在为 MultiAgentV2 消息引入加密负载传输后，Codex CLI 出现了显著的回归问题。尽管 pull request #26210 的加密更新旨在增强隐私性，但它无意中消除了对 agent 任务的人类可读审计链。将主要消息内容标记为已加密后，系统在本地历史、 trace 日志和 parent-side debug 界面中不再显示纯文本的任务细节。

该改动导致开发者无法方便地查看分派给 subagents 的具体任务，增加了调试工作流的难度，也让人在 rollout 期间难以弄清为何会启动某些 child threads 。目前系统只在加密字段中保存通信负载，而可读内容字段被保空。因此用户在查看交互历史时看到的只是不可读的 ciphertext，而不是实际发送给 subagents 的指令。

拟议的解决方案是在继续保留 model 所需的加密传输路径的同时，引入一个用于审计的强制性非加密纯文本伴随字段。通过为任务提供明文字段，系统既能保持对模型的安全传输支持，又能为开发者保留可见性。该审计字段应持久化到本地历史和 trace 元数据中，确保用户无需解密主负载就能随时检查被委派的指令。

用户对当前实现表示强烈不满，指出这使得高级 agent 功能难以使用和维护。鉴于部分 frontier models 现在强制采用 MultiAgentV2 及其加密 schema，可观测性的丧失已不再是可选权衡，而成为限制性的默认行为。社区普遍认为，健全的 introspection 对于运维透明至关重要，隐藏 agent 指令并未带来实质性的安全收益，反而妨碍了高效开发。 A significant regression has been identified in the Codex CLI following the implementation of encrypted payload delivery for MultiAgentV2 messages. While the encryption update, introduced in pull request #26210, aims to improve privacy, it inadvertently removes the human-readable audit trail for agent tasks. By marking the primary message content as encrypted, the system no longer displays plain-text task details in local history, trace logs, or parent-side debug interfaces.

This change prevents developers from easily inspecting what tasks are being delegated to subagents, making it difficult to debug workflows or understand why specific child threads were initialized during a rollout. Currently, the system stores the communication payload solely in an encrypted field, leaving the readable content field empty. Consequently, when users attempt to review their interaction history, they are presented with opaque ciphertext instead of the instructions actually sent to the subagents.

The proposed solution involves maintaining the encrypted delivery path for the model while simultaneously introducing a mandatory, non-encrypted plaintext companion field for auditing purposes. By including a clear text field for tasks, the system could preserve visibility for developers without compromising the secure delivery mechanism required by the model. This audit field would be persisted in local history and trace metadata, ensuring that users can inspect delegated instructions at any time without needing to decrypt the primary payload.

Users have expressed deep frustration with the current implementation, noting that it renders advanced agent features difficult to use and maintain. With certain frontier models now forcing the use of MultiAgentV2 and its encrypted schema, the loss of observability is no longer an optional tradeoff but a restrictive default. The consensus among the community is that robust introspection is essential for operational transparency, and that concealing agent instructions hinders effective development rather than providing meaningful security.

将 Google Play Integrity API 和 Apple App Attestation 集成到 EU Digital Identity Wallet 项目中用于年龄验证的提案，遭到开发者社区的强烈反对。批评者认为，依赖美国科技巨头的专有技术会削弱欧盟追求数字主权的努力。把基本政府服务绑定到特定操作系统和硬件供应商上，被视为加深对第三方公司的依赖，并把控制权交给那些常在不透明政策与利益冲突下运作的企业。

讨论者指出，这一做法违反了项目自身强调的互操作性、用户控制和普及可及性等原则。多位参与者表示，如荷兰身份识别应用 Yivi 等开源方案已经能实现有效的年龄验证，证明对外部专有依赖并非必需。还有人质疑必须为每次交互都使用独立应用的前提，认为现代基于 Web 的方法可以在不将用户锁入封闭生态的情况下实现同样目标。

安全与隐私是反对的核心。许多业内专家认为，硬件认证是朝着被锁定和审查的计算环境迈出危险一步。批评者称，这类措施并不会有效阻止恶意行为，反而会惩罚那些选择 root 设备或使用替代操作系统的合法用户。有观点认为，若强制执行，未来可能为控制而牺牲通用计算，使个人设备实质上变成受管控的客户端。

除技术层面外，讨论也反映出对规范起草者缺乏技术远见的强烈不满。许多参与者担心，该方案会造成类似 IE6 的锁定局面，使欧洲基础设施长期受制于外国垄断企业的任意要求。他们担忧，欧盟若采纳这些做法，不仅无法保护公民的数字权利，反而助长专有的"反功能"成为常态。

最后，参与者警告称，实施这种侵入性验证系统可能带来全球性后果，其他国家或会效仿。异见者普遍认为，若一个系统必须以牺牲用户自主权、隐私和互联网开放性为代价才能构建，那么该项目当前方向可能存在根本性缺陷。许多人呼吁回归开放标准和服务器端验证模型，尊重硬件所有者的自主权，而非强加由公司控制的集中限制。 The proposal to integrate Google Play Integrity API and Apple App Attestation for age verification within the EU Digital Identity Wallet project has faced significant backlash from the developer community. Critics argue that relying on proprietary technologies from American tech giants undermines the European Union's push for digital sovereignty. By tying essential government services to specific operating systems and hardware vendors, the project is seen as deepening dependence on third-party corporations and relinquishing control to entities that often operate under opaque policies and conflicting interests.

Participants in the discussion highlighted that this approach contradicts the project's own design principles, which emphasize interoperability, user control, and accessibility for everyone. Several contributors pointed out that effective age verification is already possible through alternative, open-source solutions like the Dutch identity app Yivi, which demonstrates that external, proprietary dependencies are not a functional necessity. Others questioned the entire premise of needing an app for every interaction, suggesting that modern web-based approaches could achieve the same goals without forcing users into restrictive ecosystems.

Security and privacy concerns are central to the objections, with many experts in the field describing hardware attestation as a dangerous step toward locked-down, censored computing. Critics argue that these measures do not effectively prevent malicious activity but instead serve to penalize legitimate users who choose to root their devices or use alternative operating systems. Some argue that if such systems are strictly enforced, they could lead to a future where general-purpose computing is sacrificed for the sake of forced control, effectively turning personal devices into managed clients.

Beyond the technical arguments, the conversation reflected deep frustration with the perceived lack of technical foresight among the specification drafters. Many contributors expressed concern that this implementation would create an "IE6-style" lock-in scenario, where European infrastructure becomes permanently chained to the arbitrary requirements of foreign monopolies. By adopting these methods, they fear the EU is not only failing to protect its citizens' digital rights but is also actively fostering an environment where proprietary "antifeatures" become the standard.

Finally, participants warned that the implementation of such intrusive verification systems could have global consequences, as other nations may imitate these practices. The consensus among the dissenting voices is that if a system cannot be built without sacrificing user agency, privacy, and the open nature of the internet, then the project may be fundamentally flawed in its current direction. Many called for a return to open standards and server-side verification models that respect the autonomy of hardware owners rather than imposing centralized, corporate-controlled constraints.

从 2026 年 7 月 1 日起，一项名为 Solar Sharer Offer 的政府支持计划将为 New South Wales 、 South Australia 和 South-East Queensland 的家庭每天提供至少三小时的免费电力。该倡议旨在把中午时段太阳能带来的好处直接让利给消费者，因为这段时间往往会导致电力批发价下降。居民只需安装智能电表并通过其电力零售商选择加入，即可参加该计划，因此无论是否自有住房或屋顶是否安装太阳能板，均有资格参与。

最初提案承诺提供不设限制的免费电力，但经过公众咨询后，决定设定每天 24 kWh 的上限。制定这一上限是为了在保证零售商财务可持续性的同时，对所有电网用户保持公平。政府指出，24 kWh 足以覆盖一个五口之家的平均日用电量，因此在正常情况下大多数家庭不太可能受该上限影响。如果某户当天用电超过上限，其超出部分将按照标准日间电价计费，但不会受到额外惩罚。

对于能把高耗能活动（如洗碗机、洗衣机或泳池水泵）转移到通常在中午左右的免费时段的家庭，Solar Sharer Offer 尤为划算。节省金额取决于家庭能迁移多少用电量，年节省额估计可能从 100 美元到超过 1,000 美元不等。

对于已经安装太阳能电池板和储能电池的房主来说，该计划还能提高整体效率：即便在阴天发电量低时，用户也能在免费时段从电网免费为家用电池充电，从而在傍晚高峰时段依靠储存的免费电力。同样，电动车车主也可在这段时间为车辆充电，进一步降低用电开支。

该项目标志着能源市场的一次重要转变，把太阳能繁荣带来的利益扩展到此前受限的租户和公寓居民。准备参与的家庭应确保已安装智能电表，并熟悉主要电器的定时功能。该计划将先在部分州推出，预计未来几年其他地区会陆续跟进，可能在一些私营零售商已提供的自愿性免费电力计划基础上进一步发展。 Starting July 1, 2026, a new government-backed scheme known as the Solar Sharer Offer will provide households in New South Wales, South Australia, and South-East Queensland with at least three hours of free electricity every day. This initiative is designed to distribute the benefits of midday solar energy, which often causes wholesale electricity prices to drop, directly to consumers. To participate, residents simply need a smart meter and to opt in through their energy retailer, making the program accessible regardless of homeownership status or the presence of rooftop solar panels.

While the original proposal promised free power without limitations, a public consultation process led to the introduction of a 24 kWh daily cap. This limit was implemented to ensure the financial sustainability of the scheme for retailers while remaining fair to all grid users. The government notes that 24 kWh is sufficient to cover the total daily usage of an average five-person household, meaning the cap is unlikely to affect most residents during typical operations. If a household happens to exceed this limit, their electricity usage simply reverts to standard daytime rates without penalties.

The Solar Sharer Offer is particularly beneficial for those who can shift their high-energy activities, such as running dishwashers, washing machines, or pool pumps, into the designated free window, which typically occurs around midday. Savings estimates vary based on how much energy usage a household can relocate, with potential annual savings ranging from $100 to over $1,000 for those who maximize their usage during these hours.

For homeowners already utilizing solar panels and battery storage, this scheme adds an extra layer of efficiency. Users can charge their home batteries from the grid at no cost during the free window, even on overcast days when solar generation is low, allowing them to rely on stored, free energy during expensive peak evening hours. Similarly, electric vehicle owners can use this period to charge their cars, further reducing their overall energy expenses.

The program represents a major shift in the energy market by extending the advantages of the solar boom to renters and apartment dwellers who previously faced barriers to entry. To prepare, interested households should ensure they have a smart meter installed and familiarize themselves with the scheduling capabilities of their major appliances. While the scheme is launching in select states first, other regions are expected to follow in the coming years, potentially building on the existing voluntary free-power plans already offered by some private retailers.

YouTrackDB 是 JetBrains 开发并内部使用的一款通用的面向对象图数据库。它专为高效处理复杂图关系而设计，通过消除昂贵的运行时 JOIN 实现高速数据处理——链接遍历的复杂度为 O(1) 。数据库支持既包含图概念又包含面向对象特性的丰富数据模型，数据库层面直接支持继承与多态。

一个关键的技术基础是默认启用快照隔离。每个事务都在事务开始时的数据库稳定快照上运行，有效避免幻读、不可重复读和脏读等常见并发问题。查询方面，系统原生支持 Gremlin 查询语言和 TinkerPop API；另有 YQL，一种基于 SQL 的查询语言，使用直观的点式表示进行遍历，并提供强大的图模式匹配函数。

该数据库支持灵活部署，兼容无模式、混合模式和全模式。它具备完善的安全能力，包括基于用户、角色与细粒度策略的权限体系，并支持对静态数据的透明加密。为便捷使用，YouTrackDB 可作为嵌入式数据库通过 shaded Maven 依赖引入，也可作为独立服务器运行，并提供 Docker 镜像，方便快速试验或生产部署。

YouTrackDB 的开发采用高度结构化的工作流程，强调质量与可预测性。每项重要贡献在进入逐步实现前，需经过严格的研究、设计与计划审查。为帮助新贡献者，项目维护了一本详尽的开发工作流程手册。代码库主要使用 Java 编写，并通过系统化的静态分析以维持高工程标准和软件可靠性。 YouTrackDB is a general-purpose, object-oriented graph database developed and utilized internally by JetBrains. It is designed to handle complex graph relationships efficiently, offering high-speed data processing by eliminating expensive run-time JOINs, as link traversals are handled with O(1) complexity. The database supports a rich data model that integrates both graph and object-oriented concepts, including support for inheritance and polymorphism directly at the database level.

A key technical foundation of YouTrackDB is its commitment to snapshot isolation by default. Every transaction operates on a stable snapshot of the database as it existed at the start of the transaction, which effectively eliminates common concurrency issues such as phantom reads, non-repeatable reads, and dirty reads. For query capabilities, the system provides native support for the Gremlin query language and the TinkerPop API. Furthermore, it features YQL, a SQL-based query language that uses intuitive dot notation for traversals and includes powerful matching functions for graph patterns.

The database is built for flexible deployment, supporting schema-less, schema-mixed, and schema-full modes. It includes robust security features, such as a profiling system based on users, roles, and granular security policies, alongside the capability for transparent data encryption at rest. Designed for ease of use, YouTrackDB can be deployed as an embedded database via a shaded Maven dependency or as a standalone server, with Docker images readily available for quick experimentation or production setup.

Development on YouTrackDB is characterized by a highly structured workflow that emphasizes quality and predictability. Every non-trivial contribution passes through a rigorous cycle of research, design, and plan review before moving to track-by-track implementation. To assist new contributors, the project maintains a comprehensive development workflow book that details these processes. The codebase itself is primarily written in Java and undergoes systematic static analysis to maintain high engineering standards and software reliability.

日本研究人员开发出一种突破性方法，可从废旧电动汽车电池中回收高达 90% 的锂。这比传统工艺（通常回收率不足 50%）有显著提升，如此高的回收率可能从根本上改变锂离子电池的制造与再利用方式。

该技术的核心在于一种特殊的化学改性：回收阶段不使用常规的氢氧化钠，而是采用回收得到的氢氧化锂（白色粉末）。这一工艺能将通常称为"黑粉"（black mass）的电池废料转化为可用于新电池的高纯度锂。除了效率提升，据称该方法还可比传统工艺减少约 40% 的碳排放，更具环保优势。

锂是电动汽车生产中关键且昂贵的原料，当前开采既耗能又常伴随复杂的地缘政治依赖。对几乎全部依赖进口电池矿产的日本而言，这项技术有望通过国内回收稳定供应链、降低对进口的依赖。

但专家也指出仍有重大挑战。目前日本只有约 14% 的废旧锂离子电池进入官方回收体系，这表明必须大幅改善回收渠道与基础设施，才能充分发挥新工艺的作用。

展望未来，研究人员计划扩大该技术的规模，目标是在 2027 年前提升产能，并争取到 2035 年实现每年提取数万吨电池材料的目标。若该方法能在全球推广，预计将大幅减少废弃物，为电动汽车产业提供更可持续的基础。 Japanese researchers have developed a groundbreaking method to recover up to 90 percent of lithium from used electric vehicle batteries. This innovation represents a significant leap forward in battery recycling, as traditional methods typically recover less than 50 percent of the material. By successfully extracting such a high percentage, the process could fundamentally change how lithium-ion batteries are manufactured and reused in the future.

The core of this new technique involves a specific chemical modification. Rather than using the standard sodium hydroxide, the engineering team utilizes recovered lithium hydroxide, appearing as a white powder, during the recycling phase. This process allows them to convert battery waste, commonly referred to as black mass, into high-purity lithium suitable for use in new batteries. Beyond its efficiency, the method is also environmentally beneficial, as it reportedly cuts carbon emissions by approximately 40 percent compared to conventional approaches.

This breakthrough is particularly important given that lithium is a critical and expensive component in electric vehicle production. Currently, mining is energy-intensive and often involves complicated geopolitical dependencies. For Japan, which imports nearly all of its battery minerals, this technology offers a path to stabilize supply chains and reduce reliance on foreign imports through domestic recovery.

Despite the potential of this technology, experts acknowledge that significant challenges remain. Currently, only about 14 percent of used lithium-ion batteries in Japan are successfully entering official recycling systems. This indicates that the country must drastically improve its collection infrastructure to fully realize the benefits of the new recovery process.

Looking ahead, there are ambitious plans to scale this technology further. Researchers aim to enhance production capabilities by 2027, with the ultimate goal of extracting tens of thousands of tons of battery materials annually by 2035. If this method is adopted on a global scale, it has the potential to significantly reduce waste and provide a more sustainable foundation for the future of the electric vehicle industry.

Satellitemap.space 提供了一个交互式的实时可视化平台，用于追踪卫星星座，覆盖 Starlink 、 GPS 等主要网络。该服务通过由 WebGL 支持的 3D 地球界面，展示数千个绕地球运行的人造物体的精确实时轨道位置。自 2019 年上线以来，该平台已发展为一个全面的监测资源，适用于从提供互联网的卫星网络到地球成像和天气监测等多种用途。

该应用不仅仅是简单的轨迹追踪，还提供了完善的分析工具和多种数据分类。用户可以按功能筛选卫星，涵盖全球定位、通信、业余无线电以及专用的 IoT 网络等。平台还具备观察轨道动力学的详细功能，例如追踪发射历史、监测再入风险，以及计算卫星相对于太阳、月球或其他行星的凌掠时刻。此外，网站提供专门的计算器，帮助识别潜在干扰并查看历史高度变化，对研究轨道衰减尤为有用。

对于关注这些星座配套基础设施的人，网站还列出地面站位置和技术规格等数据，并提供用于分析 Two-Line Element 集的工具，以及基于 GPU 加速的近距接近搜索，用以评估卫星之间的碰撞风险。该平台面向普通太空爱好者与需要深入技术细节的用户，均提供易用的访问方式，并支持导出数据以便进一步分析。

为提升用户体验，界面支持多种可视化叠加层，包括昼夜交替、大气渲染和实时地面站映射。平台还提供原生移动应用，集成增强现实功能并能推送天体事件通知。通过将复杂的航天数据以直观且可交互的方式呈现，该项目旨在让日益拥挤的轨道环境变得更易理解。 Satellitemap.space provides an interactive, real-time visualization platform for tracking satellite constellations, including major networks like Starlink, GPS, and others. The service utilizes a 3D globe interface powered by WebGL to display precise, live orbital positions for thousands of man-made objects orbiting the Earth. Since its launch in 2019, the platform has evolved into a comprehensive resource for monitoring everything from internet-providing satellite shells to Earth imaging and weather-monitoring networks.

The application goes beyond simple tracking by offering a robust suite of analytical tools and data categories. Users can filter satellites by function, ranging from global positioning and communications to amateur radio and specialized IoT networks. The platform also includes detailed features for observing orbital dynamics, such as tracking launch history, monitoring re-entry risks, and calculating specific satellite transits across the Sun, Moon, or planets. Furthermore, the site provides specialized calculators to help users identify potential interference and view historical altitude changes, which is particularly useful for studying orbital decay.

For those interested in the infrastructure supporting these constellations, the site includes data on ground station locations and technical specifications. It also features dedicated tools for analyzing Two-Line Element sets and conducting GPU-accelerated searches for close approaches between satellites to assess collision risks. The platform is designed to be accessible for both casual space enthusiasts and those requiring more detailed technical insights, with options to export data for further analysis.

To enhance the user experience, the interface supports various visual overlays, including day/night cycles, atmosphere rendering, and live ground station mapping. It also provides mobile accessibility through a native app that includes augmented reality features and push notifications for celestial events. By organizing complex aerospace data into an intuitive, interactive environment, the project aims to demystify the increasingly crowded landscape of modern orbit.

Arvind Narayanan 把 AI 看作既具变革性又属常态的技术。他认为，AI 并不会在短期内以某种超级智能全面取代人类工作，而更像电力等历次革命性技术，经历发明、创新和长达数十年的逐步扩散。尽管 AI 能力在提升，其经济与社会影响更多取决于组织如何随时间调整、重组并将这些工具融入人类的工作流程。

他强调能力与可靠性的区别。模型在多项任务上的表现大幅改进，但在一致性、鲁棒性和运行安全方面仍存不足。基于这些局限，Narayanan 预测在可预见的未来，AI 更可能作为协作工具发挥作用，而非完全自动化的"替代者"。他以起重机操作员为喻：机器承担认知上的"重活"，而人类仍掌控决策、规划和最终交付。

对于 AI 取代工作的担忧，他反驳了"效率提升必然导致就业减少"的观点，指出这是劳动总量谬论。随着任务变得更快、更便宜，相关需求往往会增长而非减少。他以软件工程、法律和放射学等领域为例，尽管 AI 被迅速采用，这些行业的就业仍保持稳定甚至增长。他强调，许多职业中编写代码或完成基础任务并非真正瓶颈，真正的限制在于更广泛的需求、规划和整合过程——也就是专业能力所在。

关于递归自我改进与超级智能的讨论，他认为常常缺乏细致区分，并指出在创造力和人类水平推理方面仍存在重大外部障碍，不是简单在实验室就能解决的。为此，他倡导"协同超级智能"的愿景，即人类将 AI 作为工具来扩展自身潜能。他主张研究社区应从单纯追求更强大模型的竞赛，转向优先建立严格的评估体系。只有通过严谨评估，才能引导 AI 朝有利于社会的方向发展，并确保人类判断在未来经济中继续处于核心且受重视的地位。 Arvind Narayanan frames AI as a transformative, yet normal technology. He argues that rather than being an imminent superintelligence that will replace all human work, AI follows a path similar to previous revolutionary technologies like electricity. This transition involves invention, innovation, and a slow process of diffusion that takes decades. He emphasizes that while AI capabilities are increasing, the actual economic and societal impact is determined by how organizations adapt, restructure, and integrate these tools into human workflows over time.

A key point of his framework is the distinction between capability and reliability. While AI models have seen dramatic improvements in performance on various tasks, they often struggle with consistency, robustness, and operational safety. Because of these limitations, Narayanan predicts that AI will be more successful as a collaboration tool rather than a fully automated worker for the foreseeable future. He uses the metaphor of a crane operator to describe the future of knowledge work, where machines perform the cognitive heavy lifting while humans remain in control of the decision-making, planning, and ultimate delivery.

Regarding anxieties about AI replacing jobs, Narayanan disputes the common notion that efficiency gains automatically lead to labor displacement. Instead, he points to the lump-of-labor fallacy, noting that as tasks become faster and cheaper to perform, the demand for them often grows rather than shrinks. He points to fields like software engineering, law, and radiology, where despite rapid AI adoption, employment has remained stable or even increased. He stresses that in many professions, writing code or performing basic tasks was never the true bottleneck; the bottleneck lies in the broader requirements, planning, and integration processes that define professional expertise.

Finally, Narayanan addresses the concepts of recursive self-improvement and superintelligence. He suggests that these are often discussed with too little nuance and that significant barriers, particularly regarding creativity and human-level reasoning, remain external to what can be solved in a lab. He advocates for a vision of co-superintelligence, where humans leverage AI as a tool to expand their own potential. He concludes that the community should shift its focus from purely building ever-more-capable models to prioritizing evaluation. Rigorous evaluation is necessary to steer the development of AI in a direction that benefits society and ensures that human judgment remains a central, valued component of the future economy.

递归性自我改进是指人工智能系统的输出被用作后续迭代的输入，从而可能形成自我维持的反馈循环。虽然这一概念常与"智能爆炸"相联系，但该模型专门考察实现 AI 能力自我持续加速所需的条件：当 AI 研究产出相对于模型能力的总体弹性超过 1 时，就会出现加速。在这个框架下，反馈强度等于开发过程中各阶段弹性的乘积，涵盖从算法效率到最终产出新 AI 系统的整个链条。

研究者用一系列有向图来描述这一过程，以追踪人类劳动力、训练算力、算法效率等投入如何驱动进展。核心的反馈回路在于 AI 能力自身促进算法效率的进一步提升。当各项弹性合并后，它们决定系统是否能达到自我维持增长的临界点。如果综合效应足够强，即使没有更多外生投入（比如更多人力或更可用的硬件），系统也能以加速的速度自我改进；反之若弹性偏低或出现收益递减，增长可能趋于稳定或逐渐消失。

分析中的一个重要区分是狭窄能力与广泛能力之间的差别。 AI 系统完全可能在狭窄的技术任务上出现快速且自我维持的改进，例如优化研究基准或代码，但这些进展未必能转化为广泛的经济价值。这样的狭义加速可能只推动 AI 研发本身显著进步，而对经济其他部门影响有限。由于加速的技术条件主要取决于模型能力与算法效率之间的联系，模型允许出现研究速度大幅提升而现实世界的生产率收益仍然受限或受不同约束的情形。

研究还强调各种瓶颈会如何破坏这些反馈回路。即便理论上存在自我持续改进的逻辑，物理和运营上的限制——例如高质量数据的可得性、实验性算力的限制，或对专业人力的持续依赖——都可能限制进展速度。这些瓶颈会削弱有效弹性，从而阻止理论上的加速在实践中出现。比如当实验性算力与模型能力是强互补关系时，硬件可用性的停滞就会成为递归过程的硬性上限。

最后，作者用现有的实证数据对模型进行了校准，指出目前尚未出现自我维持的加速，但反馈回路正在增强。粗略估算显示，目前 AI 能力每提高一个单位，AI 研发生产力大约提升 9%，低于模型中触发自我维持加速所需的大约 15% 阈值。尽管如此，趋势呈上升。研究者呼吁 AI 公司提高透明度，建议公开研发支出和 AI 驱动技术进展等具体数据点，以便更好衡量我们是否接近递归性自我改进的临界点。 Recursive self-improvement occurs when outputs of an artificial intelligence system are used as inputs to future iterations, potentially creating a self-sustaining feedback loop. While the concept is often associated with the idea of an intelligence explosion, this model focuses specifically on the conditions required for a self-sustaining acceleration in AI capabilities. Such an acceleration occurs if the total elasticity of AI research productivity with respect to model capabilities exceeds one. In this framework, the strength of the feedback is the product of elasticities across various stages of the development process, ranging from algorithmic efficiency to the eventual production of new AI systems.

The researchers model this process using a series of directed graphs to track how inputs like human labor, training compute, and algorithmic efficiency contribute to progress. The core feedback loop is established when AI capabilities themselves facilitate further improvements in algorithmic efficiency. When these elasticities are combined, they determine whether the system reaches a point of self-sustaining growth. If the combined effect is strong enough, the system can improve at an accelerating rate even without an increase in exogenous inputs like human researchers or available hardware. However, if the elasticities are low, or if diminishing returns set in, the system may instead see growth that stabilizes or fizzles out over time.

A critical nuance in the analysis is the distinction between narrow and broad capabilities. It is entirely possible for AI systems to demonstrate a rapid, self-sustaining improvement in narrow technical tasks, such as optimizing research benchmarks and code, without those advancements translating into broad economic value. This narrow acceleration could lead to significant progress in AI R&D while leaving other sectors of the economy relatively untouched. Because the technical conditions for acceleration depend specifically on the link between model capabilities and algorithmic efficiency, the model allows for a scenario where research speed increases dramatically even if broader, real-world productivity benefits remain limited or subject to different constraints.

The study also highlights how various bottlenecks can disrupt these feedback loops. Even if the internal logic for self-sustaining improvement exists, physical and operational constraints—such as the availability of high-quality data, the limitations of experimental compute, or the ongoing necessity for specialized human labor—could cap the speed of progress. These bottlenecks serve as dampeners that can lower the effective elasticities, potentially preventing an otherwise theoretical acceleration from manifesting in practice. The researchers note that if experimental compute and model capabilities are strong complements, for instance, then any stagnation in hardware availability acts as a hard limit on the recursive process.

Finally, the authors calibrate their model using current empirical data and suggest that while a self-sustaining acceleration is not currently underway, the feedback loops appear to be strengthening. A back-of-the-envelope calculation indicates that a one-unit increase in AI capabilities currently yields about a 9% improvement in AI R&D productivity, which falls short of the roughly 15% threshold required to trigger self-sustaining acceleration in their model. Nonetheless, the trend is moving upward. The researchers conclude by calling for more transparency from AI companies, suggesting that sharing specific data points on R&D expenditure and AI-driven technical advances would be invaluable for better measuring whether we are approaching the tipping point of recursive self-improvement.

在 Git 中同时处理多条并行更改常常变成一场复杂的杂耍，频繁需要进行压力山大的交互式变基（interactive rebasing），并可能把仓库弄坏。虽然诸如 jujutsu（即 jj）之类的版本控制工具因能处理这类工作流而备受关注，但有些用户觉得难以长期适应。一个值得注意且内嵌于 Git 核心发行版的替代方案是实验性的 git history 命令——它无需额外安装软件，就能提供强大的历史重写能力。

git history 包含三个主要子命令：fixup 、 reword 和 split 。 fixup 允许把暂存的更改折叠进较早的提交，并同时更新从该提交衍生出的所有本地分支。它比标准的 git rebase --update-refs 更进一步，会自动在所有相关分支上完成重写。由于该工具以原子方式执行，凡可能引发冲突的操作都会被拒绝，从而保证仓库始终处于稳定状态。

reword 用于更新提交信息：在特定提交上运行它，可以编辑该提交的信息，之后 Git 会自动重建后续提交栈并更新受影响的分支指向。 reword 仅在提交图上操作，不触及工作树或索引，因此即便目标分支未被检出也能执行，是随设计演进清理项目历史的高效手段。

split 用于把一个过于庞杂的提交拆成两个：通过交互式地从 diff 中挑选特定 hunk，用户可以决定哪些变更放入第一个提交，剩下的放入第二个。与其他子命令一样，split 会对下游提交执行必要的变基，从而在不需要手动且易出错的变基操作下，生成更清晰有序的提交历史。

虽然这些命令不如 jj 那样具备完整功能，比如高级冲突处理或操作日志，但对常规 Git 用户来说已是实质性进步。当前对冲突处理的限制是刻意为之，开发者把更复杂的冲突处理留待将来改进。 git history 以更安全、更直观的方式管理历史，是希望简化日常开发流程用户的一个强大内置工具。 Working with multiple parallel changes in git can often become a complex juggling act, frequently requiring stressful interactive rebasing that risks leaving a repository in a broken state. While tools like the version control system jujutsu, or jj, have gained significant traction for addressing these workflows, some users find the transition difficult to maintain. A compelling alternative currently embedded directly within the core git distribution is the experimental git history command, which offers powerful history-rewriting capabilities without requiring the installation of new software.

The git history tool features three primary subcommands: fixup, reword, and split. The fixup command allows a developer to take staged changes and fold them into an older commit, while simultaneously updating all local branches that descend from that point. This functionality goes beyond the standard git rebase --update-refs, as it performs the rewrite across every relevant branch automatically. Because the tool is designed to be atomic, it refuses any operation that might lead to a conflict, ensuring that the repository remains in a stable state throughout the process.

For updating commit messages, the reword command provides a streamlined experience. By running this command on a specific commit, a developer can edit the message, after which git automatically rebuilds the subsequent commit stack and adjusts the affected branch tips. Since reword operates strictly on the commit graph rather than the working tree or index, it can be executed on branches that are not currently checked out, making it a highly efficient way to clean up project history as designs evolve.

The final subcommand, split, is particularly useful for breaking down a single, over-complicated commit into two separate ones. By interactively selecting specific hunks from a diff, a user can designate what belongs in the first commit, while the remaining changes are relegated to the second. Much like the other commands, split handles the rebasing of any downstream commits, resulting in a cleaner and more organized commit history without the need for manual, error-prone rebase gymnastics.

While these commands may not offer the full suite of features found in jj, such as first-class conflict handling or an operation log, they represent a meaningful advancement for standard git users. The current limitation regarding conflicts is a deliberate design choice that the developers have left open to future development. By providing a safer, more intuitive way to manage history, the git history command serves as a powerful, built-in utility for those looking to simplify their daily development workflow.

Emily Eden 是一位著名的英国艺术家兼作家，在摄影尚未成为视觉记录普遍手段之前，就以独特的第一手视角记录了 19 世纪的 India 。她出身于英国显赫的政治家族，1836 年至 1842 年间随兄长 George Eden（当时的 governor-general of India）在印度北部广泛旅行。她的作品题材丰富，从贵族与战士到随从与动物，成为这一正在经历深刻政治变革社会的重要史料。

Eden 抵达 Calcutta 后经历了一段艰难的适应期，极度思乡且难以适应眼前的文化差异。她花了一段时间才走出孤独，开始记录周遭的一切。随着好奇心占上风，她逐渐超越传统风景画的框架，转而专注于随兄长随行队伍所见的人物、服饰与陌生场景。

这一时期的艺术成果以她的作品集 Portraits of the Princes and People of India 得以保存，该集于 1844 年以一系列手工上色的石版画出版。如今这些素描成为 Delhi 一场大型展览的焦点。艺术史学家称赞她卓越的观察力，指出她是那个时代少数能如此全面、细致记录印度人生活的英国女性之一。

尽管她观察入微并欣赏所遇之人，Eden 仍深受其时代的殖民心态影响。她把自己在 India 的存在视为一种责任，坚定地相信 Britain 的文明使命。这种双重性构成了她遗产的主要特征：她的艺术既是那个时期宝贵的民族志记录，也是塑造她世界观的帝国主义态度的反映。

1842 年回到 England 后，Eden 将创作重心转向写作，先后在 Up the Country 等书中发表了她在 India 的书信。她虽未放弃绘画，但对 Indian 题材的强烈专注随着她重回熟悉的家庭题材而逐渐减弱。今天，她的影响主要通过这些经久不衰的文字记录和那套引人注目的素描集得以确立，记录着 Indian subcontinent 上一个时代的暮色。 Emily Eden was a notable English artist and writer who captured a unique, firsthand perspective of 19th-century India long before photography became the standard for visual documentation. As a member of a prominent British political family, she travelled extensively across the northern region between 1836 and 1842 while accompanying her brother, George Eden, the governor-general of India. Her work, which included a diverse array of subjects from noblemen and warriors to everyday attendants and animals, serves as a significant historical record of a society undergoing major political transformation.

Her journey was marked by a difficult initial period of adjustment. Upon arriving in Calcutta, Eden struggled with extreme homesickness and the jarring cultural differences she encountered. It took time for her to overcome her isolation and begin documenting her surroundings. However, her natural curiosity eventually prevailed, leading her to move beyond traditional landscape painting. She began to focus intently on the people, costumes, and unfamiliar scenes she experienced while traveling with her brother's official party.

The artistic output from this period is best preserved in her collection, Portraits of the Princes and People of India, which was published as a series of hand-coloured lithographs in 1844. These sketches are now the focus of a major exhibition in Delhi. Art historians highlight her work for its exceptional observational quality, noting that she was one of the few British women of the era to produce such a comprehensive and detailed account of Indian life.

Despite her keen eye for detail and appreciation for the people she encountered, Eden remained firmly rooted in the colonial mindset of her time. She viewed her presence in India as an obligation and held an unshaken belief in Britain's civilising mission. This duality defines much of her legacy, as her art functions both as a valuable ethnographic record of the period and as a reflection of the imperial attitudes that shaped her worldview.

Following her return to England in 1842, Eden shifted her creative energy toward writing, eventually publishing her letters from India in books like Up the Country. While she continued to paint, the intense focus she had applied to her Indian subjects gradually lessened as she turned back toward more familiar domestic themes. Today, her legacy is primarily established through these enduring written accounts and her striking portfolio of sketches that captured the twilight of an era on the Indian subcontinent.