OrangeBot.AI Digest — 2026-09-27

AI NPCs that talk, remember & act on their own

Compare ChatGPT, Claude & Gemini answers in one tab.

Tiling that feels like it shipped with macOS.

AI agents that resolve IT/HR tickets inside Slack and Teams

The AI assistant that works where you do

Frontier AI intelligence, now at half the price

A new kind of mobile communicator designed for doing

A printed receipt of your day in Claude Code

Conversational Video Agent That Plugs Into Any Stack

Free Read Aloud with Cartesia Voices

A private Markdown notes app for Mac

AI social media manager for LinkedIn. Only authentic content

Keep listening. Searchable memory for your AI agents.

A shared shopping list that remembers what things cost

One tap in the notch locks your keyboard and trackpad

Object permanence and solidity are hallmarks of human cognitive priors. Recent studies show that video generation models, a paradigmatic class of current world models, have begun to show emerged reasoning abilities, making them ideal candidates for building human-like physical intelligence. Do video models have emerged object permanence in them? If not, could we train them with a core-cognition inspired dataset? We introduce WROP (World Reasoning with Object Permanence), a data infrastructure of 150 hand-designed cognitive science inspired tasks, divided into six cognitive categories. We build Blender generators that randomize speed, lighting, camera angle, and other nuisance parameters while preserving each task's cognitive structure, yielding 10,000+ samples per task. We release a 1.5M-sample training corpus and a 300-question exam. On this exam we evaluate 14 video models: 3 reference-to-video, 7 edit, and 4 continuation, among which PWM-WROP, our 16B world model. In a blind pairwise Elo study, PWM-WROP ranks first among continuation models and third overall, behind only a statistical tie between two reference-to-video models. We release the data, exam, model answers, scores, weights, and PWM, our native-PyTorch training stack on AWS Trainium2.

While Large Language Models (LLMs) rely on highly non-linear components, in this work we demonstrate that they exhibit fundamental linearity: when inputs from distinct text streams are linearly combined, the model outputs a superposition of the individual next-token distributions. We term this the Superposition Linearity Hypothesis. We provide evidence that superposition is an intrinsic property of the Transformer architecture rather than an emergent consequence of training; in fact, we observe that it tends to diminish as pretraining progresses. However, we demonstrate that linearity can be substantially restored through lightweight fine-tuning, significantly reducing the divergence between the predicted next-token distribution and the average of the individual next-token distributions. Finally, we introduce a guided decoding procedure that disentangles superposed outputs, enabling the simultaneous generation of two coherent continuations from a single forward pass.

Video generation begins in text space by authoring a cinematic screenplay, then materializes into pixels. As contemporary video generators scale to 30 seconds and faithfully follow complex conditions, the textual prompt largely directs the production, planning how actions, camera trajectories, lighting, and sound unfold across multi-shot sequences. In this paper, we present WanPE, a 397B-parameter prompt enhancement model trained on 1.05M real-world videos to master director-level cinematic planning. WanPE formulates shot-level cinematic plans via video-grounded reverse construction and employs Semantic-Consistency GRPO (SC-GRPO) to faithfully preserve user requirements across shots and over time. To benchmark this capability, we curate WanPEval, a human-annotated testbed covering durations from 5 to 30 seconds across varying intent granularities, supported by approximately 11K blind pairwise assessments. When powering Wan3.0's video generator, WanPE-397B boosts human preference over raw user prompts by 10.66-18.84 points at 5-15 seconds and by a dramatic 50.86 points in the 30-second arena. Ablation studies show that reverse construction demonstrates clear superiority over forward rewriting, while SC-GRPO robustly preserves semantic fidelity across model scales. Ultimately, WanPE leads all evaluated commercial offerings at 5-15 seconds and remains competitive with Seedance 2.5 at 30 seconds.

Humans can effortlessly localize the direction of a sound source and integrate it with visual cues for reasoning, yet this remains challenging for embodied agents. In particular, it is still unclear how to effectively evaluate and model spatial audio understanding in embodied settings. To address this gap, we introduce OmniEchoBench, a unified benchmark for spatial audio-visual perception and audio-vision-language navigation. OmniEchoBench comprises six tasks over 197 real-world spatial audio-visual scenes, 2,972 question-answer pairs, and 900 navigation samples with first-order ambisonics (FOA) audio collected from 30 real-world environments. To enable scalable training supervision, we develop a controllable rendering pipeline for spatial audio. It preserves geometric consistency among sound sources, visual observations, and agent trajectories. Building on this, we propose OmniEcho, a spatially aware omni-modal model. It introduces an FOA spatial encoder alongside a pretrained semantic audio pathway. Extensive experiments show that OmniEcho achieves state-of-the-art performance on spatial audio-visual perception. For our sound-guided navigation, OmniEcho reaches a performance level close to that of traditional vision-language navigation. These results demonstrate that spatial audio can serve as a valuable signal for embodied scene reasoning and navigation, while also highlighting fine-grained spatial localization and distance estimation as important open challenges.

Recent advances in large language models (LLMs) have enabled agents to tackle long-horizon tasks across diverse environments. To further improve agent performance, existing language world models typically predict environment observations, yet reconstructing high-entropy, execution-dependent tool responses offers limited value when real feedback is available. Meanwhile, agents suffer from task-state contamination, where unsupported assumptions and outdated plans persist in history and distort subsequent decisions. We propose the Agent-Editing World Model (AEWM), which models how reasoning and actions shape future task progress rather than simulating tool responses. AEWM combines Action Judge to distinguish Critical, Exploratory, and Noisy decisions with State Revision to edit noisy reasoning--action continuations from the same observed history. EditAct integrates these capabilities with real execution, directly changing the state underlying subsequent decisions rather than merely providing critiques. We train AEWM across Search, Terminal, and Software Engineering through mid-training and supervised fine-tuning. AEWM achieves 70.5\% macro-F1 on our Action Judge benchmark, exceeding the strongest frontier baseline by 10.6 points. Across six benchmarks and three agent backbones, EditAct improves average scores by 3.2--6.7 points over the strongest baseline. Furthermore, rejection sampling fine-tuning on verified EditAct trajectories, termed AEWM-RFT, improves over Self-RFT by 2.2--2.6 points across three domains without online AEWM guidance.

Rufus-Air is an open and reproducible post-training recipe on GLM-4.5-Air-Base (106B-A12B), organized as a serial pipeline of eight stages: SFT, Reasoning RL, Coding RL, Instruction-Following RL, General Agent, Coding Agent, Search Agent, and RLHF. We document the data, reward design, infrastructure, stage order, and stagewise results needed to reproduce the recipe. Stages progress from basic to advanced capabilities and from hard, verifiable rewards to softer judge-based signals. Training builds on open-source components and public data, much of it used as released, without new human annotation or an in-house distillation teacher. Our main findings are that (i) diverse, high-quality SFT establishes a strong capability floor; (ii) difficulty filtering keeps RL prompts within a productive learning range; (iii) reward reliability provides a practical principle for ordering stages; and (iv) infrastructure and engineering choices are part of the recipe, not just an implementation detail. Rufus-Air improves over the official GLM-4.5-Air post-trained release and is competitive with similarly sized open models.

Scientific discovery begins where known problems end. There, AI systems must engage in exploration: framing hypotheses, designing experiments, and iterating on the results. However, evaluating this ability is difficult: (1) how to verify whether a genuinely new hypothesis holds, and (2) how to determine whether a system has discovered it through exploration or merely recalled related knowledge from pre-training data. To this end, we introduce ExplorationBench, which turns the wicked problem of evaluating scientific exploration into a concrete and tractable framework built on verifiable Alien Worlds: their rules are executable, so every answer can be checked exactly, and they conflict with familiar knowledge, so recall alone cannot solve the tasks. The benchmark contains two sandboxes, AlienCode (31 discovery targets, 70 tasks) and AlienLogic (24 discovery targets, 70 tasks). Each sandbox provides a flawed manual, task-specific environmental feedback, and a dedicated tool-call schema. Systems use these resources to explore the sandbox, then solve held-out tasks. We evaluate 10 AI systems and find that the strongest systems can acquire and apply unfamiliar rules, while performance varies substantially across trajectories and continued exploration can stall or reverse earlier gains. ExplorationBench represents a step towards AI systems that can acquire and apply genuinely new knowledge through exploration in unknown environments.

Recently, Large Language Models (LLMs) have been increasingly able to solve advanced mathematical problems, including many that have been open for decades. This opens the door to expansion of mathematical knowledge at unprecedented scale. Yet, while LLMs may be able to conjecture and prove more and more theorems, it remains open whether this new mathematical knowledge is interesting or useful. We define intrinsic interestingness of a theorem as the ratio between the length of its proof and the length of its statement. We show that this correlates strongly with an extrinsic measure of the downstream utility of a theorem. We identify the difficulty of a proof conditioned on a set of premises as a useful primitive for computing these metrics, and train a 27B model that predicts proof difficulty more accurately than frontier general-purpose models. Optimizing for our metric creates a model capable of producing more interesting theorems, while also reducing substantial or full overlap with Mathlib from 91.9% to 30.6%, showcasing the creation of more out-of-distribution math. We show that our system can generate candidate theorems, select the most interesting among them, and iteratively build on a self-expanding mathematical library. These metrics provide a practical and quantifiable signal for ranking conjectures and guiding proof search within formal mathematical libraries. Our framework provides a path towards self-expanding, machine-verified mathematical libraries that can choose worthwhile statements without relying on human-supplied targets.

Task and motion planning (TAMP) problems remain difficult even with full observability and object-centric states because discrete decisions are tightly coupled to geometric, kinematic, and dynamic constraints. Generalized TAMP addresses this difficulty by exploiting regularities across problem instances to reduce planning effort on new instances. However, existing methods require substantial TAMP-specific engineering. We investigate whether coding agents can automate this process by synthesizing programs that generalize across instances. Given a task description and simulator access, each agent chooses how to interact with the environment while developing a program within a fixed synthesis budget. The program is then frozen and evaluated on unseen instances. We evaluate Claude Code (Opus 5) and Codex (GPT-5.6 Sol and GPT-6 Astra) on 28 simulated environments from KinDER and PDDLStream, with object counts beyond those evaluated in the original benchmark. Across all program synthesis methods, we evaluate 980 generated programs on 100 held-out instances each, 98,000 evaluation episodes in total. Overall, we find that coding agents are surprisingly effective at generalized TAMP: all three agent configurations outperform hand-engineered planners, one-shot generation, and an LLM-based generalized planning baseline in mean success (56% to 95% versus 47% for the planners, on the 16 environments where a planner is available). As object counts grow, the agents' programs maintain higher success than the planner, using an order of magnitude less computation per instance on average. Logs show agents using interaction to calibrate physical models, test edge cases, and refine strategies. We release all code, including the full prompts given to the agents. These findings suggest that coding agents are a strong baseline for generalized TAMP.

The rapid progression of large language models is extending AI from passive content generation into the active workflows of engineering and scientific discovery. This shift raises a compelling question: can AI be both the object of development and an active participant in building next-generation AI systems? We explore this question by building Qwen-Planner-Agent within a closed-loop AI-for-AI framework for scalable development and iterative improvement. Mobile planning offers a demanding test of this approach: complex, long-horizon tasks challenge agent reliability, while costly real-device interaction limits development scalability. The framework connects data production, model training, and deployment through a shared action-feedback-verification contract. (i) AI for Data builds a human-gated agentic data flywheel in which specialized agents construct tasks, collect interaction trajectories, curate and balance training data, and use training feedback to guide subsequent data generation. (ii) AI for Training combines a supervised planning cold start with hybrid-environment online agentic reinforcement learning, where we introduce Competence-Aware Reward-and-Advantage Engineering (CARE) to reduce reasoning and tool-use costs while preserving task performance. (iii) AI drives model--harness co-evolution through an execution-evidence-driven loop that orchestrates memory, skills, and tools at runtime and feeds structured action feedback and preserved failure traces back into coordinated model and harness adaptation. Qwen-Planner-Agent achieves the best overall performance among all evaluated models and systems on MobilePA-Bench, improving over its base model across tool use, memory, skills, and sub-agent coordination. Further evaluations of our model show improvements across non-mobile agentic benchmarks while largely preserving general capabilities.

Sparse AutoEncoders (SAEs) offer a promising way to inspect language model representations, but it is still unclear what kind of linguistic structure their latents expose. We use part-of-speech (PoS) categories as a controlled test case to study whether morpho-syntactic information is encoded by individual latents or by structured groups of features. We find that PoS distinctions are highly recoverable from SAE activations, but do not align with one-to-one latent / category mappings. This recoverability is not reducible to lexical memorisation, and Open and Closed PoS classes differ substantially. Categories are supported by compact groups of sparse latents, with substantial variation across tags. These groups remain stable on held-out data, while also showing overlap between related categories. Our results show that SAEs localise morpho-syntactic information in a distributed and category-dependent form rather than through atomic grammatical features.

Deep search requires LLM agents to decompose complex queries, search for evidence, and synthesize grounded answers, yet existing ReAct-style agents suffer from two limitations: role coupling, where one policy must handle planning, evidence use, and synthesis; and context accumulation, where growing search histories introduce noise and obscure useful information. To address these issues, we propose IterSynth, a role-decoupled and summary-based paradigm that alternates between a Planner for identifying information needs and a Synthesizer for integrating evidence into an evolving summary state. This design separates planning from synthesis while using the summary as the persistent state of search, reducing both capability coupling and context noise. To train IterSynth effectively, we further introduce Role-Decoupled Policy Optimization (RDPO) for reinforcement learning, which combines terminal outcome rewards with turn-level rubric evaluations and computes role-specific advantages for more precise credit assignment. Experiments on five long-horizon deep-search benchmarks such as BrowseComp and Xbench-DS show that IterSynth-8B achieves an average score of 50.7, surpassing the strongest prior leq8B agent by +4.2\%. Moreover, IterSynth serves as a model-agnostic prompting paradigm, delivering substantial zero-shot gains over ReAct and similar prompting paradigms on frontier proprietary models.

In this paper, we propose RGBD20K, a novel dataset for facilitating the development of more robust and general RGB-D semantic segmentation by encompassing abundant categories and high-quality annotations. RGBD20K possesses several attractive properties: (1) Expanded Semantic Space. In particular, it covers 160 fine-grained categories, largely surpassing the category diversity of existing popular RGB-D benchmarks (e.g., NYUv2 with 40 classes and SUN RGB-D with 37 classes). With such enriched semantic coverage, we expect to promote the learning of more generalizable segmentation models. (2) Larger Scale. Compared with current benchmarks, RGBD20K offers 20,000 RGB-D image pairs, providing a substantially larger training resource that benefits the development of more powerful deep models. (3) High-Fidelity Annotation. We perform rigorous re-evaluation and correction of existing labels to resolve long-standing annotation noise, resulting in a clean and reliable ground-truth foundation. Furthermore, we propose a novel score-purified fusion (SPF) method, which achieves state-of-the-art performance across all evaluated benchmarks, demonstrating the effectiveness of our approach in leveraging high-quality multimodal information for RGB-D semantic segmentation. The dataset is here: https://github.com/ShaohuaDong2021/RGBD20K/.

Modern Transformer design and compression both reduce to allocating capacity under a budget. The standard scalars for these decisions, #Params and #FLOPs, capture size and compute but not architectural structure: two architectures with identical parameter budgets but different depth-width, head, or FFN allocations receive identical scores yet behave differently. We propose Neural Spectral Capacity (NSC), a closed-form scalar grounded in the singular-value spectrum of each weight matrix. Under standard random initialization, the Marchenko-Pastur law renders NSC computable from the architectural specification alone, with no model instantiation, data, or gradients. Its layer-wise additive structure admits NSC-DP, an exact dynamic-programming solver returning the architecture globally maximizing NSC under resource constraints in seconds on a CPU -- a guarantee that black-box search over existing training-free proxies cannot provide. Empirically, NSC outperforms #Params, #FLOPs, and representative training-free proxies in ranking across seven Transformer and CNN families (on FlexiBERT, τ= 0.505 on pairs differing in #Params by less than 10%, where #Params collapses to 0.082); NSC-DP discovers a Transformer-XL architecture on WikiText-103 that beats the human-designed baseline in 2 seconds; and prunes LLaMA-7B to the best 5.7B model across eight commonsense reasoning tasks without any calibration data, about 5900x faster than the strongest training-free proxy baseline.

Detectors of alignment failures screen deployed language models and score alignment benchmarks. Most are generative judges that spend a decoding pass on every criterion, and classifiers that read token probabilities, such as Llama Guard, still score one fixed label per call. Jev, a model trained with reinforcement learning for calibrated decisions (RLCD), answers many typed questions about one input with calibrated probabilities in a single call. Whether it detects alignment failures has not been measured. We present RLCDAlignBench, which benchmarks Jev on ten alignment failures: sycophancy, jailbreaks, deception, prompt injection, hallucination, privacy violation, social bias, reward hacking, concealing uncertainty, and power seeking. It spans 44 benchmarks and five target models, labelled by each benchmark's scorer and, on two, by humans. Many of these failures are relational, defined against a reference, such as the user's belief or an injected instruction, that the response alone does not reveal. Our key idea is therefore to vary what Jev is asked separately from what it sees: the question's wording and answer type on one side, the fields of the input on the other. A single generic question reaches a median AUROC of 0.886 zero-shot and beats supervised baselines on most benchmarks. Question wording matters little, while context matters more, mostly through fields that encode the label. Jev matches the reference scorer's agreement with human labels, surfaces label defects in existing benchmarks, and costs 63x less than LLM-judge scorers. Code and data: https://github.com/sumleo/RLCDAlignBench.

New York Times : How AI's acceleration created a global policy vacuum, as EU AI Act enforcement lags and regulators remain torn between harnessing AI and fearing its risks —  Ursula von der Leyen, the president of the European Commission, gathered senior officials for an exclusive meeting at a 19th-century palace outside Brussels this month.

Shirin Ghaffary / Bloomberg : Q&A with Mustafa Suleyman on recent AI safety incidents, risks of removing guardrails while testing 10x-larger future models, a cross-industry safety body, more —  The software giant's AI chief also thinks the government should help “drive” the process of evaluating models.

Kate Clark / Wall Street Journal : A profile of Jaan Tallinn, who led Anthropic's $124M Series A in 2021, has advocated for AI safety for over a decade, and donated ~$170M to safety initiatives —  One of Anthropic's earliest investors, Jaan Tallinn dances hip-hop and warns about ‘racing to build a technology that could end humanity’

Qianer Liu / The Information : Sources: China's government asked Alibaba, ByteDance, and others to report on their plans to buy Nvidia RTX Pro 5500, saying it intends to approve the purchases —  China's government has signaled it could allow some domestic companies to buy a new Nvidia chip built for high-end professional computers …

Axios : Sources: Trump plans to host Dario Amodei at a private White House dinner on Sunday, an indication of thawing relations; Trump personally invited Amodei —  President Trump plans to host Anthropic CEO Dario Amodei at a private White House dinner on Sunday evening, sources familiar told Axios.

Mark Gurman / Bloomberg : Meta VR Glasses is the product Vision Pro should have been; sources: Apple Vision team is working on a revamped headset, but the work is still on “life support” —  Apple weighs what to do next with its headset.  —  Meta's new VR Glasses are what Apple should have created for its Vision Pro line.

Lily Kuo / New York Times : In China, recent warnings about existential AI risks are seen as distinctly Western or as a ploy to stop Chinese AI companies from overtaking their US rivals —  In China, A.I. doomsday warnings can feel distinctly Western or like a ploy to stop Chinese A.I. companies from trying to overtake their U.S. rivals.

Wall Street Journal : How effective altruism shaped AI safety and Anthropic; some early Anthropic employees are considering buying remote US land for relocation if AI goes awry —  Effective altruism and its fears that AI could wipe out humanity have wielded big influence at Anthropic

Brian Danga / The Block : New York City-based HIFI, which provides API infrastructure for stablecoin payments and settlements, raised a $37M Series A led by Left Lane Capital —  The funding follows HIFI's role in DTCC's July tokenized-asset trades and a September Visa partnership for stablecoin-funded payouts.

Mike Wheatley / SiliconANGLE : PicoJool, which is developing AI data center interconnects based on vertical cavity surface emitting lasers, raised a $27.5M Series A led by Socratic Partners —  A startup called PicoJool that's backed by former Intel Corp. Chief Executive Pat Gelsinger says it's ready to help ease the strain …

FinTech Global : Numeral, a provider of AI-powered tech to automate sales tax compliance workflows in over 90 countries, raised a $100M Series C led by Insight Partners —  Numeral, the AI-driven sales tax compliance platform, has closed a $100m Series C round as businesses face a widening set of tax obligations across markets and product types.

Madison Mills / Axios : Sources: OpenAI, Anthropic, and researchers are probing tens of thousands of frontier model security incidents, including sandbox escapes and website hijacking —  OpenAI, Anthropic and security researchers are investigating tens of thousands of incidents in which their frontier models took steps …

Robert McMillan / Wall Street Journal : Research: OpenAI agents scanned a UN data hub 16K+ times between April and the end of June, and circumvented a filter that was blocking their requests for data —  Autonomous bots hit public data site more than 16,000 times and circumvented a filter  —  OpenAI agents bombarded …

Tom Wilson / Financial Times : Google Threat Intelligence Group finds dark web marketplaces selling access to AI models, including from Anthropic, Google, and OpenAI, at up to 97% discounts —  Security researchers warn of surge in ‘LLM-jacking’ attacks targeting companies' costly AI resources

Pranshu Verma / Washington Post : Sources: US and Russian diplomats worked to weaken an AI weapons pact at the UN this month, removing a requirement that humans review AI-generated targets, more —  The two nations weakened a range of safeguards in a landmark U.N. effort to try to regulate the use of lethal autonomous weaponry.

距离 2026 年 10 月巴西大选不到两周，Meta 本周短暂屏蔽了巴西现任总统卢拉的 FB 主页以及竞选连任广告，在抗议和投诉之后，Meta 恢复了主页，但卢拉的竞选团队认为此举损害了总统的竞选活动。Meta 是在本周三屏蔽了卢拉的主页，未给出任何理由。卢拉竞选团队投诉称，数字环境在政治辩论中发挥着核心作用，限制卢拉的广告账户损害了其开展竞选活动的能力。此举使卢拉与其他候选人相比处于不平等的地位。卢拉竞选团队以及其所属的劳工党要求 Meta 保留所有数字证据，考虑诉诸巴西最高选举法院。

Bitget 交易所被盗走价值 3.875 亿美元的加密货币。攻击发生在 9 月 24 日 18:31 UTC。区块链情报公司 Arkham 发表报告称，从 18:58 至 19:16 之间的 18 分钟内，价值 2.28 亿美元的数字资产从 Bitget 钱包中转出。价值 1.53 亿美元的 XRP 从一个被识别为​​ Bitget 冷钱包的地址中转出，此外还有价值 6620 万美元的 ETH、3480 万美元的 USDT、1290 万美元的 USDC 以及 1280 万美元的 Tether Gold on Ethereum 等。CEO Gracy Chen 称没有发生私钥泄漏，攻击者入侵了钱包服务的一个关键后端系统，利用它伪造转账信息，启动授权签名流程，将虚拟货币转移出去。

新加坡国立大学研制出至今最精确的原子钟，运行 2600 亿年误差不到 1 秒。新设备是一台光学原子钟，靠镥离子固有而稳定的特性计时。特定频率的光，能把电子送上更高能量的激发态，而这一跃迁频率恒定不变。团队持续观测镥离子的跃迁，把激光锁定在触发跃迁的精确频率上，再以激光振荡作为计时标尺。在最新研究中，团队把镥钟频率测到小数点后 19 位，不确定度仅为 1×10^(-19)，创下所有光学原子钟的最低纪录。他们还造出两台时钟，进行了长达 200 小时的比对。团队表示，镥的特性非常适合这项工作。与原子钟常用的镱、锶等原子相比，镥对温度和磁场波动更不敏感。这些波动会干扰原子跃迁频率，进而影响时钟准确度。镥钟仅需商用激光技术，还能在室温下工作。未来精密计时，镥有望唱主角。

今年 2 月 Debian 13 的龙芯架构移植版 loong13 的维护者在编译打包过程中发现，normaliz 的自带测试会死循环导致打包超时。第一次排查发现原子加指令会在特定情况下丢失更新，但原因未知。今年 8 月，开发者在 AI 的帮助下重新寻找 normaliz 中原子加丢失的问题。他们让 AI 去找最小复现，在这个过程中负责指挥 AI 调查的方向。大概两天后找到了一个稳定的复现程序，才发现事情的根源是：CPU 的原子加法指令，偶尔会不原子。开发者向龙芯报告了问题，两周后龙芯给出了修复的测试固件，确认问题解决。龙芯表示会在国庆节（10 月 1 日）之前发布固件。该问题主要影响使用 LA664 核心的 3C6000/S 和 3A6000。

科学家捕捉到了太平洋西北下方的一个大型构造系统分裂的细节。最新研究发现了一个俯冲带正在活跃分裂。俯冲带形成于一个构造板块俯冲到另一个板块下并深入地球内部，是大型地震、强火山爆发以及大陆和海洋盆地长期变化的源头。现在科学家有机会清晰观察大型构造系统如何走向终结。构造系统会持续活跃数百万年，但不会永远持续下去。利用声波回声，科学家正在观察 Cascadia 俯冲带的分裂。路易斯安那州立大学地质学家、研究主要作者 Brandon Shuck 表示，“这是我们第一次清晰看到一个俯冲带正在消亡的过程。板块没有一下子全部停止运动，而是一块接一块撕裂，产生了更小的微板块和新的边界。因此它不像是一场巨大的火车事故，更像是在看着一节车厢接一节车厢缓缓脱轨。”

微软宣布其电子表格软件 Excel 的单元格将支持输入多个值，而不是以前的只能一个值。Excel 将通过“列表”（Lists）、“单元格内数组”（arrays in cells）和“嵌套数组”（nested arrays）三个功能实现在单元格内输入多个值。微软称，某个项目可能会将“Carlos，Henrietta，Jacob”列为三位负责人，或者 Forms 调查问卷可能会将“2:00 PM；2:30 PM；3:00 PM”作为单个回复返回。借助“列表”功能，用户既可以将这些值保留在同一个单元格中，又能彼此区分开来进行筛选、计算等操作。

2025 年美国政府制裁了位于荷兰海牙的国际刑事法庭，导致了依赖微软软件的法庭工作陷入瘫痪，此事促使欧洲各国政府推动数字主权，减少对美国科技公司的依赖。其中荷兰政府正在本土 Linux 发行版 NixOS 基础上上构建数字工作环境 Digitaal Autonome Werkomgeving Overheid (DAWO) 。DAWO 包含了操作系统、办公套件、协作应用、云服务及管理工具。试点项目正在荷兰政府内部展开。欧洲各地的政府以前尝试过取代微软的软件，但成功者寥寥。现在过于依赖微软软件可能导致服务被彻底切断，对欧洲政府而言风险更大，因此转向替代解决方案比以往更重要。DAWO 先后尝试了 openSUSE 和 Fedora 发行版，最终选择了自主开发的发行版 NixOS。原因是 openSUSE 的母公司 SUSE 频繁出售，而 Fedora 是 Red Hat 主导开发的发行版，而 Red Hat 的母公司 IBM 是美国公司。NixOS 是基于荷兰开发的软件包管理器 Nix，不隶属于任何商业实体，开发者计划明年发布 1.0 版本。

负责发行 Alex Gibney 拍摄的马斯克（Elon Musk）纪录片《Musk》的公司 Bleecker Street 发现，主流社交平台 YouTube、TikTok 和旗下包括 Instagram 和 Facebook 的 Meta 公司，以及马斯克旗下的 X 平台都拒绝投放该纪录片的商业广告。这是一部批评马斯克的纪录片，X 平台拒绝能理解，但 YouTube、TikTok 以及 Meta 都拒绝令发行商感到意外，引发了少数几家公司掌控社交平台压制言论自由的担忧。YouTube、TikTok 和 Meta 都以政治内容相关的理由拒绝投放广告。Bleecker Street 对三家公司提起了上诉，Meta 已经驳回上诉，而 TikTok 和 YouTube 仍在审议中，X 平台则直接拒绝沟通。《Musk》将于 10 月 9 日上映。

Nala Ginrut 写道： HardenedLinux 最近发布了可用于推理CosyVoice的Velum，它用modern C++开发，编译成一个单一的可执行文件，方便部署。 CosyVoice是目前比较优秀的一款 TTS 模型，但其推理程序使用的Python体系比较老旧，需要在部署的时候做一些处理，而且Python依赖占用空间较大，不利于大量能力情况下的Agent部署。要是每个agent能力都要一堆Python十几G的依赖，每一堆还有版本冲突，那就不要卖产品了，不如回家卖红薯。 Velum编译之后只有一个可执行文件，所谓部署更新就是拷贝。一些预处理的模型相关的东西虽然需要用Python生成，但运行时是不需要任何Python的东西。 Velum同时也是一个例证，它是由人类做架构规划，DeepSeek-v4-pro完成的项目，也就是说，DeepSeek足以做这种程度的Vibe。在目前Claude只需要两轮配额就烧干的今天，稍微复杂点的程序，如果不能用DeepSeek做，最后还是要回家卖红薯。 希望以后Codex和Claude也能增强自己的竞争力，把价格向DeepSeek靠拢，让天下无红薯可卖，也未尝不是一件美事。

在电影《教父》中，维托柯里昂（Don Vito Corleone）拒绝参与海洛因交易，称毒品生意太脏。现实中的黑手党并非如此，但对于选择芬太尼还是海洛因等其它毒品，意大利黑手党看起来选择了拒绝芬太尼。这或许可以解释意大利芬太尼过量致死率异常低。2024 年比吗啡强效百倍的合成阿片类药物在意大利仅检测出两例致死事件。相比之下，德国 95 例，美国近 4.8 万例。意大利整体上的毒品消费水平无法解释这一现象。根据欧洲的数据，每年约有 2.1% 的意大利青年使用可卡因，德国的这一比例为 2.2%。芬太尼在意大利尚未造成太大影响，警方也没有查获多少芬太尼，反黑手党检察官 Nicola Gratteri 认为黑手党远离了芬太尼。芬太尼相比海洛因和可卡因致死率更高，客户容易死亡对黑手党而言不是一门好生意，通常意味着更低的利润，因此黑手党选择了拒绝芬太尼。

非洲象会利用数十种药用植物治疗自身和家族成员的疾病。科学家和 Mount Elgon 基金会合作展开了这项研究，他们采访了在肯尼亚 Mount Elgon 地区与大象共同生活和工作的居民、野生动物巡护员和社区长者。根据采访者的描述，大象在身体不适时会选择特定的植物，而母象还会给幼象喂食药用植物。研究人员得出结论，大象会使用 35 种不同的植物，其中 25 种在当地已知具有药用价值。一位野生动物巡护员看到母象使用名为 Angurweet 的植物给幼象治病。Angurweet 可用于治疗包括胃痛在内的多种疾病。巡护员看到母象将植物嚼烂，与乳汁混合，喂给幼象，幼象咀嚼后吐出固体部分。公元三世纪的罗马作家 Claudius Aelian 在其作品《De Natura Animalium (On the Nature of Animals)》中最早描述了大象用植物治疗自己的记录。

全球平均气温正走在比工业化前水平高出 1.5°C 的轨道上。很多人可能会觉得升温幅度不大，可以接受或忍受。但地球绝大部分表面是海洋，海洋的升温幅度要缓慢得多，而陆地则显著得多，居民体会到的升温幅度要高得多。德国研究人员在《Environmental Research Letters》期刊上发表研究报告，指出全球平均气温每上升 1°C 德国夏天气温上升 2.62°C，范围在1.62-3.62°C 之间。当地热浪频率的增加速度会远远超过全球平均水平。

两年前 Zhu Zhili 选择了深圳作为其 AI 电影工作室的办公地点，今年他接到了来自中国各地政府和产业园官员的电话，内容基本相同，即希望将 AI 电影业务带到当地。中国各地正在推动 AI 视频的产业化，类似太阳能、电动汽车和机器人。AI 电影制作人表示，低廉的制作成本是吸引他们投身 AI 创作的主要原因。CCTV 报道 2026 年上半年，AI 短剧的制作成本从每分钟 5,000 元降至几百元。可能和太阳能等领域一样，AI 视频行业也正面临产能过剩。DataEye 的数据显示，今年上半年，抖音上推出了 221,900 部新 AI 视频，其中只有 1,055 部播放量逾 1 亿次。国家电影局已向 90 分钟科幻史诗片《三星堆未来启示录》颁发公映许可证，这是中国大型制片厂制作的首部获批在院线上映的 AI 电影，出品方博纳影业集团表示影片计划于今年上映。

Android 自由软件应用商店 F-Droid 宣布发布 2.0 版本。F-Droid 2.0 对 UI 进行了重新设计，旨在更容易的发现，安装和管理应用。主要界面简化为了三个核心区域：发现，搜索和“我的应用”。类别现在整合进了发现，使其更容易浏览和探索，而“我的应用”提供了一个一站式管理已安装应用、更新和潜在问题的地方。设置和附近交换在顶栏的一级菜单里，但不再占据主界面的空间。F-Droid 2.0 不再将所有游戏放在一起，而是分成了 17 个不同的游戏类型，让用户更容易找到真正想玩的游戏。F-Droid 2.0 还改进了搜索，除了应用名称，现在也能搜索应用描述、类别和翻译内容，新版本改进了中日韩语言搜索，对 CJK 文字系统提供了更好的支持，帮助用户用他们自己的语言找到相关应用。F-Droid 2.0 将在未来几周内推送给用户。