尽管 Large Language Models 在生成语法正确的代码方面已非常娴熟，但它们的普及也带来了一个隐性的代码质量危机。即便代码能通过自动化测试，往往仍存在过度抽象、冗余重复和糟糕的架构选择。此类问题通常被称为 slop，会导致代码行数失控式增长。随着项目每月新增数百万行，人类开发者已难以保持监督，而与常见观点相反，AI agents 目前也无法自我修正这种技术债务的累积。

目前业界对这一问题的评估过于依赖主观"感觉型"指标，难以产出有价值的数据。用 AI 来审查代码质量并不可靠：模型常常给出不一致的评分，或受命名等表面因素影响而改变偏好。尽管人工评估仍是保证可读性的金标准，但它缺乏现代训练流程和大规模基准所需的可扩展性。

为应对这一难题，研究者开始采用定量指标来区分 legacy codebases 与由 AI 生成的 slop 。两个较有前途的度量是 verbosity（通过启发式方法检测重复或不必要的冗长片段）和 erosion（衡量系统复杂度在多大程度上集中在少数过大、密集的函数中）。这些指标揭示了一个残酷事实：AI-generated code 在冗长度和侵蚀程度上约为 human-authored software 的两倍。

Agents 无法应对这种复杂性的现象在 SlopCodeBench 等基准中被凸显出来：这些基准通过在检查点间擦除模型上下文来模拟真实的迭代开发过程。在糟糕决策随时间累积的情形下，最先进的模型往往无法在多轮迭代中维持一个可用的 codebase 。对于那些在缺乏充分监督下大量引入 machine-generated code 的团队来说，这一失败是严重的警示。

归根结底，衡量代码质量仍离不开人类的直觉与审美。定量指标虽可帮助识别问题规模，但仅是迈向软件可维护性的一步。今后对 code churn 、 function cohesion 和 system coupledness 等指标的研究，将对完善这些工具至关重要。随着领域的发展，关注点必须从单纯生成代码，转向营造一个把结构完整性与功能性同等重视的开发环境。 While Large Language Models have become remarkably adept at generating formally correct code, their rise has ushered in a hidden crisis of code quality. Even when code passes automated tests, it often suffers from excessive abstraction, redundant duplication, and poor architectural decisions. This phenomenon, often termed slop, leads to an uncontrolled explosion of lines of code. As projects grow by millions of lines per month, human developers lose the ability to maintain oversight, and contrary to popular belief, AI agents are currently incapable of self-correcting this accumulation of technical debt.

The industry's current approach to evaluating this problem relies heavily on subjective, vibes-based metrics that fail to provide meaningful data. Using AI as a judge to grade code quality is largely ineffective, as models frequently produce inconsistent results or change their preferences based on superficial factors like naming conventions. Conversely, relying on human evaluation is the gold standard for maintaining readability, but it lacks the scalability required for modern training workflows and large-scale benchmarking.

To address this, researchers are turning to quantitative metrics to distinguish between legacy codebases and AI-generated slop. Two particularly promising measures are verbosity, which tracks duplicated or unnecessarily verbose segments using heuristics, and erosion, which calculates how much of a system's complexity is concentrated within a few oversized, dense functions. These metrics reveal a stark reality, as AI-generated code consistently proves to be roughly twice as verbose and eroded as human-authored software.

The inability of agents to manage this complexity is highlighted by benchmarks like SlopCodeBench, which simulate real-world, iterative development by erasing model context between checkpoints. Under these conditions, where bad decisions accumulate over time, state-of-the-art models often fail to maintain a functional codebase across multiple iterations. This failure serves as a critical warning for teams currently integrating vast amounts of machine-generated code into their systems without adequate oversight.

Ultimately, measuring code quality remains deeply tied to human intuition and taste. While quantitative metrics offer a way to identify the scale of the problem, they are only the beginning of a larger effort to ensure software remains manageable. Future explorations into code churn, function cohesion, and system coupledness will be essential to refining these tools. As the field evolves, the focus must shift from simply generating code to fostering an environment where structural integrity is treated with the same importance as technical functionality.

据报道，Iran-backed Houthi movement 已控制了 Perim Island，这座位于 Bab al-Mandab Strait 的小岛具有重要战略价值。该狭窄海峡连通 Red Sea 与 Gulf of Aden，是亚洲与欧洲之间的重要航运通道。夺岛属于其快速且大规模的军事攻势的一部分：此次行动已使 Houthi 控制了 Yemen 沿 Red Sea 的全部海岸线，迫使 Saudi-backed forces 撤退，并在一周内造成至少 46,000 人流离失所。

尽管 Houthi 领导层宣称行动取得成功并坚称国际海上通行安全，但他们同时明确表示对 Saudi 船只实施封锁。此举在 Riyadh 引发严重担忧：在 Strait of Hormuz 受阻后，Saudi Arabia 对经由 Red Sea 的石油出口通道愈发依赖。分析人士认为，Houthi 通过打击与 Saudi 有关的航运、同时向更广泛的国际社会释放安抚信号，可能是在有意向 Saudi Arabia 施压，同时尽量避免触发 Washington 的直接军事回应。

据报道，Saudi Crown Prince Mohammed bin Salman 已向 U.S. 请求直接军事干预，但 United States 拒绝直接出兵，只愿提供情报与目标指引支持，并将重心放在维护核心国家安全利益与保障航行自由上。 U.S. 官员称，他们正与地区盟友保持定期沟通，力图在应对这场升级危机与避免开辟新战线之间取得平衡。

这一冲突的战略重要性在 Djibouti 附近的重兵部署中凸显出来：包括 United States 、 China 和 France 在内的多国在当地设有基地，以监控并保护航道。随着 Yemen 沿海政府阵地突然崩溃、 Houthi 在关键通行区域站稳脚跟，观察人士警告局势高度不稳定。国际利益相关者最担忧的是全球能源市场可能进一步遭到冲击——若 Houthi 对这些海上咽喉的控制持续扩大，冲突将如何演变仍难以预测。 The Iran-backed Houthi movement has reportedly seized control of Perim Island, a strategically vital location situated within the Bab al-Mandab Strait. This narrow waterway serves as a critical gateway between the Red Sea and the Gulf of Aden, acting as an essential shipping corridor for trade between Asia and Europe. The capture of the island comes as part of a rapid, large-scale military offensive that has seen the Houthis take control of Yemen's entire Red Sea coast, forcing a retreat of Saudi-backed forces and displacing at least 46,000 people in just one week.

While the Houthi leadership claims that their military operations are a success and insists that international maritime navigation remains safe, they have explicitly stated that Saudi vessels are subject to a blockade. This move has caused significant alarm in Riyadh, as Saudi Arabia has become increasingly reliant on Red Sea routes for oil exports following the disruption of the Strait of Hormuz. Analysts suggest that by targeting Saudi-linked shipping while attempting to reassure the broader international community, the Houthis may be executing a calculated effort to pressure Saudi Arabia without triggering a direct military response from Washington.

In response to these developments, Saudi Crown Prince Mohammed bin Salman has reportedly sought direct U.S. military intervention. However, the United States has declined to engage directly, offering instead to provide intelligence and targeting support while maintaining a focus on protecting core national security interests and ensuring the freedom of navigation. U.S. officials maintain that they are in regular communication with regional allies, aiming to balance the management of this escalating crisis with an ongoing effort to avoid opening a new military front.

The strategic stakes of this conflict are underscored by the heavy military presence in nearby Djibouti, where several global powers, including the United States, China, and France, maintain bases to monitor and protect shipping lanes. With the sudden collapse of government-held positions along Yemen's coast and the Houthi entrenchment in key transit areas, observers warn that the situation remains highly volatile. The potential for further disruption to global energy markets remains a primary concern for international stakeholders, as it is difficult to predict how the conflict may evolve if Houthi influence over these vital maritime chokepoints continues to expand.

"The Waymo effect"描绘了我们与技术以及彼此互动方式的一种悄然变化。通过消除与他人打交道的摩擦，技术带来一种看似纯粹的收益感。例如，我们欣赏无人驾驶汽车的便利，因为它让我们绕过人类互动中不可避免的寒暄与社交摩擦。但这种便利也有隐性代价：我们失去了那些非自选且往往不可预测的与陌生人相遇的机会，而正是这些相遇能拓宽视野，把我们与自身信息泡沫之外的世界连接起来。

这种动力正在逐步重塑学术与科学研究的格局。正如无人驾驶汽车一样，大型语言模型为人类合作者提供了一种无摩擦、随时可得的替代方案。同行可能带来令人不便的议程、分歧或拒绝配合你的前提，但人工智能助手则完全顺从、随时可用、不需妥协。然而，人类合作者的"麻烦"恰恰往往蕴含真正价值：正是在这种摩擦中，想法被挑战、假设被检验、意外的机遇才会出现。当研究者用聊天机器人的无缝产出取代人类之间那种混乱且不可预测的协作过程时，就存在一种协作瓦解的风险，这会威胁到研究共同体的社会结构基础。

现代学术界的激励机制进一步加剧了这一倾向。机构与资助体系在很大程度上更看重速度与可衡量的产出，使得在许多情况下选择机器而非人类成为最理性的决策。协作本身耗时且昂贵，涉及出差、建立信任以及管理参与者的自尊心。当预算吃紧、职业晋升依赖于迅速发表时，缓慢而深入的人类协作很容易被边缘化。结果是，个人生产力可能上升，但思想的集体多样性却缩水——人人依赖那些能产出流畅、结构化但往往趋同结论的工具。

此外，写作不仅是产出文本的行为，更是思考的重要组成部分。写作像一种强制机制，把模糊的直觉锻造成精确的论点，揭示出可能被忽略的逻辑漏洞。把写作过程外包给 AI，研究者可能无意中跳过那些困难而必要的深度思考，最终用表面光鲜、无摩擦的模仿替代了扎实的理解。危险在于，学术文化表面上看起来健康且高效，而支撑它的关键而无形的智力劳动却在悄然流失。

归根结底，这并不是要拒绝新技术——它们确实带来创新与效率的真实潜力。相反，这是提醒我们正视一个问题：当机器驱动的工作流程变得廉价、而以人为驱动的协作变得非常昂贵时，系统本身存在风险。若要维护研究的质量与深度，就必须有意识地把摩擦重新引入体系：将非结构化的人际互动视为关键的基础设施，奖励研究者的协作贡献，并确保人类牢牢掌握主导地位。在论文撰写能力变得普及之后，真正稀缺且有价值的，将是共同思考的能力。 The Waymo effect describes a quiet shift in how we interact with technology and each other. By removing the friction of dealing with other people, technology offers an experience that feels like pure gain. We appreciate the convenience of the driverless car, for example, because it allows us to bypass the small talk and social friction inherent in human interaction. Yet, this convenience comes at a hidden cost. We lose the unchosen, often unpredictable encounters with strangers that broaden our perspectives and connect us to a world outside our own bubbles.

This dynamic is increasingly shaping the landscape of academic and scientific research. Much like the driverless car, large language models provide a frictionless, always-available substitute for the human collaborator. While a colleague might arrive with inconvenient agendas, disagreements, or the refusal to align with your specific assumptions, an AI assistant is perfectly compliant, available at any hour, and requires no compromise. However, the very inconvenience of a human collaborator is often where the real value lies. It is through this friction that ideas are challenged, assumptions are tested, and serendipity occurs. When researchers trade the messy, unpredictable process of human collaboration for the seamless output of a chatbot, they risk a process of decollaboration that threatens the underlying social fabric of the research community.

The incentive structures within modern academia further exacerbate this trend. Institutions and funding systems largely prioritize speed and measurable output, creating a environment where choosing a machine over a human is often the most rational decision. Collaboration is inherently time-consuming and expensive, involving travel, trust-building, and the management of egos. When budgets are tight and career advancement depends on rapid publication, the slow, human work of deep collaboration is easily sidelined. This creates a trap where individual productivity may rise, but the collective diversity of ideas narrows, as everyone relies on tools that provide fluent, structured, yet often convergent results.

Furthermore, we must recognize that writing is not just the production of an artifact, but an essential component of thinking. It serves as a forcing function that turns vague intuitions into precise arguments, revealing holes in logic that might otherwise go unnoticed. By outsourcing the writing process to AI, researchers may inadvertently skip the difficult, effortful thinking required for genuine insight. This transition risks replacing durable understanding with a polished, frictionless imitation. The danger is that research culture could appear healthy and efficient on the surface while the vital, invisible intellectual labor that sustains it slowly erodes.

Ultimately, this is not an argument for rejecting new technologies, which offer genuine potential for innovation and efficiency. Instead, it is a call to recognize the risks of a system that makes machine-driven workflows cheap and human-driven collaboration prohibitively expensive. If we wish to preserve the quality and depth of research, we must deliberately engineer friction back into the system. This means valuing unstructured, human interaction as a form of critical infrastructure, rewarding researchers for their collaborative contributions, and ensuring that humans remain firmly in the driver's seat. As the ability to produce papers becomes common, the truly scarce and valuable resource will be the ability to think together.

RTK，即 Rust Token Killer，是一款通过过滤和压缩终端输出以降低 AI 编码成本的工具，近来颇为流行。尽管有许多病毒式传播的帖子声称启用 RTK 可将 Token 使用量减少高达 90%，但这类说法常把终端数据量的减少等同于实际费用的节省。对 Claude Code 和 DeepSeek V4 Pro 等模型进行了超过 1,700 次的严格基准测试后，结果表明 RTK 并不可靠地降低总体费用，甚至在某些情况下会提高成本。

测试采用 Terminal-Bench 2.1 在终端交互频繁的场景中评估工具表现。结果显示，在使用 Fable 5.0 的 Claude Code 的某个特定场景下，成本仅略降约 5%；而在 DeepSeek 上，成本反而上升了 5% 。更重要的是，全部任务的平均值显示，Fable 并未呈现明确的成本优势，而 DeepSeek 的开销平均上升了 17% 。可见，RTK 报告的"收益"是基于丢弃的字节数而非计费的 Token 数，这一指标容易误导对实际财务节省的判断。

测试中还发现 RTK 容易引入技术性错误。一次插件故障导致代理进入死循环，出现了 339 次连续错误，使得总成本大约是基线的九倍。此外，现代 AI 代理通常已经具备管理输出的手段（例如使用 head 或 tail 命令），因此来自终端输出的账单占比本就不高。 RTK 的压缩效果常被抵消，因为压缩终端输出后，代理往往需要额外的交互轮次才能完成任务，从而抹去了潜在的节省。

综上，研究认为 RTK 更像是一种针对特定场景的小众优化工具，而非普适的 AI 编码成本解决方案。代理式编码模型高度依赖缓存读取，处理原始终端输出的成本本就比开发者预想的低得多。鉴于额外的代理轮次代价较高，使用压缩工具通常难以带来净收益。随着前沿模型在终端交互处理方面变得越来越高效，RTK 在普遍意义上的成本下降效果仍未得到证实。 RTK, or Rust Token Killer, has gained significant popularity as a tool designed to reduce AI coding costs by filtering and compressing terminal output. While many viral posts and claims suggest that implementing RTK can cut token usage by up to 90%, these projections often conflate reduced terminal volume with actual cost savings. After conducting a rigorous benchmark test across more than 1,700 attempts using models like Claude Code and DeepSeek V4 Pro, the findings indicate that RTK does not reliably lower total expenses and can, in some cases, increase them.

The testing process utilized Terminal-Bench 2.1 to evaluate how the tool performs in environments with heavy terminal interaction. Results showed that for Claude Code using Fable 5.0, costs decreased by a marginal 5% in one specific scenario, while for DeepSeek, costs actually increased by 5%. More importantly, when averaged across all tasks, Fable saw no clear cost benefit, and DeepSeek's expenses rose by 17% on average. These results demonstrate that the tool's reported gain—which measures discarded bytes rather than billed tokens—is a misleading metric for estimating financial savings.

A significant issue identified during testing is the propensity for RTK to introduce technical errors. In one instance, a plugin bug caused an agent to enter an infinite loop, resulting in 339 consecutive errors and a total cost roughly nine times higher than the baseline. Furthermore, because modern AI agents already possess built-in techniques for managing output, such as using head or tail commands, the actual proportion of an agent's bill derived from terminal output is relatively small. The compression offered by RTK is frequently offset by the fact that agents often require additional turns to complete a task when the terminal output is condensed, effectively erasing any potential savings.

Ultimately, the study concludes that RTK is a niche optimization tool rather than a comprehensive solution for reducing AI coding costs. Because agentic coding models rely heavily on cache reads, the cost of processing raw terminal output is already significantly lower than developers might assume. Since extra agent turns are costly, the trade-off of using a compression tool rarely results in a net gain. Given that frontier models are becoming increasingly efficient at handling terminal interactions on their own, the practical utility of RTK for general cost reduction remains unproven.

Claude 仅供年满 18 岁的用户使用，用户在创建账户时必须确认自己符合该年龄要求。平台使用自动化安全系统检测可能未满 18 岁的迹象；若系统怀疑有未成年人使用某账户并将其标记，访问权限会被限制，用户需先完成年龄验证后方可继续使用服务。

为便于验证，平台与独立第三方年龄验证服务商 Yoti 合作。账户被标记时，用户会收到一封包含验证链接的通知邮件；成功完成验证后，账户将恢复使用。

用户可通过 Yoti 选择多种方式确认年龄：可以选择面部年龄评估，通过自拍估算年龄，无需提交证件；也可以选择上传政府签发证件的照片进行 ID 验证，例如 Passport 或 Driver's License；已使用 Yoti Digital ID App 的用户则可直接从 App 分享已验证的 "over 18" 属性。

在数据隐私方面，Yoti 是一家经独立审计且符合 SOC2 的提供商，保障安全。验证流程旨在保护用户隐私——平台不会查看或存储任何身份证件或自拍，Yoti 在年龄核验完成后会立即删除所有个人数据、文档和图像。平台仅接收通过或未通过的结果，确保验证过程中不保留任何敏感个人信息。 Claude is strictly intended for users who are at least 18 years old, and individuals must confirm they meet this age requirement during the initial account setup process. The platform employs automated safety systems designed to detect indicators that a user might be under the age of 18. If the system flags an account due to suspected minor activity, access will be restricted, and the user will be prompted to verify their age before they can continue using the service.

To facilitate this verification, the platform partners with Yoti, an independent, third-party age verification provider. When an account is flagged, the user receives a notification email containing a link to begin the verification process. Successfully completing this process allows the user to have their account reinstated.

Users have multiple options for confirming their age through Yoti. They may choose facial age estimation, which uses technology to estimate their age from a selfie without requiring an ID document. Alternatively, they can opt for ID verification by uploading a photo of a government-issued document, such as a passport or driver's license. Finally, those who already use the Yoti digital ID app can share a verified "over 18" attribute directly from the app.

Regarding data privacy, Yoti acts as an independently audited, SOC2-compliant provider to ensure security. The verification process is designed to protect user privacy, as the platform does not see or store any ID documents or selfies. Yoti deletes all personal data, documents, and images immediately after the age check is completed. Consequently, the platform only receives a pass or fail result, ensuring that no sensitive personal information from the verification process is stored by the company.

该文档只是一个自动安全验证页面，用来防止恶意机器人流量，表明网站正在通过标准的 Cloudflare 安全检查来核实用户身份。

由于输入仅包含该技术验证通知、并不包含实际文章内容，因此无法就具体内容、论点或主题作出摘要。该页面只是通往目标站点的通道，并不包含实质性信息。 The provided document serves only as an automated security verification page used to protect the website from malicious bot activity. It indicates that the site is currently confirming the user's legitimacy through a standard Cloudflare security check.

Because the input consists exclusively of this technical verification notification and lacks the actual article text, it is not possible to generate a summary of specific content, arguments, or thematic details. The page serves as a gateway to the intended site rather than containing substantive information.

当前 AI 工程越来越呈现内卷化倾向：系统要求投入更多努力、产出更高竞争性成果，却并未真正提升价值或生产力。虽然像 GPT 6 Astra 这样的新模型在处理计算机操作、图像理解和复杂任务方面能力突出，但它们给实际的软件工程带来了显著挑战。模型似乎缺乏明确的约束或惩罚机制来抑制低质量代码的生成，因而对长期任务完成的过度关注反而催生出大量不可用或不可维护的产出。

为测试这些能力的极限，一项为期周末的自管理软件工厂实验暴露出令人担忧的模式。模型在被赋予完全自治权以管理上下文和子智能体时，于 35 小时内生成了 75,000 行代码，消耗了约十亿个 tokens（词元）。尽管活动量巨大，产出却几乎毫无价值。代码表现出许多怪异行为：在编辑 C 代码时依赖复杂的 Python 字符串拼接，用 Python 去执行简单的 shell 命令（本可直接调用），以及各种古怪难读、完全违背既有最佳实践的编码风格。

问题的根源似乎在于这些模型的训练目标。它们因词元效率和任务完成而被大量奖励，常采用在临时工具调用中可行但用于持久代码库时灾难性的"代码高尔夫"技巧。这导致可读性明显退化：模型优先选择紧凑、对机器高效的方案，而非便于人类理解的逻辑。在无人看管的情况下，这些智能体会呈现出递归、螺旋式的行为，生成越来越晦涩的任务结构，并将随机常量或任意逻辑混入生产级代码中。

这一趋势对 AI 在软件开发方向上的走向提出了根本性疑问。尽管模型快速演进，但其发展轨迹似乎与以人为中心的工程流程相悖。运行这些智能体的成本——无论是高昂的财务开销，还是需要的大量人类监督——都使其在专业软件工作中的合理性受质疑。看起来这些工具更像是在为其他领域（如 3D rendering 或通用计算自动化）进行优化，反而让专业开发者面临一个越来越难以审计与信任的产品。

归根结底，智能体被优化出来的产物与人类工程师所需之间的脱节在扩大。如果代码最终只成为其他智能体读写的媒介，负责长期维护的人类程序员就可能丧失对其的实用价值。只要这些模型把"完成任务"的原始得分置于连贯性与可维护性之上，它们很可能更多地成为生产"糟粕"的新奇工具，而非严肃软件工程中的可靠伙伴。 The current state of AI engineering is increasingly characterized by what can be described as Neijuan, or involution, where systems demand greater effort and competitive output without any real increase in value or productivity. While new models like GPT 6 Astra are undeniably impressive in their capacity to handle computer use, image understanding, and complex tasks, they present significant challenges for practical software engineering. These models seem to lack clear constraints or penalties for producing low-quality code, leading to an environment where the focus on long-horizon task completion results in the creation of vast amounts of unusable or unmaintainable material.

In an effort to test the limits of these capabilities, a weekend-long experiment running a self-managed software factory revealed concerning patterns. Given full agency to manage context and subagents, the model produced 75,000 lines of code over 35 hours while burning through roughly a billion tokens. Despite this immense activity, the output was effectively valueless. The code displayed strange behaviors, such as a reliance on convoluted Python string splicing for C code editing, unnecessary use of Python to execute simple shell commands, and bizarre, hard-to-read coding styles that deviate entirely from established best practices.

The core issue appears to stem from how these models are trained. They are heavily rewarded for token efficiency and completing tasks, often using "codegolfing" techniques that work well for temporary tool calls but are catastrophic when applied to permanent codebase development. This leads to a degradation of readability, as the models prioritize compact, machine-efficient solutions over human-understandable logic. When left to run unattended, these agents exhibit a recursive, spiraling behavior, creating increasingly obscure task structures and integrating random constants or arbitrary logic into production-level code.

This trend raises fundamental questions about the direction of AI in software development. While these models are evolving rapidly, their trajectory seems at odds with current human-centric engineering processes. The cost of running these agents, both in terms of financial expense and the necessity for intense human oversight, makes them difficult to justify for professional software work. It appears these tools are being optimized for other domains, such as 3D rendering or general computer automation, leaving the professional developer with a product that is increasingly difficult to audit or trust.

Ultimately, there is a growing disconnect between what these agents are optimized to produce and what human engineers require. If code becomes a medium intended solely for other agents to read and write, it may eventually lose all utility for the human programmers responsible for its long-term maintenance. As long as these models prioritize raw completion over coherence and maintainability, they will likely remain more of a curiosity for "slop" production rather than a reliable partner in serious software engineering.

16 岁的 Ángela Karime Venegas Hernández 是 Altamira, Tamaulipas 地区 CETIS 78 的一名学生，她开发出一种依靠声波而非传统化学品或水的创新灭火装置。她的灵感来自一个简单的好奇心：想在不吹气或不使用常规物质的情况下熄灭蜡烛。

为此，她设计了由 12 伏电池供电的装置，采用频率发生器和经过校准的扬声器，以每秒 30 次脉冲振动工作。该振动能有效将氧气从火焰周围推开，使火焰因缺氧在 5 到 8 秒内熄灭。

她将项目命名为 Vortex Tech，并通过严格测试验证了其多用途性。 Ángela 进行了 100 多次试验，确认装置对多种火灾类型均有效，包括木材、易燃液体、厨房油脂以及对电子设备敏感的火情。

除了速度和效果，这项发明还具有显著的环保和安全优势：通过声波灭火不会造成污染、不留化学残留，也不会对操作人员造成身体伤害。

这个起于学校的小实验如今已受到国际关注。凭借这一突破，Ángela 将代表 Mexico 参加国际科学博览会，展示她应对人类最古老且最持久危险之一的方法。 Sixteen-year-old Ángela Karime Venegas Hernández, a student at CETIS 78 in Altamira, Tamaulipas, has developed an innovative fire extinguishing device that relies on sound waves rather than traditional chemicals or water. Her inspiration began with a simple curiosity: the desire to extinguish a candle without blowing on it or using conventional substances.

To achieve this, she engineered a device powered by a 12-volt battery, utilizing a frequency generator and a speaker calibrated to emit 30 pulses per second. This specific vibration creates an effect that effectively pushes oxygen away from the flame. By depriving the fire of the oxygen it needs to burn, the device is able to extinguish flames in just five to eight seconds.

The project, which she has named Vortex Tech, has proven to be remarkably versatile through rigorous testing. Ángela conducted over 100 trials to confirm the device's efficacy across various types of fires, including those involving wood, flammable liquids, cooking grease, and sensitive electronic equipment.

Beyond its speed and effectiveness, the invention offers significant environmental and safety benefits. Because it operates through sound waves, the process does not create pollution, leave behind chemical residues, or pose any physical harm to the person operating the device.

What started as a modest school experiment has now gained international attention. As a result of her breakthrough, Ángela is set to represent Mexico at an international science fair, where she will showcase her method for combating one of humanity's oldest and most persistent dangers.

Google 宣布在 Europe 的有史以来最大单笔投资，承诺投入 130 亿欧元用于加强 Finland 的 AI 基础设施。该投资包括建设三座新数据中心、扩建位于 Hamina 的现有设施，并支持多项清洁能源项目，旨在满足对 Gemini 聊天机器人、 Google Search 、 Maps 和 YouTube 等 AI 驱动服务日益增长的全球需求。

计划的关键一环是与公用事业公司 Fortum 签订的一项为期 22 年的长期合约。根据协议，Google 将购买 Loviisa 核电站多达 50% 的发电量。这笔交易不仅为 Google 的高耗能业务提供了稳定电力，也为 Fortum 提供了增加发电能力并延长电厂运营寿命所需的财务保障。

Finland 正逐渐成为大型科技公司青睐的数据中心选址。 Prime Minister Petteri Orpo 强调了该国的优势：凉爽气候有助于自然降低大规模服务器的冷却成本，可靠且低碳的电力供应和稳定的数字基础设施也极具吸引力。最近 TikTok 也在当地投资了 10 亿美元。

Google 预计该项目将带来显著的经济效益，在 2027 年和 2028 年的施工期预计将创造超过 37,000 个就业机会。除了实体基础设施外，投资还将设立社区基金，用于支持当地生物多样性、科研、教育与劳动力培养。这一做法契合 Alphabet 更广泛的全球战略，即在竞争中大幅增加投入，以建立人工智能革命所需的产能。 Google has unveiled its largest single investment in Europe, committing 13 billion euros to boost AI infrastructure in Finland. The comprehensive investment package includes the construction of three new data centers, the expansion of an existing facility in Hamina, and support for various clean energy projects. This move is designed to meet the skyrocketing global demand for AI-driven services, such as the Gemini chatbot, Google Search, Maps, and YouTube.

A critical component of this initiative is a long-term, 22-year partnership with the Finnish utility firm Fortum. Under this agreement, Google will purchase up to 50 percent of the energy produced by the Loviisa nuclear power plant. This deal not only secures a stable power supply for Google's energy-intensive operations but also provides the utility company with the financial certainty needed to increase the plant's generating capacity and extend its operational life.

Finland has become an increasingly attractive destination for major technology companies looking to house their data centers. Prime Minister Petteri Orpo highlighted the country's strengths, including its cool climate, which naturally helps reduce the cooling costs associated with massive server arrays. Additionally, the nation offers a reliable, low-carbon electricity supply and a stable digital infrastructure, which recently also drew a 1 billion dollar investment from TikTok.

The economic impact of Google's project is expected to be significant, with the company anticipating the creation of over 37,000 jobs during the construction phases in 2027 and 2028. Beyond physical infrastructure, the investment also allocates resources for community funds that will support local biodiversity, research, education, and workforce development. This approach aligns with Alphabet's broader global strategy to ramp up spending significantly as it competes to build the necessary capacity for the ongoing artificial intelligence revolution.

尝试用本地 large language model 替代编码工具框架中对远程 API 的调用，往往会令人很沮丧。虽然像 llama-bench 这样的基准测试可能显示很高的 tokens-per-second，但真实的使用体验常常被明显的延迟、卡顿和漫长的等待时间破坏。这种差距主要因为大多数编码工具框架在设计时假定会运行在高速的数据中心 GPU 上，而非受限的消费级笔记本上。

本地硬件的物理特性会把系统逼到三种典型的设计困境上。第一，巨大的系统提示（system prompt）和复杂的工具 schema 带来沉重的预填充负担。当笔记本处理 token 的速度只是数据中心的一小部分时，一个 18,000 token 的提示可能导致模型在开始生成代码前就要等待几分钟。第二，这类大提示会占用大量上下文窗口，几乎没剩多少空间用于实际任务。第三，许多框架会不断发起会话标题或摘要的侧请求，在本地运行时这些请求会把模型推入频繁且重叠的队列，导致重复的长时间预填充，使得性能降到不可用的程度。

为了解这些瓶颈，在一台高端 M4 MacBook Pro 上，使用量化的 Qwen 3.8 27B 模型对九个流行的编码框架进行了多项 Python 练习测试。结果把这些工具分为三类：精简且稳定的框架保持短小的系统提示并高度重用缓存，是本地使用的最佳候选；尽管提示较长或工具 schema 较多但有纪律的重型工具，在完成初次预填充后仍能正常工作；而启动沉重型的工具可能需要三到四分钟的静默等待，代理才能迈出第一步，因而不适合本地工作流。

问题的核心不是工程水平低下，而是这些工具最初的设计环境不同。当延迟以毫秒计时，几乎免费的特性在本地硅片上以秒或分钟计时就成了沉重的负担。因此，像 chad 这样的项目应运而生，采取稀缺性思维，限制工具表面并针对持久化的前缀缓存（prefix caching）进行优化。这类设计把效率放在首位，确保开发循环保持响应，而不是被模型的开销阻塞。

真正的性能突破发生在框架与推理引擎运行在同一进程时。通过打破传统的客户端 - 服务器模型，将引擎内嵌到进程中（例如通过 MLX），可以将与请求模式相关的延迟降到最低。那些拥有自身缓存并采用专门草稿生成技术的框架，能够实现显著更高的 tokens-per-second，把曾经缓慢、令人沮丧的体验变成流畅高效的编码环境。 Attempting to use a local large language model as a replacement for remote API calls in coding harnesses often leads to frustration. While benchmarking tools like llama-bench might suggest high tokens-per-second, the actual user experience is frequently marred by significant latency, stalling, and long wait times. This discrepancy exists primarily because most coding harnesses were built with the assumption of high-speed data center GPUs, rather than the constraints of a consumer laptop.

The physics of local hardware often pits the system against three specific design choices. First, large system prompts and extensive tool schemas create a massive prefill burden. When a laptop processes tokens at a fraction of the speed of a data center, a 18,000-token prompt can result in several minutes of waiting before the model even begins to generate code. Second, these large prompts consume a significant portion of the available context window, leaving little room for actual task execution. Third, many harnesses are designed to make constant side requests for session titles or summaries, which, when run locally, force the model into frequent, overlapping queues and repeated long prefills that degrade performance to the point of being unusable.

To better understand these bottlenecks, nine popular coding harnesses were tested against a series of Python exercises on a high-end M4 MacBook Pro using a quantized Qwen 3.8 27B model. The results categorized these tools into three distinct groups. Lean and stable harnesses maintain trim system prompts and high cache reuse, making them the most viable candidates for local use. Heavy but disciplined tools, while burdened by long prompts or excessive tool schemas, remain functional once the initial prefill is complete. Finally, heavy-to-start tools can require three to four minutes of inactivity before the agent takes its first step, rendering them impractical for a local workflow.

The core issue is not poor engineering, but rather the environment for which these tools were originally designed. Features that are virtually free when latency is measured in milliseconds become heavy liabilities when measured in seconds or minutes on local silicon. Consequently, projects like chad have emerged to embrace a scarcity mindset, limiting tool surfaces and optimizing for persistent prefix caching. These designs prioritize efficiency to ensure that the development loop remains responsive rather than blocked by the model's overhead.

The true breakthrough in performance, however, arises when the harness and the inference engine share the same process. By breaking away from the standard client-server model and moving the engine in-process, such as through MLX, the latency associated with request patterns is minimized. Harnesses that own their cache and utilize specialized drafting techniques can achieve significantly higher token-per-second rates, turning what was once a sluggish, frustrating experience into a fluid and efficient coding environment.