a dedicated team effort that involved a wide range of supporters. The campaign began with an initial push by Doytchin and gained significant momentum when Jim traveled to Bali to present the case directly to ICANN. The legal defense was spearheaded by attorneys Jan, Flip, and Diégo, who took aggressive action against ICANN to protect the interests of domain holders.

Beyond the core organizers and legal team, the cause was bolstered by formal reconsideration requests from Patrik, Troy, Huyen, Keith, Judy, Erik, and Ryan. Dozens of others provided essential letters of support, creating a unified front that made the movement difficult to ignore.

The effort also benefited from internal support, as several individuals within ICANN had the courage to challenge the organization's bureaucracy. Reflecting on the campaign, the author highlights the high caliber of the people involved, noting that the experience was defined by the remarkable dedication and integrity of everyone who contributed to the cause.

Blender 5.3 引入了一套重要的全新功能，其中以对 3D Gaussian Splats 的原生支持为核心。用户现在可以从 PLY 、 SPZ 和 glTF 文件导入这些场景，允许将它们作为点云进行操作，通过 Geometry Nodes 进行处理，并在 EEVEE 、 Cycles 和 Workbench 中进行渲染。为促进这一功能，软件包含了用于控制 splat 旋转、缩放和辐射率的新属性，并初步支持使用标准 BSDF 材质对 splats 进行重新打光。

Cycles 的渲染能力得到了扩展，在 Principled BSDF 中加入了光色散，使用户能够在玻璃和宝石中创造逼真的彩虹效果。此外，集成 NVIDIA DLSS 4.5 Ray Reconstruction 提供了强大的新降噪器，在视口中实现接近最终渲染的效果，同时显著减少动画播放过程中的闪烁。 EEVEE 也获得了重大更新，包括新的材质照明节点，赋予艺术家对材质如何响应场景光线的细粒度控制，并支持多种全景相机投影方式，如鱼眼镜头。

工作流程管理方面引入了基于项目的组织方式，允许用户整合相关文件、自定义变量，并为每个项目配套资产库。资产管理进一步改进为能够将资产库直接作为扩展进行共享和安装，可从 Poly Haven 等公共领域来源获取，而无需额外的插件。与此同时，Video Sequencer 也进行了大幅重构，引入了完整的涟漪编辑框架、重写的 blade 工具，并提升了性能，使编辑时保持时间线连续性变得更容易。

用户界面也进行了贴心的优化，例如 Outliner 中支持自定义排序，以及在所有编辑器中搜索侧边栏的能力。界面中集成了 Markdown 支持，尤其体现在全新的 Comment 节点里，帮助更清晰地标注节点树。 Grease Pencil 用户则可以使用新的选择工具、一个强大的 Carver 工具（通过布尔减法在笔触和填充中创建切口），以及同时填充多个区域的能力，从而显著简化动画与绘制流程。 Blender 5.3 introduces a significant suite of new features, headlined by native support for 3D Gaussian Splats. Users can now import these scenes from PLY, SPZ, and glTF files, allowing them to be manipulated as point clouds, processed through Geometry Nodes, and rendered in EEVEE, Cycles, and Workbench. To facilitate this, the software includes new attributes for controlling splat rotation, scale, and radiance, with rudimentary support for relighting splats using standard BSDF materials.

The rendering capabilities in Cycles have been expanded with the addition of light dispersion within the Principled BSDF, enabling users to create realistic rainbow effects in glass and gemstones. Additionally, the integration of NVIDIA DLSS 4.5 Ray Reconstruction provides a powerful new denoiser, offering near-final render quality in the viewport while significantly reducing flickering during animation playback. EEVEE also receives major updates, including new material lighting nodes that grant artists granular control over how materials react to scene lights, alongside support for various panoramic camera projections like fisheye lenses.

Workflow management is addressed with the introduction of project-based organization, allowing users to consolidate related files, custom variables, and per-project asset libraries. Asset management is further improved by the ability to share and install asset libraries directly as extensions, pulling from public domain sources like Poly Haven without needing extra add-ons. Meanwhile, the Video Sequencer has undergone a substantial overhaul, introducing a comprehensive ripple editing framework, a rewritten blade tool, and improved performance, making it much easier to maintain timeline continuity during edits.

The user interface also sees thoughtful refinements, such as custom sorting in the Outliner and the ability to search sidebars across all editors. Markdown support has been integrated into the interface, notably in the new Comment node, which helps in annotating node trees clearly. Grease Pencil users benefit from new selection tools, a powerful Carver tool that uses boolean subtraction to create cuts in strokes and fills, and the ability to fill multiple regions simultaneously, streamlining the animation and drawing process significantly.

Google 已正式发布 Gemini 4 Carbon，这是其迄今为止最先进的前沿模型。虽然之前发布的 Gemini 4 Argon 仍然是通用软件工程和常规知识任务的最佳选择，但 Carbon 是专为高风险智能任务打造的。它在复杂的多步骤编程、调试以及开放式科学探索方面表现出色。通过使用更多 tokens，它能够进行更深层的推理，并运行更长的验证循环，使其成为推动现有 AI 能力边界的首选工具。

该模型在高精度知识工作上展现出强大实力，尤其在法律、金融和企业场景中。它具备先进的视觉理解与检索能力，能够从蓝图、示意图、财务报告等密集的多模态文档中整合信息并完成综合分析。与 Google Antigravity 的新 swarm 功能结合后，它还能让用户协调多智能体团队处理复杂任务，例如从单一 prompt 生成三维建筑重建，或构建交互式、由数据驱动的应用程序。

在软件工程与网络安全领域，Gemini 4 Carbon 在漏洞发现与修复方面带来了显著提升。它在关键安全基准测试中表现稳定，能够识别缺陷并生成经过验证的修复补丁。早期测试者与内部团队已用该模型把数周的人工安全研究压缩到短短数小时，发现 0-day 漏洞，并在自动化扫描与分诊中取得了很高的成功率。它在将多个漏洞串联成攻击场景的同时仍能保持高精度，使其成为网络安全防御方的重要力量。

该模型在科学与数学推理方面同样取得了突破性进展。通过与多智能体框架整合，Carbon 在形式化验证方面发挥了关键作用，例如证明密码协议的安全属性；同时还能通过识别实验数据中先前遗漏的异常现象，加速药物毒性研究。它还能在海量数据集中发现远缘生物同源物，从而提升蛋白质结构预测能力，证明它能够成为攻克基础科学难题研究者的关键伙伴。

安全依然是此次发布的核心考量。 Google 针对模型能力增强实施了与之相匹配的强有力缓解措施。 Gemini 4 Carbon 在抵御 prompt injection 以及维持一致性方面表现更佳，内部与外部测试均确认其在阻断恶意攻击路径方面具备较高成功率。该模型现已通过 Google AI Studio 和 Gemini API 向开发者与企业提供，并为科学家与网络安全专业人士开放专门权限，标志着在满足真实生产需求方面实现前沿级性能的新里程碑。 Google has officially released Gemini 4 Carbon, its most advanced frontier model to date. While the previously released Gemini 4 Argon remains the optimal choice for general software engineering and standard knowledge tasks, Carbon is specifically engineered for high-stakes intelligence. It excels in complex, multi-step coding, debugging, and open-ended scientific discovery. By utilizing more tokens, it is capable of deeper reasoning and running extended verification loops, positioning it as the primary tool for projects that push the limits of current AI capabilities.

The model demonstrates significant prowess in high-precision knowledge work, particularly within legal, financial, and enterprise environments. It features advanced visual understanding and retrieval capabilities, allowing it to synthesize information from dense, multi-modal documents like blueprints, schematics, and financial reports. When combined with Google Antigravity's new swarm feature, it enables users to coordinate teams of agents that can manage complex tasks, such as creating 3D architectural reconstructions or building interactive data-driven applications from a single prompt.

In the realm of software engineering and cybersecurity, Gemini 4 Carbon introduces substantial improvements in vulnerability discovery and remediation. It consistently performs well on key security benchmarks, identifying flaws and generating verified patches. Early testers and internal teams have utilized the model to compress weeks of manual security research into mere hours, uncovering 0-day vulnerabilities and achieving high success rates in automated scanning and triage. Its ability to maintain precision while chaining multiple vulnerabilities into attack scenarios makes it a powerful asset for defenders in the cybersecurity landscape.

The model also showcases significant breakthroughs in scientific and mathematical reasoning. By integrating with multi-agent frameworks, Carbon has been instrumental in formal verification, such as proving the security properties of cryptographic protocols, and accelerating drug toxicity research by identifying previously missed anomalies in experimental data. Its capacity to discover remote biological homologs across vast datasets is also enhancing protein structure prediction, proving that it can serve as a vital partner for researchers tackling foundational scientific problems.

Safety remains a core focus of the release, with Google implementing robust mitigations proportionate to the model's increased capabilities. Gemini 4 Carbon shows improved performance in resisting prompt injection and maintaining alignment, with internal and external testing confirming high success rates in blocking malicious attack trajectories. The model is now available to developers and enterprises through Google AI Studio and the Gemini API, with specialized access for scientists and cybersecurity professionals, marking a new milestone in providing frontier-level performance for real-world production demands.

可靠智能家居的基础哲学是：它必须在完全不依赖厂商服务器或畅通的互联网连接的情况下正常运行。十多年来，这一主要原则始终不变：照明和安防等核心家中运作绝不应仅因为外部网络中断就失效。云服务可以为了方便或作为额外的接口而使用，但它们绝不能成为单点故障。这种本地优先的思路能让用户免受常见的当代烦恼，例如损坏的接口、证书错误以及被停止支持的产品。

要维持一个有韧性的家庭环境，需要遵循一些切实可行的准则。在加入新的自动化时，关键在于保留传统的手动控制，因为当智能功能失效时，它们能提供必要的冗余。系统设计要具备灵活性，认识到居住空间和设备需求会随着时间变化。此外，用户应当整合配置，而不是依赖零散的厂商应用；任何带进家里的设备都必须提供可访问的接口，并有清晰的日志，避免陷入专有"黑盒"带来的风险。

这一愿景的技术落地，通常会借助诸如基于 ESP 硬件的 Tasmota 、用于发布 / 订阅消息的 MQTT 代理，以及在本地树莓派上用 Node-RED 管理复杂逻辑等灵活的开源工具。通过使用 MQTT，设备之间保持解耦，使多个系统能够在不依赖复杂的一对一连接的前提下对传感器数据作出响应。这种模块化让整个搭建更易调试、也更容易扩展，无论是监测喷灌系统、自动化车库门，还是管理气候控制。

随着家庭逐步增加更多居住者和设备，更稳健的交互方式的重要性就愈发明显。仅依赖应用或语音助手是不够的，因此需要配合实体的墙面面板、 NFC 标签和二维码，确保每位住户都有可靠的方式与自己的环境互动。此外，选择电池供电设备时要格外谨慎，确保使用寿命更长。因为频繁更换传感器带来的维护负担是一种常见的坑，往往会引发挫败感，甚至导致系统被闲置。

除了个人经历，技术爱好者中还形成了一种更广泛的共识：那些依赖互联网的设备往往把厂商的商业模式放在用户价值之上。许多消费者反馈，硬件一旦在服务器离线时就会变得无法使用，比如智能厨房电器或家庭安防组件。综合建议是：把第三方依赖视为固有风险，并优先选择支持本地控制或固件灵活性的设备。归根结底，设计智能家居的目标，是打造一个无论外部企业的稳定性如何，都能保持可用，并始终由房主掌控的环境。 The foundational philosophy of a reliable smart home is that it must function entirely without reliance on a vendor's server or an active internet connection. For over a decade, the primary rule has remained constant: core household operations, such as lighting and security, should never fail simply because the external network goes down. While cloud services can be utilized for convenience or as an additional interface, they should never serve as a single point of failure. This local-first approach protects users from common modern frustrations, such as broken APIs, certificate errors, and abandoned product support.

Maintaining a resilient home environment requires adhering to several practical guidelines. When adding new automation, it is crucial to ensure that traditional manual controls remain intact, as they provide necessary redundancy when smart features fail. Systems should be designed for agility, recognizing that living spaces and device requirements change over time. Furthermore, users should consolidate configurations rather than relying on disparate vendor apps, and any device brought into the house must have an accessible API with clear logs to avoid the hazards of proprietary black boxes.

The technical implementation of this vision often involves leveraging flexible, open-source tools like Tasmota on ESP hardware, an MQTT broker for pub/sub messaging, and Node-RED for managing complex logic on a local Raspberry Pi. By using MQTT, devices remain decoupled, allowing multiple systems to react to sensor data without complex point-to-point dependencies. This modularity makes the entire setup easier to debug and scale, whether it involves monitoring a sprinkler system, automating garage doors, or managing climate controls.

As the home evolves to include more occupants and devices, the importance of robust interaction methods becomes clear. Relying solely on apps or voice assistants is insufficient, so incorporating physical wall dashboards, NFC tags, and QR codes ensures that every resident has a reliable way to interact with their environment. Additionally, battery-powered devices should be chosen carefully to ensure long life, as the maintenance burden of frequently swapping sensors is a common pitfall that can lead to frustration and system disuse.

Beyond personal experience, there is a broader consensus among technology enthusiasts that internet-dependent devices often prioritize vendor business models over user utility. Many consumers report significant issues with hardware that becomes unusable when servers go offline, such as smart kitchen appliances or home security components. The collective advice suggests treating third-party dependencies as inherent risks and prioritizing devices that offer local control or firmware flexibility. Ultimately, designing a smart home is about creating an environment that remains functional and under the owner's control, regardless of external corporate stability.

Ableton Live 13 推出了一套全新的工具和工作流程增强功能，旨在简化音乐制作、声音设计和混音。其中最引人注目的功能包括一套全新的内置 Step Sequencer，它让你可以直接在 Clip 内创建复杂的节拍、旋律以及多节奏结构。此外，软件现在支持 ARA 2，可与更先进的音频编辑插件实现更顺畅的集成，并新增 Capture Automation 功能，让用户把即兴演奏中的临时调整转化为永久的自动化曲线。

为进一步打磨创作流程，Live 13 还加入了多项专门的声音设计工具。用户现在可以使用一款配备超过 20 种引擎的专用 Drum Synth 来生成自定义打击乐；全新的 Looplet 工具则支持通过四个独立播放头，对同一个 Sample 的不同段落进行分层和操控。在混音细化方面，本次更新提供了用于对起音与延音进行精准"外科手术"式控制的 Transient Shaper，用于处理峰值或添加谐波饱和度的全新 Clipper，以及支持对 Return Tracks 上的效果器与混音器设置进行调制的 Envelope Clips 。

混音环境同样进行了大幅升级，以提升清晰度与效率。界面现在在混音器中直接加入 Device Slots，帮助更好地概览各轨道的设备效果；同时新增 Multiband 与 Mid-Side Rack 类型，用于塑造立体声宽度与音色。此外，Core Library 进一步扩充了可直接用于制作的预设，用户还可获得五个由 Softube 开发并模拟专业录音室硬件的全新效果器。与此同时，Racks 的控制也更细致：现在最多可支持 32 个 Macros，并提供自定义映射曲线。

硬件集成方面，Push 依旧是关键组成部分。 Push 3 新增 Custom Layouts，把 Session Views 、混音器控制和音符区域整合到同一个页面中。 Push 用户还可以使用全新的性能混音器以及更先进的设备模式导航；而 Push 2 则支持新的 step sequencer，并带来多项设备模式增强。整体来看，这些更新都聚焦于让想法在更灵敏、更一体化的环境中持续推进。

Ableton Live 13 预计将于 2027 年初发布。对于计划切换的用户，官方目前推出促销活动：购买或升级到 Live 12 的用户，可在 Live 13 发布后免费获得对应版本的 Live 13 。软件将继续提供 Intro 、 Standard 与 Suite 三个版本，其中 Suite 仍是最完整的套装，包含 Max for Live 以及全部乐器与效果器内容。 Ableton Live 13 introduces a suite of new tools and workflow enhancements designed to streamline music production, sound design, and mixing. Among the standout features is a new built-in Step Sequencer that allows for the creation of intricate beats, melodies, and polyrhythms directly within clips. Additionally, the software now supports ARA 2, which enables more fluid integration with advanced audio-editing plug-ins, alongside a new Capture Automation feature that lets users turn spontaneous performance tweaks into permanent automation curves.

Refining the creative process further, Live 13 adds several specialized sound design utilities. Users can now utilize a dedicated Drum Synth with over 20 engines to generate custom percussion, while the new Looplet tool offers the ability to layer and manipulate various sections of a single sample using four independent playheads. For mix refinement, the update includes a Transient Shaper for surgical control over attack and sustain, a new Clipper for managing peaks or adding harmonic saturation, and Envelope Clips that allow for modulation of effects and mixer settings on return tracks.

The mixing environment has also received significant updates to improve clarity and efficiency. The interface now features Device Slots directly in the mixer for a better overview of track effects, along with new Multiband and Mid-Side Rack types to assist in shaping stereo width and tone. Furthermore, the Core Library has been expanded with production-ready presets, and users gain access to five new Softube effects modeled after professional studio hardware. These improvements are complemented by more granular control over Racks, which now support up to 32 Macros and custom mapping curves.

Hardware integration for Push remains a central component, with Push 3 receiving Custom Layouts that combine session views, mixer controls, and note areas onto a single page. Push users also benefit from a new performance mixer and advanced device mode navigation, while Push 2 supports the new step sequencer and various device mode enhancements. These updates collectively focus on keeping ideas moving through a more responsive and integrated environment.

Ableton Live 13 is slated for release in early 2027. For those looking to transition, the company is currently offering a promotion where purchasing or upgrading to Live 12 grants users the corresponding edition of Live 13 for free upon its release. The software will continue to be available in Intro, Standard, and Suite editions, with Suite maintaining its position as the most comprehensive package, including Max for Live and the full collection of instruments and effects.

.NET 11 runtime 引入了旨在增强性能、简化代码维护并优化开发者体验的重要更新。其中一项主要变化是提高 x86/x64 和 Arm64 架构的最低硬件要求：现在它们需要更现代的指令集。这种调整通过放弃对旧版硬件的支持来精简代码库，从而实现与当前操作系统标准一致的更优化代码生成。

预览阶段的一项重点功能是 Runtime Async，它将模型从编译器生成的状态机转向由 runtime 管理。该更新通过取消对编译器输出基础设施的依赖，带来更清晰的实时堆栈跟踪、更好的可调试性，并降低开销。 runtime 现在能够原生处理异步挂起与恢复，并借助缓存延续以及延续可选择不捕获不必要的 ExecutionContext 等能力进一步提升性能，从而减少高吞吐量异步代码中的开销。

Just-In-Time（JIT）编译器同样获得了显著改进，从更积极的边界检查消除到对泛型虚拟方法的更好去虚拟化。这些增强再加上用于 SIMD 通道构建与组合的新硬件指令，使执行效率更高。 runtime 现在还可以在不依赖平台特定代码的情况下，对向量元素布局进行更精细的控制，这对图像处理和数字信号处理工作负载尤为有利。

基础设施和诊断方面也新增了多项关键能力，包括一种新的进程内崩溃报告机制，可在进程终止前捕获关键状态信息。通过允许 runtime 在逻辑处理器超过 1024 的系统上初始化，高核心数服务器的支持得到改进。另外，NativeAOT 通过为接口方法调用提供共享分发辅助程序来提升性能，减少二进制体积并提高吞吐量。

最后，这次更新还针对特定平台和场景带来了明显优化。 WebAssembly 支持现在提供更便捷的 WebCIL 有效载荷加载，并提升了互操作性能；iOS 和 tvOS 上的文件锁定行为也已更新，避免应用在后台挂起期间被终止。与此同时，ReadyToRun（R2R）镜像现在配备了专门的默认比较器，以加速集合操作。再加上多项可靠性修复，确保 .NET 11 为现代应用程序开发提供更坚实、更高效的基础。 The .NET 11 runtime introduces significant updates aimed at enhancing performance, simplifying code maintenance, and refining the developer experience. One of the primary changes is an increase in minimum hardware requirements for x86/x64 and Arm64 architectures, which now demand more modern instruction sets. This shift streamlines the codebase by moving away from legacy hardware support, allowing for more optimized code generation that aligns with current OS standards.

A major feature arriving in preview is Runtime Async, which transitions from compiler-generated state machines to a runtime-managed model. This update results in cleaner live stack traces, improved debuggability, and lower overhead by eliminating the need for compiler-emitted infrastructure. The runtime now handles asynchronous suspension and resumption natively, and performance is further boosted by features like cached continuations and the ability for continuations to opt out of unnecessary ExecutionContext capture, which reduces overhead in high-throughput async code.

The Just-In-Time (JIT) compiler has also received substantial improvements, ranging from more aggressive bounds check elimination to better devirtualization of generic virtual methods. These enhancements, coupled with new hardware intrinsics for SIMD lane construction and composition, allow for highly efficient execution. The runtime now provides finer control over vector element layout without relying on platform-specific code, which is particularly beneficial for image processing and digital signal processing workloads.

Infrastructure and diagnostics have been bolstered through several key additions, including a new in-process crash reporting mechanism that captures vital state information before a process terminates. Support for high-core-count servers has been improved by allowing the runtime to initialize on systems with more than 1024 logical processors. Additionally, NativeAOT has seen performance gains through shared dispatch helpers for interface method calls, reducing binary sizes and improving throughput.

Finally, the update brings notable refinements for specific platforms and scenarios. WebAssembly support now includes easier loading of WebCIL payloads and improved interop performance. Meanwhile, file locking behaviors on iOS and tvOS have been updated to prevent app termination during background suspension, and ReadyToRun (R2R) images now feature specialized default comparers to speed up collection operations. These changes, alongside various reliability fixes, ensure that .NET 11 provides a more robust and efficient foundation for modern application development.

Google DeepMind 在软件安全领域取得了重大里程碑，成功利用人工智能对 Signal 消息应用程序的核心加密协议进行了形式化验证。这一成就通过一个名为 AlphaProof Code 的新型代理框架实现，标志着在证明代码完全按预期运行且不存在漏洞方面迈出了巨大的一步。过去，形式化验证是一项极其耗费劳动力的任务，但借助这种新方法，研究人员仅用四周时间就完成了对 Signal 广泛部署实现的验证。

AlphaProof Code 框架通过类似专业研究团队的方式来分担验证工作量。人类专家与 AI 合作建立所需的安全规范，而一系列由 Gemini Carbon 模型驱动的 AI 代理则负责创建数学证明这项艰巨工作。该系统生成了超过 600 万行 Lean（为机器校验证明而设计的语言），以确保软件的准确性。通过自动化这一流程，框架让人类专家把注意力放在代码库中只占一小部分、但需要更细致判断的关键环节上，而 Lean 系统则对其余复杂内容进行严格校验。

这一突破尤其引人关注，因为它弥合了协议设计与实际实现之间的难题。尽管许多验证方法停留在高层设计层面，AlphaProof Code 进一步确认：在设备上运行的已编译机器代码始终遵循关键安全属性。团队还通过将已知的历史漏洞重新植入代码来对系统进行再验证，证明 AI 能够成功识别这些版本的问题，并证明其不安全。

随着这项技术走向可扩展应用，未来数字安全的意义也将更为深远：它或将把形式化验证从小众的学术研究，变成保护关键基础设施的可落地方案。凭借数学层面的安全保证，这些方法最终有望应用于保护多类系统，从医疗设备固件到电网软件。与此同时，网络威胁日益复杂，而能够将代码快速核验到具体安全规范之上的能力，将为支撑全球通信的技术提供一条更强有力的新防线。

为推动社区在这一方向开展更深入的探索与开发，Google DeepMind 已将其形式化验证制品开源。该举措旨在促进协作与研究，推动这类由 AI 驱动的安全保障方案在全行业软件加固中的落地实践。通过把自动定理证明与主动防御代理相结合，团队希望在技术快速变革的时代，为构建更具韧性、更安全的系统确立新的标准。 Google DeepMind has reached a significant milestone in software security by successfully utilizing artificial intelligence to formally verify the core cryptographic protocol of the Signal messaging application. This accomplishment, managed through a new agentic framework called AlphaProof Code, represents a major leap forward in proving that code operates exactly as intended without vulnerabilities. Historically, formal verification has been an extremely labor-intensive task, yet this new approach allowed researchers to complete the verification of Signal's widely deployed implementation in just four weeks.

The AlphaProof Code framework functions by splitting the verification workload similarly to a dedicated research team. Human experts collaborate with AI to establish the necessary security specifications, while a series of AI agents, powered by the Gemini Carbon model, undertake the arduous task of creating mathematical proofs. The system generated over six million lines of Lean, a language designed for machine-checked proofs, to ensure the accuracy of the software. By automating this process, the framework allows human experts to focus their attention on the critical one percent of the codebase that requires nuanced judgment, while the Lean system validates the remaining complexity.

This breakthrough is particularly noteworthy because it addresses the difficult gap between the design of a protocol and its actual implementation. While many verification methods stop at the high-level design, AlphaProof Code confirms that the compiled machine code, which runs on devices, consistently adheres to vital security properties. The team further validated the system by reintroducing known historical bugs into the code, demonstrating that the AI could successfully identify and prove these versions were insecure.

The implications for the future of digital security are profound, as this technology moves formal verification from a niche academic pursuit to a scalable solution for critical infrastructure. By providing a mathematical guarantee of security, these techniques could eventually be applied to protect diverse systems ranging from medical device firmware to power grid software. As cyber threats become increasingly sophisticated, the ability to rapidly verify code against security specifications provides a powerful new line of defense for the technologies that underpin global communications.

To encourage further community exploration and development in this space, Google DeepMind has made its formal verification artifacts open-source. This initiative aims to foster collaboration and research into how such AI-driven security assurance can be deployed to harden software across the industry. By combining automated theorem proving with proactive defense agents, the team hopes to establish a new standard for building resilient, secure systems in an era of rapid technological change.

AI 和 LLMs 在软件开发流程中的快速融合，已经显著改变了代码的生成方式，但这种依赖往往以牺牲长期可持续性为代价。 AI 在其上下文窗口内擅长解决局部问题，却在很大程度上忽视了大型 SaaS 应用所需要的更广泛架构要求。这种失衡会催生脆弱的代码库，使其难以承受高可用与高扩展性的严苛挑战。

导致这种脆弱性的主要有四个问题。第一，AI 往往生成"浅层测试"，把注意力放在琐碎方法上，而不是关键的用户路径；这些覆盖缺口不会被常见指标捕捉到。第二，AI 经常产出过于复杂的代码：死板地套用 DRY 这类模式，却不顾可读性，结果就是系统对人类而言难以排查与调试。第三，当开发人员越来越依赖自动化时，他们对底层数据结构与算法的理解也会逐步减弱。最后，AI 生成代码的速度极快，反而让开发团队难以维持高质量门槛，进而不断累积技术债，最终把原本稳定的项目变成一团难以理清的 spaghetti code 。

这些做法会在一些可见的信号上逐渐显现，最突出的是生产事故的增多。当代码库由 agents 而不是人类架构师来构建时，系统失效的频率往往会上升。而一旦事故发生，由于人类团队失去了对代码的深入熟悉，他们就更难理解和穿越 AI 生成结果所留下的复杂性，因此修复和缓解会耗时更久。与此同时，系统中还会滋生一些对开发人员并不明显、但对用户却立刻可感的错误，最终拉低整体产品体验。

归根结底，主要由 AI 构建的软件会变成面向机器而非面向人的软件。缺少了人类视角，就难以预判真实用户会如何与产品交互。自动化或许足以支撑那些只靠价格竞争的产品，但高质量的 SaaS 仍需要人类开发者所具备的细腻判断、直觉与用心。面向未来，更有价值的很可能是保留这种人类触感的软件：让系统更具韧性、更易维护，并且真正贴合使用它的人的需求。 The rapid integration of AI and LLMs into the software development process has created a significant shift in how code is produced, but this reliance often comes at the expense of long-term sustainability. While AI is highly effective at solving localized problems within its context window, it remains largely oblivious to the broader, architectural demands of large-scale SaaS applications. This imbalance fosters fragile codebases that are ill-equipped to handle the rigors of high availability and scalability.

Four primary issues contribute to this fragility. First, AI tends to generate shallow tests that focus on trivial methods rather than critical user pathways, leaving significant gaps in coverage that are not captured by standard metrics. Second, AI frequently produces overly complex code that adheres rigidly to patterns like DRY without regard for readability, making the system difficult for humans to debug. Third, as developers lean more heavily on automation, their own understanding of the underlying data structures and algorithms diminishes. Finally, the sheer speed at which AI produces code prevents developers from maintaining a high quality bar, leading to an accumulation of technical debt that transforms stable projects into disorganized spaghetti code.

The impact of these practices manifests in observable symptoms, most notably an increase in production incidents. When a codebase is built by agents rather than human architects, the frequency of system failures often rises. Furthermore, when these incidents do occur, they take longer to mitigate because the human team, having lost its deep familiarity with the code, struggles to navigate the complexity left behind by AI generators. This environment also breeds errors that remain hidden from developers but obvious to users, ultimately resulting in a degraded product experience.

Ultimately, software built primarily by AI becomes software designed for machines rather than humans. This disconnect removes the essential human perspective needed to anticipate how real people interact with a product. While automated development might be sufficient for products that compete solely on price, high-quality SaaS offerings require the nuance, intuition, and care that only human developers can provide. Moving forward, the most valuable software will likely be that which preserves this human touch, ensuring systems remain resilient, maintainable, and tailored to the needs of the people using them.

亚马逊工程师最近通过用一种名为 Strands Decider 的专用决策模型取代 Claude Sonnet 4.6 大语言模型，为其内部票务系统实现了显著的成本节约。尽管 Sonnet 4.6 表现不错，但切换到 Strands Decider 后，运营成本下降了 98%，公司有望每年节省数百万美元。这次变化反映出一种战略转向：在高频、结构化任务中，优先使用更精简、为特定场景打造的模型。

该票务系统需要处理大量输入，包括客户升级诉求、安全发现以及容量请求，并将它们归入 20 个不同的类别。与生成对话文本的传统 LLM 不同，决策模型会接收输入状态，并对预定义选项集合返回概率分数。借助 Strands Decider E4B 模型，团队成功用更高效的决策流程替代了冗长的基于 JSON 的 LLM 提示词，从而显著减少了所需的 Token 数量。

为确保平稳切换，团队通过 AWS AppConfig 实施了三向配置开关，能够实时对比两种模型效果。他们采用双模式监控流量：在新系统全面上线前，Sonnet 和 Decider 并行运行以验证性能。该方式比离线评估可靠得多，因为它让团队能够基于真实生产规模数据分析差异。

此次切换的效果十分突出：每个工单的平均延迟下降了 78% 。通过优化输入提示词，并将输出精简为单一选择而不是完整的 JSON 响应，每个工单的 Token 总数减少了 70% 。此外，团队还发现：在请求中对 JSON 键进行排序能够提升一致性，他们注意到键的顺序此前会影响模型输出的概率。

尽管结果令人满意，工程师们仍强调决策模型更适用于特定场景，例如类别集合稳定的高并发分类任务。他们建议，如果某项任务需要较长解释，或类别目录经常变化，传统 LLM 可能仍然更合适。最终，这次迁移的成功表明：在大型企业环境中为代理部署专用决策模型，能够带来可观的经济收益。 Amazon engineers recently achieved significant cost savings by replacing the Claude Sonnet 4.6 large language model with a specialized decision model known as Strands Decider for their internal ticketing system. While Sonnet 4.6 performed effectively, the transition to Strands Decider reduced operational costs by 98 percent, potentially saving the company millions of dollars annually. The change highlights a strategic shift toward using leaner, purpose-built models for high-volume, structured tasks.

The ticketing system handles a vast array of inputs, including customer escalations, security findings, and capacity requests, which must be sorted into 20 distinct categories. Unlike traditional LLMs that generate conversational text, a decision model accepts input state and returns probability scores for a predefined set of options. By leveraging the Strands Decider E4B model, the team successfully replaced lengthy JSON-based LLM prompts with a more efficient decision-making process that requires significantly fewer tokens.

To ensure a safe transition, the team implemented a three-way configuration switch through AWS AppConfig, allowing them to compare models in real-time. They monitored traffic in a dual-mode, running both Sonnet and the Decider in parallel to validate performance before fully committing to the new system. This method proved far more reliable than offline evaluations, as it allowed the team to analyze discrepancies in live, production-scale data.

The performance metrics of this switch were notable, with average latency per ticket dropping by 78 percent. By optimizing the input prompts and streamlining the output to a single selection rather than a full JSON response, the total token count per ticket decreased by 70 percent. Additionally, the team discovered that sorting JSON keys during requests improved consistency, noting that key order had previously influenced model output probabilities.

While the results were impressive, the engineers emphasized that decision models are best suited for specific use cases, such as high-volume classification tasks with a stable set of categories. They advised that if a task requires lengthy explanations or frequent changes to the category catalog, traditional LLMs might still be more appropriate. Ultimately, the success of this migration demonstrates the substantial economic benefits of deploying specialized decision models for agents within large-scale enterprise environments.

标准电视信号竟然可以成为一种出人意料的"运动探测器"，甚至能透过墙壁捕捉到人的移动。通过使用像 Ettus USRP B200 这类设备，对本地电视台信号的频谱图进行分析，就能看到信号与环境之间的相互作用。直接且无遮挡的信号会呈现为一条强烈、稳定不变的线；而当有人走到信号传播路径附近时，频谱图上就会出现清晰可见的扰动。

这些扰动源于选择性衰减：强反射会把直达信号抵消。由于反射的相位随信号频率变化，这种抵消并不均匀，因而在频谱图上会形成具有辨识度的特征图案。这种状态极为敏感，哪怕是家里很轻微的移动，也会在频谱图中表现得非常明显。

当人开始走动，比如从椅子上起身或在相邻房间里行走时，频谱图会显示出醒目的条纹状图案。这些条纹会以精确的间隔反复出现，因为当传播路径长度每改变一个波长时，反射就会恰好把残余信号再次抵消。通过把这些图案出现的时间与信号"瀑布图"的刷新速率对应起来，甚至可以推算出人的移动速度。

在这个例子里，计算结果显示行走速度约为每秒 1.3 米，也就是每小时 4.6 公里。由于受试者的运动方向与直达信号路径大致垂直，因此不需要进行复杂的几何修正就能得到这个数值。这说明，一套相对简单的无线电接收系统也能以相当高的精度追踪运动。

尽管电视信号能把这一效应直观地呈现出来，但它们并不是唯一可用的媒介。其他宽带无线电信号（例如 WiFi 和 LTE）同样可以被用来提取类似信息。虽然这些信号在频谱图上可能呈现得更"杂乱"，但底层原理依旧相同，使得在不同的无线电频率环境下，都能利用这种方式穿透墙壁探测运动。 A standard television signal can act as a surprising tool for detecting movement, even through walls. By analyzing the spectrogram of a local TV station signal using equipment like an Ettus USRP B200, one can observe how the signal interacts with the environment. While a direct, unobstructed signal appears as a strong and unchanging line, the introduction of a person walking near the signal path creates visible disturbances.

These disturbances are caused by selective fading, a phenomenon where a direct signal is canceled out by a strong reflection. Because the phase of this reflection is dependent on the signal frequency, the cancellation effect is not uniform, resulting in distinctive patterns on a spectrogram. This state is remarkably sensitive, reacting clearly to even minor movements within a home.

When a person moves around, such as getting up from a chair or walking in an adjacent room, the spectrogram displays clear, striped patterns. These stripes appear at precise intervals because the reflection perfectly cancels the residual signal every time the path length changes by one wavelength. By calculating the timing of these patterns against the update rate of the signal waterfall, one can even determine the speed of the person's movement.

In this instance, the math revealed a walking speed of approximately 1.3 meters per second, or 4.6 kilometers per hour. Because the subject was walking perpendicular to the direct signal path, no complex geometric adjustments were required to arrive at this figure. This demonstrates that a relatively simple radio receiver is capable of tracking motion with significant accuracy.

While television signals provide a clear visual demonstration of this effect, they are not the only medium that can be used for such surveillance. Other wideband radio signals, including WiFi and LTE, can also be exploited to extract similar data. While these signals might produce a messier visual output on a spectrogram, the underlying principle remains the same, allowing for the detection of movement through walls across various radio frequency environments.