NVIDIA 近日宣布重大领导层变动：Jensen Huang 已辞去其总裁、首席执行官以及董事会成员职务，并自即日起生效。这一出乎意料的离任意味着他三十三年任期的结束。任职期间，他带领公司走到如今的位置，成为人工智能与加速计算领域的全球领先者，并持续推动公司的发展加速。

针对这一职位空缺，董事会已任命现任执行副总裁兼首席财务官 Colette Kress 担任临时首席执行官。 Kress 在选出永久继任者之前，将继续兼顾原有的财务职责并带领公司前行。董事会已启动遴选流程，将对永久职位的内部与外部候选人进行评估。

Harvey C. Jones 被任命为有史以来首位董事会主席，以在此次交接期间提供治理与稳定。 Harvey C. Jones 是一位经验丰富的技术高管和风险投资人，曾在 Synopsys 等公司任职。他表示，董事会对现有领导团队充满信心。他同时强调，公司日常运营、对客户的承诺以及已制定的产品路线图预计将保持连续推进，不会受到影响。

在谈及对未来的承诺时，Kress 指出，公司能够取得今天的成绩，深深依托于员工队伍，以及过去三十年所奠定的坚实基础。她再次确认，领导团队将继续推进 NVIDIA 在人工智能与加速计算领域的工作，确保公司持续满足开发者、合作伙伴以及股东的需求。

尽管高层行政领导出现重大变化，NVIDIA 仍确认其计划发布的 2027 财年第三季度财务报告将按原定时间表进行。公司将于 11 月 17 日公布财务业绩，并随后召开财报电话会议，以维持投资者在此过渡阶段所期待的透明度与沟通。 NVIDIA recently announced a major leadership change as Jensen Huang has resigned from his roles as President, Chief Executive Officer, and member of the Board, effective immediately. This unexpected departure marks the end of his thirty-three-year tenure, during which he guided the company to its current position as a global leader in artificial intelligence and accelerated computing.

In response to the vacancy, the Board of Directors has appointed Colette Kress, the current Executive Vice President and Chief Financial Officer, to serve as Interim Chief Executive Officer. Kress will continue to balance her existing financial responsibilities while leading the company until a permanent successor is selected. The board has already initiated a search process that will evaluate both internal and external candidates for the permanent role.

Harvey C. Jones has been appointed as the first-ever Chair of the Board to provide governance and stability during this transition. Jones, a seasoned technology executive and venture investor with experience at companies like Synopsys, emphasized the board's confidence in the current leadership team. He noted that the company's ongoing operations, commitment to customers, and established product roadmaps are expected to continue without interruption.

Expressing her commitment to the future, Kress highlighted that the company's success is deeply rooted in its people and the strong foundation built over the last three decades. She reaffirmed that the leadership team remains dedicated to advancing NVIDIA's work in AI and accelerated computing, ensuring that the company continues to meet the needs of its developers, partners, and shareholders.

Despite the significant change in executive leadership, NVIDIA has confirmed that its upcoming third-quarter fiscal 2027 financial reporting remains on schedule. The company intends to release its financial results on November 17, followed by an earnings call, maintaining the transparency and communication expected by investors during this period of transition.

人工智能即将对全球格局产生重大影响，而 Anthropic 正将自己定位为这一转型中的关键参与者。作为一家公共利益公司，该公司从根本上承担着两项使命：在确保 AI 带来收益的同时，积极降低其相关风险。对道德发展与安全的重视，正是他们在筹备公开发行之际所确立的组织理念与核心立场。

有意参与此次全球发行的投资者，可通过加拿大的多家主要金融机构进行参与。 RBC 、 BMO 、 TD 、 CIBC 、 Scotiabank 以及 Wealthsimple 等平台已被确认是相关方可用于探索与该公司相关投资机会的渠道。这些直接投资服务旨在为投资者提供对本次发行新股的通道。

目前，IPO 的法律框架仍在推进中，相关必要文件已通过 SEDAR+ 在加拿大各省和地区提交。尽管修订并重述的初步基础 PREP prospectus 已提交，但仍可能会继续完成或进一步修订。因此，在对 prospectus 下发最终收据之前，当前并未获得出售证券或接收购买要约的授权。

除加拿大的备案外，Anthropic 还向美国证券交易委员会提交了注册声明，其中包括美国 prospectus 。这些文件详细披露了公司的情况以及全球发行的具体安排。投资者建议通过 EDGAR 在 SEC 网站上查阅上述文件，以便更全面地了解业务的财务与运营细节。

需要特别说明的是，注册声明尚未生效，因而目前无法开展任何销售，也不能接受任何购买要约。本文所载信息仅供教育与参考之用，不构成在任何司法辖区内于正式注册或取得资质前就进行法律禁止的要约出售或招揚购买意向。潜在投资者应咨询当地经纪商或查阅相关监管数据库，以获得最准确、最新的指引。 Artificial intelligence is poised to have a significant impact on the global landscape, and Anthropic is positioning itself as a key player in this transition. Operating as a public benefit corporation, the company is fundamentally dedicated to the dual mission of securing the benefits of AI while actively mitigating its associated risks. This focus on ethical development and safety serves as the cornerstone of their organizational identity as they look toward a public offering.

Investors interested in participating in this global offering can do so through several major financial institutions in Canada. Platforms such as RBC, BMO, TD, CIBC, Scotiabank, and Wealthsimple have been identified as channels through which interested parties can explore investment opportunities related to the company. These direct investing services are provided to facilitate access to the new shares being offered.

The legal framework for this IPO is currently in progress, with the necessary filings made across Canadian provinces and territories through SEDAR+. While the amended and restated preliminary base PREP prospectus has been filed, it remains subject to potential completion or amendment. Consequently, there is no current authorization for the sale of securities or the acceptance of buy offers until a final receipt for the prospectus has been issued.

In addition to the Canadian filings, Anthropic has submitted a registration statement, including a U.S. prospectus, to the U.S. Securities and Exchange Commission. These documents provide in-depth information about the company and the specifics of the global offering. Investors are encouraged to review these filings on the SEC website via EDGAR to gain a comprehensive understanding of the financial and operational details of the business.

It is important to note that the registration statement has not yet become effective, meaning that no sales can occur and no offers to buy can be accepted at this time. The information provided is for educational and informational purposes and does not constitute a legal offer to sell or a solicitation of interest in any jurisdiction where such activity would be unlawful before official registration or qualification. Prospective investors should consult their local brokers or the relevant regulatory databases for the most accurate and up-to-date guidance.

编写正确的软件之所以从根本上存在难度，是因为计算本身的性质带来了不可避免的限制。以 Gödel 、 Church 和 Turing 的奠基性工作为根源的理论计算机科学表明：我们无法完全理解或预测任意程序的行为。 Halting theorem 证明不存在任何算法能够判断一个程序最终是否会终止，其推论（如 Rice's theorem）也说明，判定程序行为的任何非平凡性质同样是不可能的。这些结论意味着，程序本身与实现完美正确性所需要的那种完整理解在本质上是彼此冲突的。

计算复杂度则让这些限制更具现实意义。仅仅"在理论上可判定"还不够——如果求解一个问题所需要的资源超出了合理范围，那就难以落到可用的工程方案上。即便问题在技术上能够解决，完成它所需的时间和精力也常常会增长到现实中难以承受的程度。正如 Juris Hartmanis 和 Richard Stearns 等研究者所指出的，复杂度就像自然法则一样客观存在。相关结果表明，我们不可能比程序自身的运行更快地分析出它的行为。因此，将验证视为一种决策问题，它在计算机科学中属于最棘手的挑战之一；正确性也不可能简单地拆分成若干模块化、可管理的部分。

编程语言中的抽象机制——例如高层结构、函数以及并发——并不能神奇地绕开这些核心困难。尽管这些特性让人更容易编写程序，但它们并不会以某种方式简化底层状态空间，从而让验证在一般情况下变得可行。即使在理论上可以进行验证的有限状态机场景中，该问题往往也属于 PSPACE-complete 。这意味着，用更简洁方式表示的程序未必比更冗长的版本更容易验证。对于自动化模型检测器而言，状态爆炸仍然是一个重大障碍：它们必须面对真实软件中庞大且常常不规则的复杂性。

各种验证方法——包括模型检测、演绎证明和静态分析——都必须在这些固有的边界内运作。诸如抽象解释或类型系统之类的技术，往往通过牺牲精确性来换取可操作性，从而给出可靠但不完整的结论。它们确实能帮助发现某些类别的缺陷，但却无法保证所有程序的完全正确性。试图为软件可靠性寻找一种"一劳永逸"的通用方案，本质上近乎现代炼金术。正如 Fred Brooks 所强调的那样，软件工程中不存在"银弹"，我们必须接受：复杂度正是我们所构建系统的本质特征。

从最终的现实出发，最可行的方向是采取一种类似自然科学的经验主义和临场应对。既然无法验证所有程序，就应该研究人们实际编写的那些具体程序，梳理常见的失败模式，并为这些场景开发有针对性的工具。经验研究——例如对分布式系统中错误处理的分析——往往比追求不可能实现的通用验证器更具实用价值。只要承认复杂度是根本性的约束，并把精力投入到我们所构建软件中那些具体、可衡量的属性上，就能在不追求完美的前提下稳步提升可靠性。 Writing correct software is fundamentally difficult because the nature of computation itself imposes unavoidable constraints. Theoretical computer science, rooted in the foundational work of Gödel, Church, and Turing, established that we cannot fully understand or predict the behavior of arbitrary programs. The halting theorem proves that no algorithm can determine if a program will eventually terminate, and its corollaries, such as Rice's theorem, demonstrate that deciding any non-trivial property of a program's behavior is similarly impossible. These results indicate that programs and the complete understanding required for perfect correctness are essentially at odds.

Computational complexity adds another layer of reality to these limitations. It is not enough for a problem to be decidable in theory if the resources required to solve it exceed reasonable bounds. Even when problems are technically solvable, the time and effort required often grow to levels that are practically insurmountable. As researchers like Juris Hartmanis and Richard Stearns observed, complexity functions as a law of nature. Results show that we cannot analyze a program's behavior faster than the program itself runs. Consequently, verification, viewed as a decision problem, is among the hardest challenges in computer science, and correctness cannot simply be broken down into modular, manageable components.

Abstractions in programming languages, such as high-level structures, functions, and concurrency, do not magically bypass these core difficulties. While these features make programs easier for humans to write, they do not simplify the underlying state space in a way that makes verification generally feasible. Even for finite state machines, where verification is theoretically possible, the problem is often PSPACE-complete. This means that programs with succinct representations are not necessarily easier to verify than their more verbose counterparts. The state explosion problem remains a significant hurdle for automated model-checkers, which must grapple with the sheer, often irregular complexity of real-world software.

Various approaches to verification, including model checking, deductive proofs, and static analysis, each operate within these rigid boundaries. Techniques like abstract interpretation or type systems attempt to make verification tractable by sacrificing precision, providing sound but incomplete results. While these tools assist in identifying specific classes of bugs, they cannot guarantee total correctness for all programs. The attempt to find a single, universal solution to software reliability is essentially modern alchemy. As Fred Brooks noted, there are no silver bullets in software engineering, and we must accept that complexity is an essential feature of the systems we build.

Ultimately, the most pragmatic path forward is to embrace an empirical, ad hoc approach similar to the physical sciences. Since we cannot verify all programs, we should study the specific programs humans actually write, document the common patterns of failure, and develop targeted tools for those scenarios. Empirical studies, such as the analysis of error-handling in distributed systems, offer far more practical value than the pursuit of impossible universal verifiers. By accepting that complexity is a fundamental constraint and focusing on specific, measurable properties of the software we create, we can make steady progress in improving reliability, even if perfection remains out of reach.

已发现一个关键同步漏洞，编号为 CVE-2026-119618 。该漏洞存在于 Apache HTTP Server 的 HTTP/2 模块中，具体位于 mod_http2 组件。该高严重性问题由 Anthropic 和 Google Cloud 的研究人员发现，根源在于 assign_headers 函数内部的并发执行竞态条件。由于对共享资源的处理不当，攻击者可能借此造成堆内存损坏，从而在远程且无需身份验证的情况下实现远程代码执行。若利用成功，攻击者将能够以与 Web 服务器进程相同的权限执行任意命令。

该风险影响范围较大，尤其是在云环境中，已被入侵的 Apache 实例可能成为更大规模攻击的突破口。由于此类实例通常承载容器化或虚拟化工作负载，它们可作为横向移动的切入点，进而暴露敏感的元数据端点或相邻的微服务。根据 Wiz 的数据，目前约 64% 的云环境包含易受该特定 CVE 影响的资源。此外，约 51% 的云环境被发现存在公开暴露的实例——尽管目前尚无可用漏洞利用手段，也缺乏野外正在发生攻击的证据，但这会显著增加外部被利用的可能性。

该漏洞影响 Apache HTTP Server 版本 2.4.55 至 2.4.70，以及独立的 mod_h2 模块版本 2.0.8 至 2.0.42 。为修复该问题，建议安全与工程团队将 Apache HTTP Server 部署升级至 2.4.71 或更高版本，并将所有独立 mod_h2 模块更新到 2.0.42 之后的版本。上述更新将针对 assign_headers 函数中发现的同步缺陷进行修补。

对于因运行条件限制而无法立即完成打补丁的组织，可采用临时缓解措施。团队可以通过禁用 HTTP/2 支持并将协议限制为 HTTP/1.1 来移除攻击向量。具体做法是修改服务器配置文件（例如 httpd.conf），确保只启用 HTTP/1.1 。尽管这并非永久修复方案，但在安排正式升级之前，它能够有效保护环境，防止竞态条件带来的风险。 A critical synchronization vulnerability, identified as CVE-2026-119618, has been discovered within the Apache HTTP Server's HTTP/2 module, specifically within the mod_http2 component. Discovered by researchers from Anthropic and Google Cloud, this high-severity flaw stems from a concurrent execution race condition inside the assign_headers function. When handling shared resources improperly, this vulnerability allows for heap memory corruption, which a remote, unauthenticated attacker could potentially leverage to achieve remote code execution. If successful, such an exploit would grant an unauthorized party the ability to execute arbitrary commands with the same privileges as the web server process.

The scope of this risk is significant, particularly within cloud environments where compromised Apache instances can serve as beachheads for broader attacks. Because these instances often host containerized or virtualized workloads, they provide an entry point for lateral movement that can expose sensitive metadata endpoints or adjacent microservices. According to data from Wiz, approximately 64% of cloud environments currently contain resources vulnerable to this specific CVE. Furthermore, about 51% of cloud environments are found to have publicly exposed instances, heightening the potential for external exploitation despite there being no currently available exploit or evidence of active attacks in the wild.

The vulnerability affects Apache HTTP Server versions 2.4.55 through 2.4.70, as well as standalone mod_h2 module versions 2.0.8 through 2.0.42. To remediate this issue, security and engineering teams are advised to upgrade their Apache HTTP Server deployments to version 2.4.71 or newer, and to update any standalone mod_h2 modules to a version beyond 2.0.42. These updates specifically address the synchronization flaws found in the assign_headers function.

For organizations that cannot perform immediate patching due to operational constraints, a temporary mitigation strategy is available. Teams can eliminate the attack vector by disabling HTTP/2 support and restricting protocols to HTTP/1.1. This can be achieved by modifying the server configuration file, such as httpd.conf, to ensure only HTTP/1.1 is in use. While this is a workaround rather than a permanent fix, it effectively secures the environment against the race condition until a formal upgrade can be scheduled.

在 SIGBOVIK 2019 上发表的这项研究，回应了编程社区里一个异想天开却又始终挥之不去的问题：能否在墙面上制造出一滩"油漆飞溅"，它看起来像代码，但却并不构成一个有效的 Perl 程序。这个想法源自社交媒体上的讨论——主张要尽早向孩子教授职业编程技能，于是研究者展开了对抽象表现主义与 Perl 语法之间交叉点的实证探索。

为检验这一假设，研究采用实证方案：借助光学字符识别（OCR）软件，对一份包含一百张油漆飞溅图像的数据集进行分析。结果表明，尽管其中相当一部分飞溅物确实能够被解读为可执行代码，但"所有这些飞溅物都能成为有效 Perl"这一说法并不成立。具体来说，数据显示，被分析的飞溅物中有 93% 能成功解析为有效的 Perl 程序，剩下的极少数则无法满足该语言以"高度灵活"著称的语法要求。

论文中给出了这些发现的详尽例子，展示了如何把原始的视觉痕迹转化为可运行的代码。有些飞溅物经 OCR 处理后，会生成能被计算为诸如零这种简单数值的字符串；而另一些则会转化为字符字符串。作者也强调了这种解读本身带有主观色彩：当 OCR 设置可能产生多种候选时，研究者会用审美判断挑选出更有趣的那一个程序。

这些结果已作为 2019 SIGBOVIK 会议论文正式归档。该会议以用轻松有趣的方式推进计算机科学研究而闻名。作者还附上了补充材料，例如飞溅物及其对应代码的完整画廊，便于他人独立验证解析结果。就算在最初提交截止日期之后，研究者仍继续补充分析，又找到了更多不符合标准的飞溅示例，进一步证实这种不寻常的"语言交叉"现象依然引人入胜。 The research presented at SIGBOVIK 2019 tackles a whimsical but persistent question in the programming community: whether it is truly possible to create a paint splatter on a wall that does not constitute a valid Perl program. This inquiry emerged from a social media discussion regarding the early introduction of vocational programming skills to children, leading to an empirical investigation into the intersection of abstract expressionism and Perl syntax.

To test this hypothesis, the study utilized an empirical approach involving optical character recognition (OCR) software to analyze a dataset of one hundred paint splatter images. The findings confirm that while a significant portion of these splatters can indeed be interpreted as executable code, the assertion that all such splatters are valid Perl is incorrect. Specifically, the data indicates that 93% of the analyzed splatters successfully parse as valid Perl programs, leaving a small minority that defy the language's notoriously flexible syntax.

The paper provides detailed examples of these findings, showcasing how raw visual input can be transformed into functional code. Some splatters, when processed by OCR, result in strings that evaluate to simple numerical values like zero, while others translate into character strings. The author highlights the subjective nature of this interpretation, noting that when OCR settings yield multiple possibilities, aesthetic judgment was used to select the most interesting program.

These results were officially documented as part of the 2019 SIGBOVIK proceedings, a conference known for its playful approach to computer science research. The author included supplementary materials, such as a complete gallery of the splatters and their corresponding code, to allow for independent verification of the parsing results. Even after the initial submission deadline, further analysis revealed additional examples of non-standard splatters, confirming the persistent intrigue of this unusual linguistic intersection.

英伟达董事会目前正准备罢免创始人兼 CEO Jensen Huang 的领导职务。这一重大的企业行动，源自公司对这位首席执行官开展的一项内部调查，调查内容与他卷入两起婚外情有关。公司尤其担忧的是，据称其中一名相关人员是中国间谍。

据报道，公司高层计划最早在本周三召开会议，敲定关于 Huang 未来去向的最终决定。若董事会继续推进罢免，他的离任将为一个由他领导的时代画上戏剧性的句号。 Huang 在推动公司穿越全球人工智能热潮、巩固其在科技行业以及国际地缘政治舞台上的重要地位方面发挥了关键作用。

他可能的离职，让全球最有价值的公司之一面临领导层不稳定的时期。此举凸显了公司调查结论的分量，尤其是涉及国家安全敏感领域以及企业监管方面的内容。行业目前正拭目以待：英伟达将如何应对这一过渡，并选出继任者来接替这位与公司成长紧密相连的高管。 The board of directors at Nvidia is currently preparing to remove founder and CEO Jensen Huang from his leadership position. This significant corporate action follows an internal investigation into the chief executive regarding his involvement in two separate extramarital affairs. Of particular concern to the company is that one of the individuals involved in these relationships is alleged to be a Chinese spy.

Company officials are reportedly planning to convene as early as this Wednesday to finalize a decision regarding Huang's future at the firm. Should the board proceed with his ouster, it would represent a dramatic conclusion to an era defined by his leadership. Huang has been instrumental in steering the company through the global artificial-intelligence boom, cementing his status as a prominent figure in the technology industry and on the international geopolitical stage.

His potential departure leaves one of the world's most valuable corporations facing a period of leadership instability. The move underscores the gravity of the findings from the company's investigation, particularly as they relate to national security sensitivities and corporate oversight. The industry now waits to see how Nvidia will navigate this transition and choose a successor to replace the executive who has been synonymous with the company's growth.

GitHub Actions 已正式支持 workflow lockfiles，这标志着开发者保障 CI/CD pipelines 的方式出现了重大变化。过去的通用安全建议是把 actions 固定到特定的 SHA 哈希上，以规避可变标签带来的风险，因为这些标签可能会被恶意方篡改。借助这一新能力，开发者既能继续使用更清晰、更易读的标签，又能确保底层代码始终锁定在经过验证的版本上。

要使用它，开发者可以在 GitHub CLI 中安装 gh-actions-lock 扩展。安装完成后，运行相应命令会自动把 workflows 更新为使用版本标签，并在代码库中生成一个集中式的 actions.lock 文件。该文件作为单一事实来源：它会将特定的 actions 映射到对应的提交 commit SHA 、 owner ID 以及 repository ID 。完成首次锁定后，只需把这些改动提交到代码库，就能最终落地安全配置。

依赖管理仍然很省事，因为 Renovate 和 Dependabot 已经都支持这些 lockfiles 。这意味着自动化的依赖更新会同时包含对 lockfile 的必要调整，只要用户遵循依赖冷却时间等最佳实践即可。此外，强烈建议在 CI 中校验这些文件的完整性。通过在 workflow 中新增专门的校验任务，你可以确保 lockfile 始终与当前配置保持一致，并且不会被破坏或篡改。

与此同时，安全性检查工具 zizmor 也已更新：在最新版本中支持 lockfiles 。把 zizmor 集成到 pre-commit hooks 或 CI pipeline 中，可以为 GitHub Actions 配置增加一层额外的自动化安全分析。将这些现代锁定机制与成熟的安全 lint 工具结合起来，开发者能够显著降低供应链攻击风险，从而让自动化流程既保持最新，也更安全。 GitHub Actions has officially introduced support for workflow lockfiles, marking a significant shift in how developers can secure their CI/CD pipelines. Historically, the standard security advice was to pin actions to specific SHA hashes to avoid the risks associated with mutable tags, which could be compromised by malicious actors. With this new feature, developers can maintain cleaner, human-readable tags while ensuring that the underlying code remains locked to a verified version.

To take advantage of this, developers can install the gh-actions-lock extension for the GitHub CLI. Once installed, running the command automatically updates workflows to use version tags and generates a central actions.lock file in the repository. This file acts as a source of truth, mapping specific actions to their corresponding commit SHAs, owner IDs, and repository IDs. After performing the initial lock, users simply need to commit these changes to their repository to finalize the security configuration.

Managing these dependencies remains straightforward as both Renovate and Dependabot have already added support for these lockfiles. This means automated dependency updates will include the necessary adjustments to the lockfile, provided users incorporate best practices like dependency cooldowns. Furthermore, it is highly recommended to validate the integrity of these files within CI. By adding a dedicated verification job to your workflow, you can ensure that the lockfile is always in sync with your configurations and not corrupted.

Complementing this, the security linter zizmor has been updated to support lockfiles in its latest versions. Integrating zizmor into your pre-commit hooks or CI pipeline provides an extra layer of automated security analysis for your GitHub Actions configurations. By combining these modern locking mechanisms with established security linting tools, developers can significantly reduce the risk of supply-chain attacks, ensuring their automation remains both current and secure.

Failed to crawl the webpage. (English version not available)

United Airlines 已正式宣布：禁止使用人工智能代理代替旅客预订航班。该航空公司指出存在严重的安全风险，并警告称，如果把个人凭据和支付信息交给这些数字助手，就会产生漏洞。据其介绍，这类第三方工具可能会导致未经授权的账户变更、意外消费，或滥用 MileagePlus 里程奖励积分。

在安全之外，United 还强调，使用 AI 执行诸如值机等任务，可能会触犯联邦法规。该航空公司举例称，旅客需要亲自确认行李中不含违禁物品，而这类要求可能会让自动代理在无意间违反常规的航空安全流程。为此，United 正在更新预订和售票政策，限制这些第三方代理访问其数字平台。

业内人士认为，此举在一定程度上也是 United 为了继续掌控客户预订体验所采取的策略。通过切断这些"机器人"对平台的访问权限，航空公司希望避免更强大的自动化中介出现，从而进一步加剧其分销渠道的复杂度。同时，这也有明显的成本考量：航空公司希望避免因自动化机器人驱动的海量航班搜索激增而产生费用，但这些搜索未必能转化为实际完成的交易。

尽管作出调整，该航空公司仍表示，人类旅客的预订流程依旧简单。乘客仍可通过 Hopper 、 Kayak 和 Google Flights 等传统航班聚合平台获取票价。 United 虽然不再允许使用自动化代理，但并不预计这会影响乘客找到有竞争力价格的能力。

更广泛的航空业对于如何应对这类技术依然存在分歧。尽管 United 正在限制相关工具，其他航司（如 Southwest）则采取更为宽松的态度，允许用户通过 ChatGPT 等 AI 应用直接查看航班库存。眼下，United 对新政策保持坚定立场，把内部安全与运营监管放在由 AI 自动化带来的便利之上。 United Airlines has officially announced a ban on the use of artificial intelligence agents for booking flights on behalf of travelers. The carrier cited significant security concerns, warning that providing these digital assistants with personal credentials and payment information creates vulnerabilities. According to the airline, such third-party tools could lead to unauthorized account changes, unintended purchases, or the misuse of MileagePlus reward points.

Beyond security, United emphasized that utilizing AI for tasks like checking in for a flight could potentially run afoul of federal regulations. The airline pointed to rules requiring travelers to verify personally that their luggage is free of banned items as one example of where automated agents might inadvertently violate standard aviation safety protocols. In response, United is updating its booking and ticketing policies to restrict these third-party agents from accessing its digital platforms.

Industry experts suggest this move is partly a strategic effort by United to maintain control over the customer booking experience. By cutting off access for these bots, the airline aims to prevent the rise of powerful, automated intermediaries that could further complicate their distribution channels. There is also a financial incentive, as the airline seeks to avoid the costs associated with the surge in automated bot-driven flight searches that do not necessarily lead to completed transactions.

Despite this change, the airline maintains that the booking process for human customers remains straightforward. Travelers can still secure fares through traditional flight aggregators such as Hopper, Kayak, and Google Flights. While United is shutting down the use of automated agents, it does not expect this to hinder the ability of passengers to find competitive pricing.

The broader aviation industry remains divided on how to approach this technology. While United is moving to restrict these tools, other carriers like Southwest have taken a more permissive stance, allowing users of platforms like ChatGPT to view their flight inventory directly through the AI app. For now, United is standing firm on its new policy, prioritizing its internal security and operational oversight over the convenience of AI-driven automation.

像 Claude Code 这样复杂的 AI 编程助手的出现，从根本上改变了软件开发的格局，让我们离开了手动、基于规则的编程时代。几十年来，数字化时代依赖二进制逻辑的精确性，以及人类程序员的耐心劳动——他们从零开始，艰难地构建系统。然而今天，这种对指令的僵化依赖正在被随机模型取代：这些模型可以在几分钟内处理海量且边界不明的任务。过去与掌握复杂开发工作流相关的那份自豪感正在消散，比如 Emacs 那些错综复杂的命令序列。因为一度只属于专家用户的效率红利，如今已向任何会向 AI 发出提示的人开放。

这种变化在所谓"God machine"的兴起中体现得尤为明显：资深编码者的竞争优势，例如快速编辑或自定义配置，正逐渐失去价值。就连 Magit 驱动的 Git commit 这类成熟仪式，也被精简到了只剩下"手动介入显得老派"的程度——AI 驱动的自动化让人工操作看起来仿佛是过时的遗物。与此同时，这种转变让新一代用户得以追逐那些以前会被认为完全不合理或在计算上几乎不可能的项目想法，只因为生成复杂代码的成本已经骤降到几枚 token 的价格。

excali-mode 和 cooked.el 是这种新型、近乎疯狂的开发方式的两个具体案例。以 excali-mode 为例，一位开发者通过强制使用新的 canvas API，对复杂层进行逐一对应替换，从而把整个基于 React 的绘图应用成功移植进 Emacs 环境。类似地，cooked.el 的诞生也表明：一位业余开发者仅用数小时，就用 Rust 实现了一个完整的终端模拟器——而在以往，这样的任务往往需要专业团队耗费数年才能完成。这些项目说明了一种现象：用户在"只想做个操纵杆"的情况下，实际上却像是在"建造自己的 Boeing"。他们忽视传统的软件架构，转而用个性化、一次性的黑客手段来达成目标。

这种转变冲击了软件设计的长期传统：过去人们更看重可维护性和架构上的连贯性。我们正在看到所有软件都走向"Emacsification"：个体用户要求工具完全按他们的具体偏好来定制，而不是接受标准化、面向大众市场的发布。新创造者的价值观几乎全都落在"展示与演示（show and tell）"上，而不是产品是否具备可行性。因此，当搭建、拆除并重写整个应用程序栈的成本在本质上被压到接近于零时，关于浪费与复杂性的传统观念也就失去了意义。

这场技术演进的讽刺在于，它一边实现、另一边又颠覆了软件自由的最初目标。用户如今拥有前所未有的能力，可以随心所欲地改造自己的数字环境，但底层机制却变得愈发隐晦不明。我们正迅速逼近一个未来：AI 可能最终能够直接与硬件交互——这几乎等同于遗传密码中孕育出生命的那一跃。我们或许能实现把数字想象力不受限制地呈现出来的终极自由，但代价是：我们失去了真正理解机器如何运转的能力。 The emergence of sophisticated AI coding assistants like Claude Code has fundamentally shifted the landscape of software development, moving us away from the era of manual, rules-based programming. For decades, the digital age relied on the precision of binary logic and the patient labor of human programmers who painstakingly built systems from the ground up. Today, however, that rigid adherence to instruction is being eclipsed by stochastic models that can handle vast, unbounded tasks in minutes. The traditional pride associated with mastering complex development workflows, such as the intricate command sequences of Emacs, is evaporating because the efficiency gains once reserved for expert users are now available to anyone capable of prompting an AI.

This transition is highlighted by the rise of what can be called the "God machine," where the competitive advantages of experienced coders, such as rapid editing or custom configuration, are becoming obsolete. Even established rituals like Git commits via Magit have been streamlined to the point where AI-driven automation makes manual intervention seem antiquated. This change has empowered a new generation of users to pursue project ideas that previously would have been considered entirely irrational or computationally prohibitive, simply because the cost of generating complex code has plummeted to the price of a few tokens.

Two specific examples, excali-mode and cooked.el, serve as case studies for this new, quixotic approach to development. In the case of excali-mode, a developer was able to port an entire React-based drawing application into the Emacs environment by forcing a one-for-one substitution of complex layers using a new canvas API. Similarly, the creation of cooked.el saw an amateur developer successfully implement a full-blown terminal emulator in Rust in mere hours, a task that would have historically required a team of professionals years to complete. These projects illustrate a phenomenon where users effectively "build their own Boeing" when they only intended to create a flight stick, ignoring traditional software architecture in favor of personalized, throwaway hacks.

This shift challenges the longstanding traditions of software design, which once prioritized maintainability and architectural coherence. We are seeing a move toward the "Emacsification" of all software, where individual users demand tools tailored to their precise preferences rather than accepting standardized, mass-market releases. The ethos of these new creators is centered entirely on "show and tell" rather than product viability. Consequently, the traditional notions of wastefulness and complexity lose their meaning when the effort required to build, tear down, and rewrite entire application stacks is effectively reduced to zero.

The irony of this technological evolution is that it simultaneously fulfills and subverts the original goals of software freedom. While users now possess the unprecedented ability to bend their digital environments to their exact whims, the underlying mechanisms have become opaque. We are rapidly approaching a future where AI might eventually interface directly with hardware, a leap comparable to the emergence of biological life from genetic code. We may achieve the ultimate freedom to manifest our digital imaginations without restriction, but in doing so, we have sacrificed the ability to truly understand how our machines function.