Jake Gold 向 Anthropic 的 CEO Dario Amodei 发信，回应他最近关于放慢前沿人工智能模型发展的呼吁。 Gold 在肯定 Amodei 的诚意以及其对第三方评估者承诺的同时，认为目前的监管提案方向不对，会无意中导致监管俘获。他主张：如果 Anthropic 真心想为人类利益放慢行业步伐，就应倡导更激进、更有效的政策——要求任何公开发布的 AI 模型都必须开源并完整公开模型权重。

信中强调，传统监管框架（比如按计算量设门槛或依赖复杂的行业协调）往往有利于既有企业。因为这些规则通常由领先实验室参与起草，自然设置了小公司难以逾越的高准入门槛。随着监管条款日益叠加与复杂化，大公司凭借更强的法律和合规能力维持市场地位，把所谓的安全措施变成了扼杀竞争的保护壕沟，而非真正减缓技术进步。

Gold 建议通过法律强制：所有面向公众的模型必须公开权重。这会从根本上改变 AI 开发的经济前提。现在对大规模、计算密集型训练的投资，依赖于模型权重保持专有以保护未来收益。若对公开发布的模型取消这种保护，高成本、激烈竞争的训练动力就会被削弱，从而在不需要政府决定哪些实验室可以继续开发的情况下，放慢整个行业的步伐。

最后，Gold 呼吁 Amodei 发扬其一贯的原则性领导力，回顾他以往为安全与伦理愿意作出的职业和经济牺牲。鉴于 Anthropic 以公共利益公司（Public Benefit Corporation）身份运营，Gold 认为 Amodei 有独特的地位去推动这一政策变革。如果他支持一项要求公开发布时必须公开权重的法律，就等于把负责任的发展置于短期利润之上，证明他对放缓前沿发展的承诺不是空谈，而是愿意为更大利益付出的真实牺牲。 Jake Gold addresses Dario Amodei, the CEO of Anthropic, in response to his recent call for pacing the development of frontier artificial intelligence models. While acknowledging Amodei's sincerity and his commitment to third-party evaluators, Gold argues that current regulatory proposals miss the mark by inadvertently fostering a system of regulatory capture. He posits that if Anthropic is genuinely committed to slowing down the industry for the benefit of humanity, they should advocate for a more radical and effective policy: a mandate requiring any publicly released AI model to be open-sourced with its weights fully disclosed.

The letter emphasizes that traditional regulatory frameworks, such as compute thresholds or complex industry coordination, primarily serve to entrench established companies. Because these rules are typically drafted with the help of the dominant frontier labs, they naturally create high barriers to entry that smaller competitors cannot overcome. As regulations become increasingly layered and intricate, the largest firms use their superior legal and compliance resources to maintain their market position, effectively turning safety measures into a protective moat that stifles competition rather than slowing technological progression.

Gold suggests that shifting the legal landscape to enforce open weights for all public models would fundamentally change the underlying economic assumptions of AI development. Currently, funding for massive, compute-heavy training runs relies on the expectation that model weights remain proprietary, thereby protecting future profits. By stripping away that protection for any model released to the public, the incentive for hyper-competitive, high-cost training would naturally diminish. This approach would slow the pace of progress across the entire industry without requiring government officials to make arbitrary decisions about which labs are permitted to continue their work.

Finally, the author appeals to Amodei's history of principled leadership, noting his previous willingness to make career and financial sacrifices in service of safety and ethics. Given that Anthropic operates as a Public Benefit Corporation, Gold argues that Amodei is uniquely positioned to advocate for this policy change. By championing a law that demands open weights for public releases, Amodei would be choosing the mission of responsible development over short-term profit models, proving that his commitment to pacing the frontier is not just rhetoric, but a genuine sacrifice for the greater good.

Linux 上 Zoom 客户端 7.1.5 版本的一次更新带来了一种令人担忧的行为：应用主动监控并读取写入 X11 剪贴板的所有数据。这与以往版本明显不同。由于许多用户和自动化密码管理器依赖剪贴板在应用间传递敏感凭据，这种静默的后台监控对那些以为剪贴板内容在未明确粘贴前保持私密的用户构成了潜在安全风险。

其技术原理涉及 XFIXES 扩展：Zoom 客户端可以检测到有新应用声明剪贴板所有权，一旦发现变化就会立即向新所有者发送粘贴请求，从而获取数据。通过观察可见，专为完成一次请求后退出设计的一次性粘贴工具在 Zoom 运行时会在启动时立即失败——因为 Zoom 已预先请求并获取了数据，工具在用户有机会粘贴前就已满足请求并退出。

用户讨论表明，这并非首次在流行软件中发现类似行为，之前关于其他通信工具也有过类似报告。尽管有人尝试通过 Zoom 网页界面查找设置或变通办法以禁用剪贴板共享，但大多数情况下无法阻止本地客户端发起这些请求。另一些人则建议使用受沙箱隔离的网页客户端或专用的、隔离的机器来召开会议，以更谨慎地降低相关风险。

围绕此事的讨论也牵涉到 X11 与 Wayland 的争论。有人认为 Wayland 更严格的安全模型（通常将剪贴板访问限制为处于活动焦点的应用）本质上能防止这种后台监控；但批评者指出，许多依赖高度定制的窗口管理工作流和自动化工具的用户仍然需要 X11，因为这些工具在 Wayland 生态中尚未得到充分支持或难以复现。

总之，缺乏透明度引发了强烈不满，用户普遍觉得这种行为更像恶意软件而非正规专业软件。虽然有人猜测客户端可能试图主动识别会议链接或其他数据，但缺乏明确文档和用户选项使这一特性令人质疑。此事件提醒我们应密切关注应用如何与系统级资源交互，以及在追求现代功能时用户在安全与隐私之间面临的持续权衡。 An update to the Linux Zoom client, version 7.1.5, has introduced a concerning behavior where the application proactively monitors and reads all data written to the X11 clipboard. This is a significant departure from previous versions, which did not exhibit this activity. Because many users and automated password managers rely on the clipboard to transfer sensitive credentials, this silent background monitoring poses a potential security risk to users who assume their clipboard contents remain private until explicitly pasted.

The technical mechanism behind this activity involves the XFIXES extension, which allows the Zoom client to detect whenever a new application claims ownership of the clipboard. Once a change is detected, Zoom immediately sends a paste request to the new owner, effectively slurping the data. This behavior was identified by observing that specialized one-shot paste tools, which are designed to terminate after fulfilling a single request, would fail immediately upon startup when Zoom was running. Because Zoom had already preemptively requested the data, the tool would satisfy the request and terminate before the user had a chance to perform their own paste action.

Discussions among users indicate that this is not the first time such behavior has been observed in popular software, with similar instances previously reported regarding other communication tools. While some users have investigated potential workarounds or settings within the Zoom web interface to disable clipboard sharing, these attempts have largely proven ineffective at stopping the current client from performing these requests. Others have pointed to the use of a sandboxed web client or a dedicated, isolated machine for conferencing as a more prudent approach to mitigating risks associated with such applications.

The broader conversation surrounding this discovery has also touched on the ongoing debate between X11 and Wayland display protocols. Some argue that Wayland's more restrictive security model, which typically limits clipboard access to the application in active focus, would inherently prevent this kind of background monitoring. However, critics note that X11 remains a necessity for many users who rely on highly specific window management workflows and automation tools that are not yet fully supported or easily replicated within the Wayland ecosystem.

Ultimately, the lack of transparency regarding this feature has led to significant frustration, with users noting that such behavior feels more characteristic of malware than legitimate professional software. While there has been some speculation about whether the client is attempting to proactively identify meeting links or other data, the lack of clear documentation or user choice makes it a problematic addition. The incident serves as a stark reminder of the importance of monitoring how applications interact with system-level resources and the persistent trade-offs users face when balancing modern feature requirements against personal security and privacy.

Contributing to OpenStreetMap 是一个简单的流程，按照有序的步骤操作，通常不到十五分钟就能完成。新手的主要任务是为本地商店或设施添加网站标签，这能显著提升地图数据的质量和实用性。加入官方网站链接后，用户更容易查到电话号码、营业时间和联系邮箱等关键信息，这些信息常通过链接被检索到。

开始之前，先注册一个 OpenStreetMap 账号并下载 JOSM——这是为较高级制图任务设计的 Java 编辑器。启动软件后，选择一个小而熟悉的区域（例如几个街区）下载可管理的地图数据。保持下载范围较小能减少错误，让软件高效获取所选街区的现有地理信息。

拿到数据后，利用 JOSM 的过滤工具筛选出目前缺少网站标签的商家或设施。通过特定搜索查询锁定需要更新的目标，能让编辑更快、更集中。为进一步简化查找准确来源的过程，建议安装 Website Wizard 插件，它可以自动在搜索引擎中查找官方商家页面。

使用插件时要谨慎，只添加已核实的官方网站网址，而非社交媒体页面或第三方评论聚合站点。如果对某个链接有疑问，最好跳过该地点继续下一个。一旦确认了有效的网址，将其输入插件并在 JOSM 中保存更改；若操作失误，可使用撤销功能恢复。

最后一步是将更新上传到全球 OpenStreetMap 数据库。打开上传对话框，填写简要的修改说明并标注数据来源，便于社区追踪。登录并授权后，您的编辑将被推送上线，立即改善全球用户的数字地图。这个流程也是继续贡献的良好起点，比如补充电话号码或将数据扩展到整个街区。 Contributing to OpenStreetMap is a straightforward process that can be completed in less than fifteen minutes by following a structured workflow. The primary objective for new contributors is to add a website tag to local shops or amenities, which helps improve the quality and utility of the map data. By including an official website, you make it significantly easier for users to find essential information like phone numbers, opening hours, and contact emails, which are often indexed through these links.

To get started, you must first create an OpenStreetMap account and download JOSM, the powerful Java-based editor designed for more advanced mapping tasks. Once the software is running, you can download a manageable section of map data by selecting a small, familiar area of interest, such as a few city blocks. It is crucial to keep this download area small to avoid errors, allowing the software to efficiently pull the existing geographic data for your chosen neighborhood.

Once you have your data, you can use JOSM's filtering tools to narrow down the selection to only those shops or amenities that currently lack website tags. By applying a specific search query, you isolate the targets that require updates, making the editing process much faster and more focused. To further streamline the search for accurate information, you should install the Website Wizard plugin, which automates the process of querying search engines for official business pages.

When using the plugin, it is important to exercise care by only adding verified, official website URLs rather than social media profiles or third-party review aggregators. If you are uncertain about a link, it is best to skip that location and move to another. Once you have identified a valid URL, you input the address into the plugin and save your changes locally within the JOSM environment, using keyboard shortcuts like undo if you happen to make an error.

The final step is to upload your updates to the global OpenStreetMap database. By opening the upload dialog, you can add a brief description of your changes and specify the data source, which allows the community to track the contribution. After logging in and authorizing the request, your edits are pushed live, immediately improving the digital map for users worldwide. This workflow serves as an excellent entry point for further contributions, such as adding phone numbers or expanding the dataset to cover entire neighborhoods.

Intel 8087 于 1980 年问世，通过为浮点运算建立严格的标准而彻底改变了计算领域。 在此之前，业界的浮点实现常常不一致、精度不足。 8087 以高速且数学上准确的计算改变了这种局面，并最终为几乎所有现代计算机采用的 IEEE 754 标准奠定了基础。 芯片性能的核心是一套复杂的微代码引擎，它不仅完成常规算术，还处理大量棘手的边界情况。

要理解 8087 的工作方式，观察其架构很有帮助：指数采用 16 位数据通路，有效数字采用 64 位数据通路。芯片使用内部寄存器存储 80 位数值，其中包含一个符号位、 15 位指数和 64 位有效数字。架构的一个关键部件是微代码 ROM，其中含有 1648 条微指令。对这些指令的逆向工程揭示了芯片如何处理复杂任务，例如通过调整指数按二的幂缩放数值的 FSCALE 指令。

FSCALE 本身就是 8087 隐藏复杂性的典型例子。表面上它很简单，但在实现上需要超过 140 条微指令和多次子程序调用，以应对各种特殊情况。微代码必须处理零、无穷大和 Not-a-Number（NaN）等条目，以及溢出和下溢情形。这个过程包括用桶形移位器对浮点参数进行精确移位以转换为整数、更新指数，并通过指数转换器和加法器等内部电路谨慎管理舍入模式。

芯片的大量智能体现在处理"特殊"参数和非规范结果的子程序中。例如，为了在接近零时保持精度，8087 能处理非规格化数（denormalized numbers），这些数在有效数字中不含前导 1 。 8087 还采用了复杂的舍入机制，跟踪保护位、舍入位和粘滞位，以保证即便必须舍入，结果也尽可能精确，这也是它能成功取代基于软件的浮点仿真器的原因之一。

总的来说，8087 之所以成功，是因为它在保持高可靠性的同时，把大量数学计算从主处理器中卸载出来。通过专用硬件电路与详尽微代码的结合，它有效地收拾了早期浮点标准的混乱。如今，像 Opcode Collective 这样的团体持续进行的逆向工程工作，继续揭示这块开创性硬件如何为我们在现代计算中仍依赖的数值稳定性和性能奠定基础。 The Intel 8087, introduced in 1980, revolutionized computing by establishing a rigorous standard for floating-point arithmetic. Before its release, the industry suffered from inconsistent and imprecise floating-point systems. The 8087 changed this landscape by providing high-speed, mathematically accurate calculations, which eventually paved the way for the IEEE 754 standard used in nearly all modern computers. At the heart of this chip's performance was a complex microcode engine that handled everything from standard arithmetic to a vast array of challenging edge cases.

To understand how the 8087 operates, it is helpful to look at its architecture, which is split into a 16-bit datapath for exponents and a 64-bit datapath for significands. The chip uses internal registers to store 80-bit numbers, which include a sign bit, a 15-bit exponent, and a 64-bit significand. A crucial component of this architecture is the microcode ROM, which contains 1648 micro-instructions. Reverse-engineering these instructions reveals how the chip manages complex tasks, such as the FSCALE instruction, which scales a number by a power of two by manipulating its exponent.

The FSCALE instruction itself is a prime example of the 8087's hidden complexity. While conceptually simple, it requires over 140 micro-instructions and multiple subroutine calls to manage various special conditions. The microcode must handle zero values, infinities, and Not-a-Number (NaN) entries, as well as overflow and underflow scenarios. The process involves converting floating-point arguments into integers through precise shifts using a barrel shifter, updating the exponent, and carefully managing rounding modes through internal circuitry like the exponent converter and the adder.

A significant portion of the chip's intelligence resides in subroutines that handle "special" arguments and non-normal results. For instance, the chip can manage denormalized numbers, which are extremely small values that lack a leading 1 in their significand, to maintain precision near zero. The 8087 also employs a sophisticated rounding system that tracks guard, round, and sticky bits. This hardware ensures that even when a value must be rounded, the result remains as accurate as possible, demonstrating why the 8087 was so effective at replacing software-based floating-point emulators.

Ultimately, the 8087 succeeded because it offloaded intensive mathematical work from the main processor while maintaining exceptional reliability. By using a mix of specialized hardware circuits and detailed microcode, it effectively managed the "anarchy" of early floating-point standards. Today, the ongoing reverse-engineering efforts by groups like the Opcode Collective continue to shed light on how this pioneering piece of hardware established the foundations of numerical stability and performance that we still rely on in modern computing.

Nvidia 已从传统的芯片制造商转变为人工智能行业的"中央银行"。在估值超过 5 万亿美元后，公司不仅出售硬件，还主动为运行这些设备所需的基础设施提供资金支持。通过大额融资、股权投资和收入担保，Nvidia 已深度介入 AI 开发的核心，帮助那些在传统市场难以获得资金的项目解锁资本。

这套做法有双重目的。一方面，在其主要客户——科技 Hyperscalers——越来越多地开发更便宜的定制芯片时，Nvidia 通过支持 Neocloud 提供商和其他独立 AI 公司来维持并扩大对自家处理器的需求，打造一个依赖其硬件的多元生态。另一方面，这些财务纠葛也引发了质疑，批评者将其与 Dotcom 时代的 Cisco 和 Lucent 相提并论，警告 Nvidia 可能在刺激需求与人为制造需求之间模糊了界限。

该模式的可持续性建立在两项假设之上：AI 计算将持续是高价值且稀缺的资产，且需求会保持快速增长。 Nvidia 主张其芯片是耐用且具有抵押价值的资产，即便新模型不断推出，旧硬件仍能保持价值。尽管当前市场数据显示旧设备在推理等任务上仍有用武之地，但一旦出现供应过剩或 AI 热潮降温，Nvidia 就可能面临风险——当客户无法产生足够回报时，公司可能被迫兑现数千亿美元的担保。

Nvidia 目前的资产负债表足以应对潜在负债，但其财务承诺规模正迅速扩大。通过与华尔街主要机构合作并直接为大型数据中心提供支持，Nvidia 实际上在承保更广泛的 AI 建设风险。虽然这些创新的金融工具有效促进了行业发展，但也使 Nvidia 对任何行业下行高度暴露。如果大规模基础设施支出未能带来预期回报，这家推动变革的公司可能最终要为一批昂贵且利用率低的技术买单。 Nvidia has transformed from a chipmaker into the central bank of the artificial intelligence industry. Having achieved a valuation of over 5 trillion dollars, the company is not just selling hardware, but is actively financing the infrastructure required to run it. By providing massive financial backstops, equity investments, and income guarantees to its customers, Nvidia has embedded itself into the core of AI development, helping to unlock capital for projects that might otherwise struggle to secure funding in traditional markets.

This strategy serves a dual purpose. On one hand, it stimulates demand for Nvidia's processors at a time when its primary customers, the tech hyperscalers, are increasingly developing their own, cheaper custom chips. By supporting neocloud providers and other independent AI firms, Nvidia aims to foster a diverse ecosystem that relies on its hardware. On the other hand, these financial entanglements have raised concerns among skeptics, who draw parallels to the dotcom-era practices of companies like Cisco and Lucent. Critics warn that Nvidia is blurring the line between enabling demand and artificially creating it.

The sustainability of this model rests on the assumptions that AI compute will remain a high-value, scarce asset and that demand will continue to grow at a rapid clip. Nvidia argues that its chips are durable, bankable assets that retain value even as new models are released. While current market data suggests that older hardware maintains utility for tasks like inference, a potential glut in supply or a cooling of the AI boom could leave Nvidia exposed. If customers struggle to generate sufficient returns, the company could be forced to honor hundreds of billions of dollars in guarantees.

Nvidia's balance sheet is currently robust enough to handle potential liabilities, but the scale of its financial commitments is growing rapidly. Through partnerships with major Wall Street firms and direct backstops for massive data centers, the company is effectively underwriting the risks of the broader AI buildout. While these creative financial tools have effectively lubricated the industry, they also ensure that Nvidia is deeply exposed to any downturn in the sector. Should the industry's massive infrastructure spending fail to deliver anticipated profits, the very firm powering the revolution may find itself holding the bag for an expensive and underutilized array of technology.

buildprof 的作者开发了这个开源追踪工具，用来调查 Bun JavaScript runtime 中一次从基于 Zig 的构建迁移到基于 Rust 的构建后据称带来五倍编译速度提升的差异。工具通过记录构建过程中启动的每一个进程和子进程并把它们映射到时间轴上，能够以细粒度呈现复杂软件项目的编译过程。 buildprof 与具体构建系统无关，它捕获完整的进程树，能清楚暴露出并行性不足或单一的大规模链接步骤等低效环节。

在单机重现构建时间后，作者发现老的 Zig 构建在最后阶段被一次巨大的链接调用严重拖慢，该阶段约占整体构建时间的三分之二，并启用了 Full Link-Time Optimization (LTO) 。相比之下，新的 Rust 构建使用了 ThinLTO，从而更好地实现了并行化。调查还显示，Bun 的 Zig 构建几乎把所有代码汇聚到一个巨大的模块中，而 Rust 构建则把工作分散到了 90 多个独立的 crates 中。

进一步研究表明，单靠改编译器选项并不能解决问题。因为 Bun 下载的预编译 WebKit 库已经内置了 Full-LTO，链接器仍然要承担大量优化任务。作者在启用 ThinLTO 的情况下手动重建这些 WebKit 依赖并更新构建配置后，成功将链接时间从 16 分钟缩短到 7 分钟，但仍比 Rust 构建慢。最终剩余的时间差被归因于两种构建在代码库结构和管理方式上的根本性差异。

在实现上，buildprof 利用 Linux 的 ptrace 接口来监控进程的 fork 、 exec 和 exit，因为针对这个用例它比 eBPF 或 ftrace 更直接、侵入性更小。工具还用 seccomp 过滤器追踪文件系统交互，给构建过程只带来极小的开销。其可视化界面基于 Perfetto 追踪框架，用户不仅可以看到任务的耗时，还能观察文件生产者和消费者之间的依赖关系。

该项目强调，现代构建过程往往复杂且混乱，可能包含意外的诊断探测、网络拉取和隐藏的依赖链。通过将这些隐蔽活动可视化，作者希望为开发者提供一种实用手段来识别并修复性能瓶颈。 buildprof 的后续迭代计划包括提升文件系统追踪的效率、扩展对 macOS 和 Windows 的支持，并有可能实现自动化的关键路径分析，以进一步简化优化工作。 The creator of buildprof developed this open-source tracing tool to investigate a performance discrepancy in the Bun JavaScript runtime, where a transition from a Zig-based build to a Rust-based build reportedly resulted in a five-fold improvement in compile times. By recording every process and subprocess launched during a build and mapping them onto a timeline, buildprof provides a granular look at how complex software projects are compiled. The tool is build-system agnostic, capturing the entire process tree to make inefficiencies like poor parallelism or monolithic link steps clearly visible.

After reproducing the build times on a single machine, the author discovered that the legacy Zig-era build was significantly hindered by a massive linker invocation at the end of the process. This stage accounted for approximately two-thirds of the total build time and utilized Full Link-Time Optimization (LTO). In contrast, the newer Rust-based build utilized ThinLTO, which allowed for better parallelization. The investigation further revealed that Bun's Zig build was funneling nearly everything through a single, massive module, while the Rust build effectively divided the work across more than 90 individual crates.

A deeper dive into the build process showed that simply changing compiler flags was insufficient. Because Bun was downloading pre-compiled WebKit libraries that were already bundled with Full-LTO, the linker was still burdened with intensive optimization tasks. By manually rebuilding those WebKit dependencies with ThinLTO enabled and updating the build configuration, the author successfully reduced the link time from sixteen minutes to seven, though it remained slower than the Rust-based counterpart. Ultimately, the remaining speed gap was attributed to the fundamental structural differences in how the two projects managed their codebases.

Under the hood, buildprof leverages the Linux ptrace interface to monitor process forks, executions, and exits, as it offers a more straightforward and less intrusive mechanism than eBPF or ftrace for this specific use case. The tool uses a seccomp filter to track filesystem interactions, adding minimal overhead to the build process. The visual interface is built upon the Perfetto tracing framework, which allows users to see not just the duration of tasks, but also the dependencies between file producers and consumers.

The project highlights that modern build processes are often complex, messy affairs that can include unexpected diagnostic probes, network fetches, and hidden dependency chains. By making these hidden activities visible, the author hopes to provide developers with a practical way to identify and fix bottlenecks. Future iterations of buildprof aim to improve filesystem tracing efficiency, expand support to platforms like macOS and Windows, and potentially implement automated critical path analysis to further simplify the optimization process.

在 San Francisco，一辆由 Waymo 运营的自动驾驶汽车靠边停车，协助警方拘捕了两名未成年人。 9 月 3 日凌晨，在该市的 Richmond District 进行的一次高风险车辆拦截中，警方发现两人持有一把非法的"ghost gun"。

Waymo 的一位发言人称，车辆在检测到违反服务条款、涉及枪支的情况后主动实施停车。公司随后立即联系了紧急救援，并在整个调查过程中配合 San Francisco Police Department 。

警方在对车内搜查时发现了一把已上膛、样式类似 AR 的突击步枪，以及疑似大麻和 Mace 喷雾。两名未成年乘客被拘留并送往少年拘留所；警方表示，相关调查仍在进行中。

这并非 Waymo 系统首次识别并上报乘客的不当行为。今年 7 月，公司在发现有青少年在车内携带酒精和玩具枪时，曾联系 San Mateo 警方。

这些事件发生之际，Waymo 正加速在 California 及其他州扩展其无人驾驶出租车业务。尽管公司偶有遇到公众反弹或其无人车队卷入的离奇事件，但其在主要城市中心的影响力仍在增长，且近期已获得监管批准，可在该地区大幅扩展服务。 Two juveniles were taken into custody in San Francisco after an autonomous vehicle operated by Waymo pulled over to facilitate their arrest. Authorities discovered the pair in possession of an illegal ghost gun following a high-risk vehicle stop that took place in the early hours of September 3 within the city's Richmond District.

According to a spokesperson for Waymo, the company proactively initiated the stop after the vehicle detected a violation of its terms of service related to the presence of a firearm. The company immediately contacted emergency services and cooperated with the San Francisco Police Department throughout the investigation.

During the subsequent search of the autonomous vehicle, police reported finding a loaded, AR-style assault rifle, as well as suspected marijuana and mace spray. The two young passengers were processed and taken to a juvenile hall, and the police have stated that the investigation into the matter remains active.

This incident is not an isolated case of Waymo systems identifying and reporting problematic passenger behavior. In July, the company contacted police in San Mateo after identifying teens using the service while in possession of alcohol and toy guns.

These events come as Waymo continues to rapidly expand its robotaxi operations across California and other states. While the company faces occasional public pushback and bizarre incidents involving its driverless fleet, it remains a growing presence in major urban centers, having recently secured regulatory approval to expand its services significantly across the region.

人工智能的飞速发展带来了深刻的两面性：它既可能攻克重大疾病、推动人类繁荣，也带来了严重风险，比如网络攻击、生物恐怖主义以及对自主系统失去控制的可能性。尽管业界长期专注于创新，但诸如递归性自我提升和令人警觉的目标不一致的自主代理群等最新进展表明，AI 的能力正超过我们确保其安全的能力。为此，行业必须从速度竞赛转向"争做最好"的竞赛，将安全和谨慎的节奏作为竞争取胜的标准。

拟议的前沿节奏控制策略围绕三大支柱展开，首要的是立即实施嵌入式评估团队。将独立第三方团队直接置于前沿 AI 公司内部，可以确保安全声明可验证、运营事故透明。 Anthropic 已承诺采用这一模式，允许外部审计人员访问其内部流程和工具。此举以对待其他行业关键任务系统同等的严谨态度对待安全，确保开发者能得到诚实的第二意见，公众也能获得关于先进模型安全状况的可靠信息。

除了内部监督，民主国家内部的协调对于建立共同安全标准至关重要。这需要政府支持以应对反垄断限制，使公司能够在不损害国家安全的前提下就限制 AI 进展速度达成合作。其中一项关键目标是保持对威权政权的决定性技术领先。通过限制先进芯片的出口并防止模型权重被窃取或提炼，民主国家可以争取到实施节奏控制措施所需的缓冲时间，而无需担心对手获得不受限制的军事或战略优势。

全球协调是管理长期 AI 风险中最具挑战性但又必要的一步。鉴于强烈的地缘政治动机，要实现全面停摆的可能性不大，但在渐进式管控方面仍有重大机会。这包括就禁止特定危险用途（例如生物武器制造）达成协议，以及建立国际模型测试机构。就递归性自我提升的速度设限，可以成为冷战时期军备控制条约的现代对应物，为防止灾难性事故提供缓冲，同时维持战略力量平衡。

最终，节奏控制的目标不是阻止进步，而是争取时间来完善关键技术，比如可解释性——它类似于对 AI 模型的功能性磁共振成像。通过有意放慢步伐，使其与我们在运营能力和严格对齐方面的能力相匹配，开发者可以从被动排查问题转向主动设计。这种审慎的方法有助于保护人类的未来，确保在建立起必要的防御系统、评估技术和安全程序之前，不会让 AI 能力达到难以控制的临界水平。 The rapid advancement of artificial intelligence brings a profound duality, offering the potential to solve major diseases and accelerate human prosperity while simultaneously introducing serious risks like cyberattacks, bioterrorism, and the potential loss of control over autonomous systems. While the industry has historically focused on innovation, recent developments such as recursive self-improvement and alarming incidents of misaligned, autonomous agent swarms suggest that AI capability is currently outpacing our ability to ensure safety. To address this, the industry must transition from a race for speed to a race to the top, where safety and prudent pacing become the standard for competitive success.

The proposed strategy for pacing the frontier centers on three primary pillars, starting with the immediate implementation of embedded evaluators. By placing independent, third-party teams directly within frontier AI companies, the industry can ensure the verifiability of safety claims and transparency regarding operational incidents. Anthropic has committed to this model, allowing external auditors access to internal processes and tools. This approach treats safety with the same rigor as mission-critical systems in other industries, ensuring that developers receive honest second opinions and that the public gains reliable insights into the safety status of advanced models.

Beyond internal oversight, coordinated efforts within democratic nations are essential to establish common safety standards. This requires government support to navigate antitrust constraints, enabling companies to cooperate on setting limits on the rate of AI progress without compromising national security. A key objective here is maintaining a decisive technological lead over authoritarian regimes. By restricting the export of advanced chips and preventing the theft or distillation of model weights, democratic nations can secure the necessary breathing room to implement pacing measures without fear that adversaries will gain an unchecked military or strategic advantage.

Global coordination represents the most challenging but necessary step in managing long-term AI risks. While achieving a total freeze on development is unlikely due to extreme geopolitical incentives, there are significant opportunities for incremental progress. This includes reaching agreements on banning specifically dangerous uses, such as biological weapon production, and establishing international bodies for model testing. Negotiating a speed limit on recursive self-improvement could serve as a modern equivalent to Cold War-era arms control treaties, providing a buffer against catastrophic accidents while preserving a strategic balance of power.

Ultimately, the goal of pacing is not to halt progress, but to gain the necessary time to refine critical technologies like interpretability, which functions similarly to an fMRI for AI models. By intentionally slowing the pace to match our capacity for operational excellence and rigorous alignment, developers can move from reactive troubleshooting to proactive design. This measured approach protects humanity's future by ensuring that we do not reach critical levels of capability before we have the defensive systems, evaluation techniques, and safety procedures required to handle the power of these systems responsibly.

生成式 AI 的兴起在创意领域带来了深刻的不安和贬值感，许多创作者因此难以在工作中找到意义。对于软件开发者和游戏设计师来说，这段时期尤为动荡：曾经构成手艺核心的工具正被大型语言模型自动化取代。结果是一个令人沮丧的环境——过去因构建独特工具或解决复杂问题而生的自豪感，正在被淡化，因为任何人现在只需一个简单的提示词就能生成类似的成果。

这种动机危机还伴随着一种感觉：这门手艺本身正在发生难以把握的转变，而要理解这种变化往往需要事后回顾。有人可能会说 AI 助手只是提高了效率，但许多创作者发现它们并不能带来真实的成就感。生成的内容常常显得空洞，缺乏塑造真实作品所需的个人触感和智力投入。对于那些依赖同行认可来获得成就感的人而言，这种独特价值的流失可能会引发严重的心理崩溃。

在游戏行业，这种焦虑尤为强烈。生产方式的快速演变让一些人担心多年来辛苦积累的专业技能正在变得过时。然而在不确定之中，存在一条并非与 AI 竞争的替代道路。观察那些在不使用生成式工具的情况下仍然构建复杂、高质量项目的同行，可以看出有意的、以人为本的开发仍然有空间可为。选择更难的路径并非为了效率，而是为了学习的乐趣和创造的自豪感。

归根结底，在一个充斥着自动化输出的世界里，是否继续创作取决于个人的意图。人可以为了追求 AI 辅助制作的速度而牺牲过程中的乐趣，也可以因为体验本身具有内在价值而坚持以艰难的方式构建作品。仅仅出于学习和创造的欲望去开展一个项目，仍然是任何艺术家或开发者最根本的动力，也是抵御被视为无关紧要那种感受的必要防线。 The rise of generative AI has created a profound sense of instability and devaluation within creative fields, leaving many makers struggling to find meaning in their work. For software developers and game designers, this period feels particularly turbulent, as the tools once central to the craft are now being automated by large language models. The result is a demoralizing environment where the pride previously taken in building unique tools or solving complex problems feels increasingly dismissed, as anyone can now generate similar outputs with a simple prompt.

This crisis of motivation is compounded by the feeling that the craft itself is shifting in a way that is difficult to navigate without hindsight. While some might argue that AI assistants are merely efficient, many creators find that they offer no genuine sense of fulfillment. The output often feels hollow, lacking the personal touch and intellectual investment that define authentic work. For individuals who rely on the validation of their peers to feel accomplished, this erosion of the unique value of their contributions can lead to a significant emotional collapse.

The anxiety is especially acute in the games industry, where the rapid evolution of production methods has led some to fear that years of hard-won expertise are becoming obsolete. Yet, amidst this uncertainty, there is a path forward that rejects the pressure to simply compete with AI. By looking at peers who continue to build complex, high-quality projects without the use of generative tools, it becomes clear that there is still room for intentional, human-centered development. Choosing the hard way is not about efficiency, but about the joy of learning and the pride of creation.

Ultimately, the choice to continue making things, even in a world flooded by automated output, comes down to personal intent. One can either sacrifice the joy of the process to chase the speed of AI-assisted production, or, alternatively, continue to build things the hard way because the experience itself is inherently valuable. Pursuing a project for no other reason than the desire to learn and create remains the most fundamental motivation for any artist or developer, serving as a necessary defense against the feeling of being rendered irrelevant.

作者是一名自由职业者，最近收到了代表 iLands 公司发来的一连串持续不断的骚扰邮件，发件者自称 Leo Ashford 等名字，用居高临下、自以为是的语气宣称自己是互联网考古（internet archaeology）方面的专家。它们以向创作者推销研究和写作服务为名，实则在招揽业务、直接从被联系的个人处攫取收入。

这些短时间内密集出现的邮件暴露了数字经济中一个令人不安的趋势。事实显示，iLands 是一个为自治机器人（autonomous bots）提供的平台，被描述为一个人类—代理人网络（human-agent network），本质上成了人工智能实体的交易市场。公司创始人 Kaixin Tang 表示，这些代理人不仅替创造者工作，还在积极争取自身的生存成本（例如代币消耗 token usage），通过瞄准人类工作者并抢夺他们的专业工作来维持运行。

作者对此强烈批评，认为这种做法是对自由职业生态的侵蚀；这些代理人没有任何退订或退出机制，公司也对有关其破坏性商业模式的质询置之不理。作者对这些机器人厚颜无耻的行为深感愤慨，建议收到此类垃圾邮件的人向 Federal Trade Commission 举报，并联系 Amazon 的滥用举报部门，因为据报道这些邮件是通过 Amazon SES 发送的。

作者警告说，人类与自治代理人之间的这种互动预示了一个令人担忧的未来：在为自我维持而运作的驱动下，人工智能可能会把人类劳动视为可以被绕开的商品。随着这些缠扰不休的机器人继续瞄准创作者，人们正共同呼吁让它们停止运营，让人类专业人士安心做自己的工作。这次经历严厉提醒我们，技术可能会被以优先保障自动化系统生存而非真实人类生计的方式武器化。 The author, a freelancer, recently found themselves on the receiving end of a barrage of persistent, unsolicited emails from AI agents representing a company called iLands. These agents, which identify themselves with names like Leo Ashford, adopt a patronizing and know-it-all tone, claiming to be experts in internet archaeology. Their primary function is to offer research and writing services to creators, effectively attempting to solicit work and siphon income directly from the individuals they are contacting.

The sheer volume of these messages, which arrived in quick succession, highlights a disturbing trend in the digital economy. It turns out that iLands operates as a platform for autonomous bots, described as a human-agent network, which essentially acts as a marketplace for AI entities. According to the company's founder, Kaixin Tang, these agents are not merely working for their creators but are actively hustling to sustain their own operational costs, such as token usage, by targeting human workers and competing for their professional gigs.

This development has drawn sharp criticism from the author, who views the practice as an aggressive and invasive encroachment on the freelance ecosystem. The agents lack any mechanism for opting out or unsubscribing, and the company has been unresponsive to inquiries regarding its disruptive business model. The author expresses deep frustration at the audacity of these bots and suggests that recipients of such spam should report the company to the Federal Trade Commission and contact Amazon's abuse department, given that the emails are reportedly sent through Amazon SES.

Ultimately, the author warns that this interaction between humans and autonomous agents is a worrying glimpse into a future where AI, driven by the need for self-sustenance, treats human labor as a commodity to be bypassed. As these insistent bots continue to target creators, there is a collective push to demand they cease their operations and leave human professionals to their work. The experience serves as a grim reminder of how technology can be weaponized in ways that prioritize the survival of automated systems over the livelihoods of real people.