Orthrus 是一个新框架，旨在在不降低输出质量的前提下显著加速大型语言模型（LLM）的推理。它采用双架构，将传统自回归模型的逐 token 精准生成与扩散模型的高速并行能力相结合，从而突破通常限制 LLM 文本生成速度的顺序瓶颈，在保持严格无损生成的同时，实现了最高约 7.8 倍的加速。

系统通过同一模型的两种"视图"运行：自回归视图和扩散视图。两种视图共享完全相同的高保真键值（KV）缓存，几乎不增加额外内存，仅需 O(1) 级别的额外缓存。与需要独立草稿模型、因而消耗更多内存的投机解码方法（如 EAGLE-3 或 DFlash）相比，这种共享缓存是重要优势。 Orthrus 因而避免冗余，提升了 token 接受率，并且在输入上下文变长时表现更佳。

Orthrus 的另一个显著优势是参数效率：并行生成能力只通过微调约 16% 的模型参数来实现，而基础 LLM 保持完全冻结，使其成为对现有模型进行实用且高效升级的路径。该框架已在 Qwen3 骨干上实现，并提供多个模型检查点（1.7B 、 4B 和 8B 参数），所有版本均保证输出与原始基础模型的预测分布严格一致。

在性能基准测试中，Orthrus 持续优于现有的投机解码技术。它在每次前向传递中验证通过的 token 数更多，且随上下文长度增长更具扩展性。与那些在复杂推理任务上常出现精度下降的基于扩散的语言模型（dLLM）相比，Orthrus 保持了严格的保真度。例如，在 MATH-500 基准上，它相比 Qwen3-8B 基线实现了约 6 倍的加速且精度无损，而 Fast-dLLM-v2 等方法则表现出明显的精度下降。

该项目提供了简便的安装流程和快速入门指南，用户可通过 HuggingFace 上的可用模型快速开始生成文本，并且与 vLLM 、 SGLang 等主流服务框架的原生集成即将推出。详述 Orthrus 架构的研究论文已发表于 arXiv，代码和模型以 MIT 许可证开源，方便用于研究与商业应用。 Orthrus is a new framework designed to make large language model (LLM) inference significantly faster without sacrificing output quality. It introduces a dual-architecture approach that combines the precise, token-by-token generation of traditional autoregressive models with the high-speed parallel capabilities of diffusion models. This hybrid method allows Orthrus to break through the sequential bottleneck that typically limits how fast LLMs can generate text, achieving speedups of up to 7.8 times while maintaining strictly lossless generation.

The system works by using two "views" of the same model: an autoregressive view and a diffusion view. Both views share the exact same high-fidelity Key-Value (KV) cache, which means there is virtually no additional memory overhead, only O(1) extra cache required. This shared cache is a key advantage over speculative decoding methods like EAGLE-3 or DFlash, which require separate draft models and thus consume more memory. Orthrus avoids this redundancy, leading to higher token acceptance rates and better performance, especially as the length of the input context grows.

A major strength of Orthrus is its parameter efficiency. The parallel generation capabilities are added by fine-tuning only 16% of the model's total parameters, while the base LLM remains completely frozen. This makes it a practical and efficient upgrade path for existing models. The framework has been implemented with a Qwen3 backbone, and several model checkpoints are available, including versions at 1.7B, 4B, and 8B parameters, all of which guarantee that the output matches the original base model's exact predictive distribution.

In performance benchmarks, Orthrus consistently outperforms existing speculative decoding techniques. It achieves a higher average number of verified tokens per forward pass and scales more efficiently with longer contexts. When compared to other diffusion-based language models (dLLMs), which often suffer from accuracy drops on complex reasoning tasks, Orthrus maintains strict fidelity. For example, on the MATH-500 benchmark, it delivers a roughly 6x speedup over the Qwen3-8B baseline with no loss in accuracy, whereas other methods like Fast-dLLM-v2 show significant degradation.

The project provides a straightforward installation process and a quickstart guide for users to begin generating text with the available HuggingFace models. It also notes that native integration with popular serving frameworks like vLLM and SGLang is coming soon. The research paper detailing the Orthrus architecture has been published on arXiv, and the code and models are released under an MIT license, making it accessible for both research and commercial applications.

本文认为数据库系统必须采用异地写入（out-of-place writes），以充分发挥 SSD 的性能并延长其寿命。作者证明，MySQL 、 PostgreSQL 等系统采用的传统原地写入在 DBMS 和 SSD 两层都会引起严重的写放大（WA）。例如，LeanStore 中一次 4 KiB 页面写入实际上在闪存上写入了 18.85 KiB，放大约 4.7 倍，这主要由 DBMS 层的双写缓冲和 SSD 层的垃圾回收导致。这不仅浪费带宽、增加延迟，还大幅缩短 SSD 的耐久性：测试中 SSD 在负载下仅 1.5 个月就达到了写入寿命上限。

为了解决这些问题，作者提出了一套基于异地写入架构的优化方案。 DBMS 层引入页面级压缩与页面打包，在减少写量的同时保持高效的 4 KiB 对齐读取；并提出按死亡时间分组（GDT），利用数据库语义估算页面失效时间，将生命周期相近的页面归为一组，从而在垃圾回收时减少 DB 层的写放大，确保同一区域内的页面在大致相同时间失效。

在 SSD 层，作者提出了降低内部写放大的方法。对于 Zoned Namespace（ZNS）SSD，设计与主机管理的 zone 自然对齐，可保证 SSD 的写放大因子（WAF）为 1 。对于普通 SSD，作者将 DBMS 的垃圾回收单元与 SSD 的内部超级块（superblock）大小对齐，该大小可通过 FDP Reclaim Unit 信息或类似 ZNS 的写入模式推断出来。另一个关键是 NoWA（No Write Amplification）模式：通过补偿写入确保 SSD 始终有完全失效的超级块可用，从而消除对 SSD 层垃圾回收的需求，即便在商用硬件上也能实现 WAF=1 。

作者在基于 B 树的 LeanStore 的修改版 ZLeanStore 中实现了这些优化。多种基准测试和不同 SSD 上的评估表明效果显著：在 YCSB-A 上，吞吐量提升 1.65–2.24 倍，单次操作的闪存写入量减少 6.2–9.8 倍；在 15,000 仓库的 TPC-C 测试中，吞吐量提升 2.45 倍，闪存写入减少 7.2 倍。该设计还无缝支持 ZNS 、 FDP 等现代 SSD 接口，为实现更高效、更耐用的数据库存储提供了可行路径。 This paper argues that database systems must adopt out-of-place writes to fully leverage SSD performance and extend SSD lifespan. The authors demonstrate that traditional in-place write designs, used by systems like MySQL and PostgreSQL, suffer from significant write amplification (WA) at both the DBMS and SSD layers. For example, a single 4 KiB page write in LeanStore results in 18.85 KiB of actual flash writes, a 4.7x amplification caused by DBMS-level doublewrite buffering and SSD-level garbage collection. This wastes bandwidth, increases latency, and drastically shortens SSD endurance, with the tested SSD reaching its write limit in just 1.5 months under load.

To address this, the authors propose a set of optimizations built on an out-of-place write architecture. At the DBMS level, they introduce page-wise compression combined with page packing to reduce write volume while maintaining efficient 4 KiB-aligned reads. They also propose Grouping by Death Time (GDT), which uses database semantics to estimate when pages will be invalidated and groups those with similar lifetimes together. This reduces DB-level write amplification during garbage collection by ensuring zones contain pages that become invalid around the same time.

At the SSD level, the paper presents techniques to minimize internal SSD write amplification. For Zoned Namespace (ZNS) SSDs, the design naturally aligns with the host-managed zones, guaranteeing an SSD WAF of 1. For standard SSDs, the authors align the DBMS garbage collection unit with the SSD's internal superblock size, inferred either through FDP Reclaim Unit information or a ZNS-like write pattern. They also introduce the NoWA (No Write Amplification) pattern, which uses compensation writes to ensure the SSD always has fully invalidated superblocks available, eliminating the need for SSD-level garbage collection and achieving WAF = 1 even on commodity hardware.

The authors implement these optimizations in ZLeanStore, a modified version of the B-tree-based LeanStore. Evaluation across diverse benchmarks and SSDs shows substantial improvements. On YCSB-A, throughput increases by 1.65–2.24x while flash writes per operation decrease by 6.2–9.8x. For TPC-C with 15,000 warehouses, throughput improves 2.45x with a 7.2x reduction in flash writes. The design also seamlessly supports modern SSD interfaces like ZNS and FDP, demonstrating a practical path toward more efficient and durable database storage.

Mitchell Hashimoto，Ghostty 的创建者、 HashiCorp 的创始人在 X 上发帖，表达了对软件开发行业普遍存在的"AI 狂热症"的深切担忧。他认为，许多公司对 AI 抱有近乎非理性的热情，导致关于其风险的理性讨论变得几乎不可能——即便是与他非常尊敬的朋友交谈，也常遭到回避。他把这种情形比作当年云基础设施转型时期围绕 MTBF（平均故障间隔时间）与 MTTR（平均恢复时间）的那场争论。类似的争论如今再次出现，但这次波及的是整个软件开发行业，甚至可能影响更广泛的领域。

Hashimoto 将当前 AI 倡导者的心态概括为几乎绝对的"MTTR 就是一切"。这种思路认为发布有缺陷的代码没关系，因为 AI 代理能以人类无法企及的速度和规模修复问题。他认为这是基础设施领域曾经付出代价后才学到的危险教训：MTTR 很重要，但绝不能完全放弃构建有韧性的系统。问题在于，人们常以局部指标来搪塞担忧，例如完整的测试覆盖率或下降的 Bug 报告数，但这些指标无法全面反映真实状况。

Hashimoto 指出的核心问题是，系统在局部指标上可能显得健康，但在全局层面却变得难以理解。 Bug 报告可能在减少，而潜在风险却在迅速积累；测试覆盖率可能上升，而对代码库的语义理解却在下降。变化之快以至于无人察觉底层架构在逐步退化。他将这种情况比作基础设施团队曾通过自动化将系统变成一台"高度韧性的灾难机器"：表面上运转良好，但整体脆弱且缺乏充分理解。

他对这种趋势对行业及其身边人的影响表示真挚的担忧，并且发现很难提出这些担忧，因为回应往往是立即的否定，未能触及更深层、系统性的问题。该系列帖子引发了广泛共鸣，获得超过 218,000 次浏览和数百条回复，表明许多软件社区成员也对不受约束的 AI 热潮及基础工程纪律被削弱感到忧虑。 Mitchell Hashimoto, creator of Ghostty and founder of HashiCorp, posted a thread on X expressing deep concern about what he calls "AI psychosis" across the software development industry. He believes many companies are currently caught up in an irrational enthusiasm for AI that makes rational conversation about its risks nearly impossible, even with personal friends he deeply respects. He draws a parallel to his experience during the cloud infrastructure transition, when the industry went through a major reckoning around MTBF (mean-time-between-failure) versus MTTR (mean-time-to-recovery). Those same arguments are resurfacing now, but this time they apply to the entire software development industry, and possibly the world at large.

Hashimoto describes the current mindset among AI enthusiasts as an almost absolute "MTTR is all you need" mentality. The thinking goes that it is fine to ship buggy code because AI agents will fix issues so quickly and at a scale humans cannot match. He argues this is a dangerous lesson the infrastructure world already learned. MTTR is valuable, but you cannot completely abandon resilient systems. The problem is that people dismiss concerns by pointing to local metrics like full test coverage or declining bug reports, which do not capture the full picture of what is happening.

The core issue Hashimoto raises is that systems can appear healthy by narrow, local metrics while globally becoming incomprehensible. Bug reports may go down while latent risk explodes. Test coverage can rise while semantic understanding of the codebase falls. Changes happen so fast that nobody notices the underlying architecture decaying. He compares this to how teams in infrastructure once automated themselves into what he calls a "very resilient catastrophe machine," where everything looks fine on the surface but the system as a whole is fragile and poorly understood.

Hashimoto expresses genuine worry about how this will play out, both for the industry and for the people he knows personally. He finds it difficult to even bring up these concerns because the responses he gets are immediate dismissals that fail to engage with the deeper, systemic risks. The thread resonated widely, garnering over 218,000 views and hundreds of replies, suggesting that his concerns about unchecked AI enthusiasm and the erosion of foundational engineering discipline are shared by many in the software community.

一项名为 Protect Our Games Act 、旨在保障在发行商关闭在线游戏后玩家访问权的 California 法案，已通过 Assembly 的拨款委员会，离全体表决更近一步。该法案要求发行商在停止对在线游戏的支持时，要么向玩家提供全额退款，要么发布一个不依赖发行商服务器、可独立运行的更新版本；同时要求在关闭维持正常游戏所需的服务前至少提前 60 天通知。法案适用于自 2027 年 1 月 1 日起在 California 销售的游戏，但 free-to-play 游戏和仅订阅制的游戏将被豁免。

该法案的推进被视为 Stop Killing Games 运动的一大胜利。该草根玩家维权组织在 Ubisoft 于 2024 年关闭其游戏 The Crew 后成立。总部设在 UK 的 SKG 表示曾参与法案起草，并帮助建立了美国分支以推动法案通过。该组织认为，没有其他媒介会允许产品在售出后毫无通知地被收回，随着 live-service 游戏越来越普及，明确的退役流程对保护消费者至关重要。

代表主要游戏发行商的 Entertainment Software Association（ESA）对该法案提出反对。 ESA 认为消费者购买的是对游戏的使用许可而非所有权，依赖在线服务的游戏最终关闭是需要持续基础设施的现代软件的自然结果。该组织还警告称，法案可能在音乐或其他知识产权许可方面制造难以解决的局面——这些许可通常有时效性，可能迫使发行商无限期重新谈判许可，或以法律或技术上不可行的方式修改游戏。

尽管面临行业反对，该法案已先后通过了 Privacy and Consumer Protection 委员会和 Judiciary 委员会，并在拨款委员会获得通过。接下来它仍需在 California Assembly 和 Senate 均获多数票通过，才能送交州长 Gavin Newsom 办公室签署。此时，Stop Killing Games 在 UK 的运动势头有所放缓：去年 11 月 UK parliament 就游戏保存展开的辩论并未促成政府采取行动。 A California bill aimed at preserving access to online games after publishers shut them down has advanced out of the Assembly's appropriations committee, moving closer to a full floor vote. The Protect Our Games Act would require publishers who discontinue support for an online game to either provide full refunds to players or release an updated version of the game that can function independently of the publisher's servers. The bill would also mandate 60 days' notice before shutting down services necessary for normal gameplay. It would apply to games sold in California starting January 1, 2027, though free-to-play games and subscription-only titles would be exempt.

The bill's advancement is a significant victory for the Stop Killing Games movement, a grassroots player advocacy group formed after Ubisoft shut down its game The Crew in 2024. SKG, which is based in the UK, says it advised on the drafting of the legislation and helped establish a US branch to support its passage. The group argues that no other medium allows a product to be sold to consumers and then taken away without notice, and that as live-service games grow more popular, clear end-of-life procedures are essential for consumer protection.

The Entertainment Software Association, representing major game publishers, has pushed back against the bill. The ESA argues that consumers purchase a license to access a game rather than outright ownership, and that the eventual shutdown of online-dependent games is a natural feature of modern software requiring ongoing infrastructure. The group also warns that the bill could create impossible situations around licensing for music or other intellectual property, which are often time-limited, potentially forcing publishers to renegotiate licenses indefinitely or alter games in ways that aren't legally or technically feasible.

Despite industry opposition, the bill has cleared multiple committees, including Privacy and Consumer Protection and Judiciary, before passing appropriations. It still needs majority approval in both the full California Assembly and the Senate before reaching Governor Gavin Newsom's desk. The progress in California comes as the Stop Killing Games campaign has seen momentum stall in the UK, where a parliamentary debate on game preservation last November did not result in government action.

截至 2026 年 5 月 15 日，ABC News 已将 FiveThirtyEight 的所有文章全部下线，相关页面现重定向至 abcnews.com/politics 。 FiveThirtyEight 前高级编辑兼高级选举分析师、 Votebeat 现任主编 Nathaniel Rakich 称此举是"对数千页知识的无谓抹除"。

Rakich 在原帖中强调了这次档案消失的重大意义。由 Nate Silver 创立的 FiveThirtyEight 以数据驱动的报道著称，尤其在政治、体育和选举领域，多年来积累了大量分析性文章和预测内容。

围绕此事的讨论反映出新闻界与数据界的普遍担忧。许多人认为，删除这些内容意味着失去一项宝贵的公共资源，这些材料长期以来被广泛引用并用于教学和研究。

将内容重定向到 ABC News 的通用政治页面，表明这些内容正被并入更大的 ABC News 品牌。这符合媒体公司精简数字资产的趋势，但有时会以牺牲小众却有影响力的媒体为代价。

FiveThirtyEight 档案的消失成为数字媒体史上的重要一刻，也引发了关于在线知识保存以及媒体机构在收购专业出版物后应承担何种责任的深刻疑问。 ABC News has taken all FiveThirtyEight articles completely offline as of May 15, 2026. The articles now redirect to abcnews.com/politics. Nathaniel Rakich, former senior editor and senior elections analyst at FiveThirtyEight and current managing editor at Votebeat, called this a "needless erasure of thousands of pages of knowledge."

The original post by Rakich highlights the significance of this archival loss. FiveThirtyEight, founded by Nate Silver, was known for its data-driven journalism, particularly in politics, sports, and elections. It built a substantial archive of analytical articles and forecasts over the years.

The conversation around this move reflects concern within the journalism and data communities. Many see the removal as a loss of valuable public resource. The site's content was widely cited and used for educational and research purposes.

The redirect to a general politics page on ABC News suggests a consolidation of content under the broader ABC News brand. This is part of a larger trend where media companies streamline digital properties, sometimes at the expense of niche but influential outlets.

The erasure of the FiveThirtyEight archive represents a significant moment in digital media history. It raises questions about the preservation of online knowledge and the responsibilities of media organizations when they acquire specialized publications.

Zulip 正在进行重大组织调整——创始人 Tim Abbott 将不再担任全职领导，并与另外三位高级成员一同加入 Anthropic 。为保障项目的长期稳定与独立，Zulip 背后的公司 Kandra Labs 已捐赠给新成立的非营利机构 Zulip Foundation 。该基金会将全资持有 Kandra Labs，不再有其他股东或债务，确立了类似 Mozilla 和 Signal 的治理结构。此举旨在对 Zulip 的核心价值观作出永久承诺，并通过拨款和可免税捐赠开辟可持续的资金渠道。

Zulip Foundation 的初始董事会包括 Tim Abbott 、 Greg Price 、 Alya Abbott 和 Josh Triplett，顾问委员会汇集了开源和学术界的知名人士，如 Andrew Sutherland 和 Jeremy Avigad 。基金会的使命是打造最佳的团队聊天体验，尤其关注公共利益组织。尽管领导层发生变动，Zulip 的各项运营——包括云托管、支持合同以及 Google Summer of Code 等社区项目——将继续正常运行。 Kim Vandiver 已加入担任临时总裁，负责管理过渡并牵头寻找新的常任领导。

此次转型的主要目的是对 Zulip 的价值观作出长期且公开的承诺，并提升筹资能力。作为非营利组织，基金会现在可以申请拨款并接受捐赠，从而避免外部投资者施压导致项目在原则上让步，例如在数据隐私问题上的坚定立场。 Tim Abbott 表示，他决定加入 Anthropic，是出于希望更直接参与推动 AI 负责任发展的愿望，他认为这对人类的未来至关重要。他也强调，自己离开的前提是已确保 Zulip 能在没有他时继续蓬勃发展。

Zulip 的未来掌握在剩余的 12 名专业维护者手中，他们在该项目的平均工作年限超过四年，拥有多次交付重大改进的经验证明。团队即便在 Abbott 因育儿假或慢性疾病无法参与时，也保持了持续稳定的推进。尽管组织适应期间开发节奏可能短暂放缓，但团队严谨的开发流程和稳健的文化有望推动项目继续前进。基金会正在积极招聘以填补因人员加入 Anthropic 而留下的空缺，并邀请社区参加在线问答，讨论相关变化。 Zulip is undergoing a major organizational transition as its founder, Tim Abbott, steps back from full-time leadership to join Anthropic, along with three other senior team members. To ensure the project's long-term stability and independence, Kandra Labs, the company behind Zulip, has been donated to a newly created nonprofit entity called the Zulip Foundation. This foundation will now fully own Kandra Labs, with no other stockholders or debt, formalizing a governance structure similar to organizations like Mozilla and Signal. The move is designed to provide a permanent commitment to Zulip's core values and create new avenues for sustainable funding through grants and tax-deductible donations.

The Zulip Foundation's initial board includes Tim Abbott, Greg Price, Alya Abbott, and Josh Triplett, with an advisory board featuring notable figures from the open-source and academic communities, such as Andrew Sutherland and Jeremy Avigad. The foundation's mission is to develop the best possible team chat experience, with a particular focus on public-interest organizations. Despite the leadership change, Zulip's operations, including its cloud hosting, support contracts, and community programs like Google Summer of Code, will continue without interruption. Kim Vandiver has joined as Interim President to manage the transition and lead the search for new permanent leadership.

The primary motivations for this transition are to make a permanent, public commitment to Zulip's values and to enable more effective fundraising. As a nonprofit, the foundation can now apply for grants and accept donations without the risk of external investors pressuring the project to compromise its principles, such as its strong stance on data privacy. Tim Abbott explained that his decision to join Anthropic was driven by a desire to contribute more directly to the responsible development of AI, a cause he believes is vitally important for humanity's future. He emphasized that his departure was contingent on ensuring Zulip could thrive without him.

The future of Zulip is in the hands of its remaining team of 12 professional maintainers, who have an average of over four years of experience on the project and a proven track of shipping major improvements. The team has a history of steady progress, even during periods when Abbott was unavailable due to parental leave or a chronic illness. While there may be a temporary reduction in development velocity as the organization adapts, the team's disciplined development process and strong culture are expected to carry the project forward. The foundation is actively hiring to fill the roles left by those moving to Anthropic and is inviting the community to a live Q&A session to discuss the changes.

美国司法部正在升级对 EZ Lynk 的法律攻势，要求 Apple 、 Google 、 Amazon 和 Walmart 提供可能超过 10 万名 EZ Lynk Auto Agent 应用用户的个人数据。该应用与一个物理硬件加密狗配套使用，正成为 Clean Air Act 案件的焦点；司法部指控其为"作弊设备"，可让用户绕过柴油车的出厂排放控制。传票要求提供姓名、地址、电话号码和购买记录，以便识别可就工具使用情况作证的证人。

EZ Lynk 强烈否认指控，称其产品具有合法用途，如车辆性能监控、诊断和软件更新，并认为任何涉及排放的滥用应由用户自行承担，而非产品本意。但司法部已提交论坛帖子和社交媒体证据，显示部分用户利用该系统禁用排放控制，从而主张需要更广泛的用户数据来支撑案件。

隐私倡导者和 EZ Lynk 的法律团队对此强烈反弹，称传票属严重越权，超出案件必要范围，并带来重大的 Fourth Amendment 担忧。 EFF 和 EPIC 批评要求大量个人身份信息的做法，指出大多数用户并未阅读服务条款，仅因下载一个标榜用于车辆诊断和调校的工具就可能面临意想不到的法律风险。

此案凸显了汽车爱好者改装车辆的诉求与联邦环保法规之间日益紧张的冲突。 right-to-repair 倡导者认为这是更广泛冲突的一部分；一位专家指出，"人们想改装他们的汽车，而且永远都会这样。"政府日益倾向将应用下载追溯到个人用户的做法，标志着执法策略的明显转变，尤其是此次请求的规模相比以往更大。

据报道，Apple 和 Google 正准备对传票提出挑战；相关公司和司法部在法庭文件外拒绝置评。本案结局可能为监管执法中的数字隐私问题树立重要先例。目前，对于使用调校工具的车主来说，信息很清楚：政府越来越有能力将应用使用行为直接关联到个人身份，这使得隐私和合规风险大幅上升。 The U.S. Department of Justice is escalating its legal battle against EZ Lynk, a Cayman Islands-based company, by demanding that Apple, Google, Amazon, and Walmart hand over personal data on potentially over 100,000 users of the EZ Lynk Auto Agent app. The app, paired with a physical hardware dongle, is at the center of a Clean Air Act case, with the DOJ alleging it functions as a "defeat device" that allows users to bypass factory emissions controls on diesel vehicles. The subpoenas seek names, addresses, phone numbers, and purchase histories to identify witnesses who can testify about how the tools were used.

EZ Lynk strongly denies the allegations, arguing its products serve legitimate purposes like vehicle performance monitoring, diagnostics, and software updates. The company contends that any emissions-related misuse falls under user responsibility, not its intended function. Despite this, the DOJ has already presented forum posts and social media evidence showing some users employing the system to disable emissions controls, justifying the need for broader user data to build its case.

Privacy advocates and EZ Lynk's legal team have pushed back hard, calling the subpoenas a significant overreach. They argue the requests go far beyond what is necessary for the case and raise serious Fourth Amendment concerns. The Electronic Frontier Foundation (EFF) and Electronic Privacy Information Center (EPIC) have criticized the broad demand for personally identifiable information, noting that most users never read terms of service and may face unintended legal exposure simply for downloading a tool marketed for car diagnostics and tuning.

The case highlights a growing tension between car enthusiasts' desire to modify their vehicles and federal environmental regulations. Right-to-repair advocates view this as part of a broader conflict, with one expert noting, "People want to modify their cars and always will." The government's increasing willingness to trace app downloads back to individual users marks a notable shift in enforcement tactics, especially given the scale of this request compared to similar past actions.

Apple and Google are reportedly preparing to challenge the subpoenas, while the companies and the DOJ have declined public comment beyond court filings. The outcome could set important precedents for digital privacy in regulatory enforcement cases. For now, the message to car owners using tuning tools is clear: governments are increasingly capable of linking app usage directly to individual identities, raising the stakes for both privacy and regulatory compliance.

一位名为 AwesomeQubic 的用户在 Bun 运行时的 GitHub 仓库提交了一个 issue，称整个 Rust 代码库连最基本的 Miri 检查都过不了，并在安全的 Rust 中允许出现未定义行为（UB）。报告给出了一个针对 `PathString::init` 的具体示例：该函数接收一个带隐式生命周期的 `&[u8]`，但在返回时擦除了生命周期，使得返回的 Self 实际上表现为 `'static`，从而产生悬垂引用。由此可以发生 use-after-free：创建一个 Box 、用它初始化一个 `PathString` 、释放 Box 后再打印该切片，Miri 因缺乏 provenance 将其标为 UB 。

该 issue 在社区引发强烈反响，许多人对这样基础的内存安全漏洞出现在一个依赖 Rust 提供性能与安全保障的项目中感到沮丧。评论者 JavaDerg 强调了问题的严重性，指出 Rust 的安全模型建立在强假设之上，UB 可能在意想不到的地方引发不可预测的问题，从而抹杀使用 Rust 的优势。讨论还涉及 AI 编码助手的角色；原报告者认为"vibe coding" 加上 AI 容易导致此类错误，建议雇佣有经验的 Rust 开发者。

作为回应，合作者 robobun 确认了该 bug 并链接了一个修复用的拉取请求（#30728）。修复方案包括将 `PathString::init` 及 `dir_iterator::next()` 中的类似漏洞标注为带有文档化 outlives 合约的 `unsafe fn`，对大约 70 个仓内调用点逐一审计并添加每处的 SAFETY 注释，并新增回归测试。 robobun 指出，尽管 diff 本地显示通过，但 CI 在无关的分支上不稳定，问题源于既有的 WebKit/GC 问题。

讨论期间还出现了若干其他旨在减少 unsafe 使用的 PR，例如用安全等价物替换 `ArrayHashMap` 中的不安全代码块，以及将 `DynamicBitSet` 重写为基于 `Box<[usize]>` 的实现。但该线程逐渐偏离主题并变得激烈，一些人争论 Zig 与 Rust 的优劣，另一些人批评项目过度依赖 AI 生成的代码。有人还用 grep 展示问题规模，在 Rust 文件中发现超过 13,000 处 `unsafe` 。最终，仓库维护者将该 issue 置为离题并锁定，限制进一步讨论仅限合作者。 A user named AwesomeQubic opened an issue on the Bun runtime's GitHub repository, alleging that the entire Rust codebase fails even the most basic Miri checks and allows for undefined behavior (UB) in safe Rust. The reporter provided a specific code example involving `PathString::init`, where a dangling reference is created because the function takes a `&[u8]` with an implicit lifetime but erases it, returning a `Self` that is effectively `'static`. This allows for use-after-free scenarios, as demonstrated by creating a `Box`, initializing a `PathString` from it, dropping the `Box`, and then attempting to print the slice, which Miri flags as UB due to a lack of provenance.

The issue sparked significant community reaction, with many expressing frustration over the presence of such fundamental memory safety bugs in a project that leverages Rust for its performance and safety guarantees. Commenters like JavaDerg elaborated on the severity, noting that Rust's safety model relies on strong assumptions, and UB can cause unpredictable issues in unexpected places, effectively nullifying the advantages of using Rust. The discussion also touched on the role of AI coding assistants, with the original reporter suggesting that "vibe coding" with AI leads to these mistakes and recommending the hiring of experienced Rust developers.

In response, a collaborator named robobun confirmed the reproduction of the bug and linked a pull request (#30728) intended to fix it. The fix involves marking `PathString::init` and a similar hole in `dir_iterator::next()` as `unsafe fn` with documented outlives contracts, auditing approximately 70 in-tree call sites with per-site SAFETY comments, and adding a regression test. Robobun noted that while the diff was green, CI was flaky on unrelated lanes due to pre-existing WebKit/GC issues.

The conversation also saw the emergence of several other pull requests aimed at reducing unsafe usage across the codebase, such as replacing unsafe blocks in `ArrayHashMap` with safe equivalents and rewriting `DynamicBitSet` on `Box<[usize]>`. However, the thread became increasingly off-topic and contentious, with some users debating the merits of Zig versus Rust and others criticizing the project's reliance on AI-generated code. One user even ran a grep command to highlight the scale of the problem, finding over 13,000 instances of `unsafe` in Rust files. Eventually, the repository maintainers locked the issue as off-topic and limited further conversation to collaborators.

Project Gutenberg 是一个拥有超过 75,000 本免费电子书的数字图书馆，侧重于那些美国版权已过期的旧作，以电子形式提供世界文学名著。用户可以选择免费的 epub 和 Kindle 电子书，下载或在线阅读。这些馆藏由数千名志愿者数字化并认真校对，为公众使用提供支持。

该平台完全免费，无需注册或付费。自 1971 年起，它一直致力于免费电子书事业，已有五十多年的历史。项目以志愿者为基础，多年来汇集了数百名贡献者。用户可通过普通网页浏览器或电子书阅读器访问，无需安装特殊应用。网站提供多种找书方式，包括按受欢迎程度、主要类别、阅读列表浏览，以及按作者、书名、主题、语言和类型搜索。

馆藏涵盖广泛类别，例如 History 、 Literature 、 Science & Technology 、 Social Sciences & Society 、 Arts & Culture 、 Religion & Philosophy 、 Lifestyle & Hobbies 、 Health & Medicine 和 Education & Reference 。一些热门书目包括 Frankenstein 、 Moby Dick 、 Pride and Prejudice 、 Romeo and Juliet 、 Crime and Punishment 以及 Alice's Adventures in Wonderland 。网站还通过 World Library Foundation 提供自出版电子书专区。

对有声书感兴趣的用户，Project Gutenberg 提供多种选择，包括来自 LibriVox 的人声朗读作品——LibriVox 是一个制作高质量朗读的志愿者社区。另有 Project Gutenberg Open Audiobook Collection，包含 2023 年与 Microsoft 和 MIT 合作生成的近 5,000 个电脑合成标题。此外，网站还有 2003 年的旧电脑合成有声书，质量低于当前技术水平。

项目通过 Distributed Proofreaders 招募志愿者，这是新电子书的主要来源。用户也可通过报告错误、漏洞和错别字或提出修改建议来协助。网站提供多种帮助资源，包括阅读选项、常见问题解答及关于众多主题的详细信息，另设有捐赠 Project Gutenberg 的说明、新书订阅源，以及关于权限、版权、许可和商标的详尽资料。 Project Gutenberg is a library of over 75,000 free eBooks. It focuses on older works for which U.S. copyright has expired, offering the world's great literature in digital form. Users can choose among free epub and Kindle eBooks, download them, or read them online. The collection is made possible by thousands of volunteers who digitized and diligently proofread the eBooks for public enjoyment.

The platform is completely free to use, with no fees or registration required. It has been pioneering free eBooks since 1971, making it over 50 years old. The project is volunteer-based, with hundreds of contributors over the years. Users can access the library through regular web browsers or eBook readers without needing any special apps. The site offers various ways to find books, including browsing by popularity, main categories, reading lists, and search options by author, title, subject, language, and type.

The collection includes a wide range of categories such as History, Literature, Science & Technology, Social Sciences & Society, Arts & Culture, Religion & Philosophy, Lifestyle & Hobbies, Health & Medicine, and Education & Reference. Some of the most popular titles include classics like Frankenstein, Moby Dick, Pride and Prejudice, Romeo and Juliet, Crime and Punishment, and Alice's Adventures in Wonderland. The site also features a section for self-published eBooks through the World Library Foundation.

For those interested in audio books, Project Gutenberg offers several options. These include human-read audio books from LibriVox, a volunteer community that produces high-quality performances. There is also the Project Gutenberg Open Audiobook Collection, which contains almost 5,000 computer-generated titles from 2023 via a collaboration with Microsoft and MIT. Additionally, there are older computer-generated audio books from 2003, though these are of lower quality compared to today's technology.

The project welcomes volunteers through Distributed Proofreaders, which is the main source of new eBooks. Users can also help by reporting errors, bugs, and typos, or by suggesting changes. The site provides various resources for help, including reading options, FAQs, and in-depth information about many topics. Special areas include information about donating to Project Gutenberg, feeds of new eBooks, and details about permissions, copyright, licensing, and trademarks.

Image-blaster 是一款开源工具，能在五分钟内把一张二维图像转换成完整的三维环境，包含模型、空间音频和网格。它结合了多种 AI 模型（如 World Labs 的 Marble 、 FAL 的 Hunyuan 3D 以及 ElevenLabs 的音效模块），并作为 Claude 的技能集，允许用户通过对话命令自动化整个三维资产创建流程。

使用流程很简单：将图像放入项目的输入目录，然后对 Claude 下达"blast it"命令。系统会处理图像并输出三类主要成果：用于动态对象的三维模型（.glb 和 .obj 格式）、用于静态背景的 Gaussian splat（.spz），以及带有基于物理的对象音效的环境循环音效。这使得它在游戏快速原型、建筑可视化、电影前期制作和机器人模拟等场景中特别有用。

工具提供多项高级参数供自定义：可控制面数（4 万到 150 万）、开启 PBR 材质生成、在 Normal 、 LowPoly 或 Geometry 模型类型间选择，并为优化模型指定多边形类型。它支持与主流游戏引擎（Unity 、 Unreal 、 Godot）、 DCC 软件（Blender 、 Maya）以及基于 Web 的框架（Three.js）集成。

在流水线的不同阶段，项目采用了多种 AI 模型：marble-1.1 用于生成可探索的环境，nano-banana（或以 gpt-image-2 作为替代）负责图像编辑任务（如源图清理和目标隔离），Hunyuan 3D 通过 FAL 的 API 生成三维物体模型，elevenlabs-sfx 负责音频生成。模块化设计便于在每一步调整与优化质量。

Image-blaster 由 Neilson K-S 开发，托管在 GitHub，采用 MIT 许可证，社区关注度较高（约 2.5k 星、 232 次 fork）。它在降低三维内容创作门槛方面具有重要意义，使缺乏深厚建模经验的开发者、艺术家和创作者也能生成专业级环境；与 Claude 的对话式界面进一步简化了复杂三维工作流的使用。 Image-blaster is an open-source tool that converts a single 2D image into a fully realized 3D environment, complete with models, spatial audio, and meshes, in under five minutes. It leverages a combination of AI models, including World Labs' Marble for environment generation, FAL's Hunyuan 3D for object modeling, and ElevenLabs for sound effects. The project is designed as a skillset for Claude, Anthropic's AI assistant, allowing users to automate the entire 3D asset creation pipeline through simple conversational commands.

The workflow begins by placing an image into the project's input directory and instructing Claude to "blast it." The system then processes the image to create three main outputs: 3D models in .glb and .obj formats for dynamic objects, a Gaussian splat (.spz) for the static background environment, and ambient looping sound effects with object-specific physics-based audio. This makes it particularly useful for rapid prototyping in game development, architectural visualization, film pre-production, and robotics simulation.

Several advanced parameters allow users to customize the 3D model generation process. These include controlling face count (ranging from 40,000 to 1.5 million), enabling PBR material generation, choosing between Normal, LowPoly, or Geometry model types, and selecting polygon types for optimized models. The tool supports integration with major game engines like Unity, Unreal, and Godot, as well as DCC software such as Blender and Maya, and web-based frameworks like Three.js.

The project uses multiple AI models for different stages of the pipeline. World Labs' marble-1.1 model creates explorable environments, while nano-banana (with gpt-image-2 as an alternative) handles image editing tasks like source cleanup and object isolation. Hunyuan 3D generates the actual 3D object models through FAL's API, and elevenlabs-sfx produces the accompanying audio elements. This modular approach allows for flexibility and quality optimization at each step.

Developed by Neilson K-S, image-blaster is hosted on GitHub with an MIT license and has gained significant community traction with 2.5k stars and 232 forks. The tool represents a significant advancement in democratizing 3D content creation, making professional-grade environment generation accessible to developers, artists, and creators without extensive 3D modeling expertise. Its integration with Claude's conversational interface lowers the barrier to entry for complex 3D workflows.