OrangeBot.AI Digest — 2026-08-24

Every local project gets a real address.

Create macOS native automations using plain language

Instruments for composing and experiencing sound as a space

Connect Your AI Spend Directly to Your Roadmap

Drive Antigravity agents from any browser

A shared workspace where people and AI agents get work done

The AI runtime that remembers, learns, and acts everywhere

Treebar gives you one view of every active worktree

Billing that survives a processor shutdown

The memory layer that decides what's worth remembering

Synth, drum kit, a 4-track tape loop, and FX app

Search and act across all your work apps

Control your company's internal AI agents and tools

Proximity-powered dating, social & networking

Build system design for new and existing codebase

Mixture-of-Experts (MoE) architectures significantly expand model capacity without a proportional increase in computational cost. However, optimizing their hyperparameters---particularly the learning rate---at extreme scales of both model size and token budget via sweeping remains computationally prohibitive. In this paper, we propose a compute-efficient, two-step hyperparameter transfer framework that estimates optimal learning rates for training large MoE models by transferring them across scaling model widths, and subsequently extrapolating to trillion-token horizons. First, we formulate a Maximal Update Parameterization (μP) adaptation for MoE architectures utilizing Multi-head Latent Attention (MLA) and the Muon optimizer, demonstrating that optimal learning rates transfer consistently across width-scaled models. Second, we extend this transferability along the token dimension by establishing a predictive scaling law. By applying linear regression to the optimal values derived from small proxy models on limited budgets, we successfully extrapolate the ideal learning rate to massive training horizons (e.g., 10 trillion tokens) with high fidelity (R^2=0.95). Consequently, this indicates that proxy training on small models is sufficient to determine the optimal learning rate for the extensive training of large-scale MoEs. We apply the proposed methodology to pretrain our foundation model (155B total, 17B active parameters) from scratch, and the stable training and evaluation results validate that optimal configurations for full-scale target models can be accurately predicted with minimal ablation costs.

LLMs have evolved from language generators to autonomous agents capable of complex, long-horizon tasks. This evolution has produced paradigms including Prompt Engineering to elicit model capabilities, Context Engineering to manage information access, Harness Engineering to organize external tools and resources, and Loop Engineering to support continual reflection and self-improvement. Yet as tasks grow more complex, individual intelligence faces a fundamental limit: many tasks require heterogeneous expertise, interdependent subtasks, parallel execution, independent verification, and persistent state, exceeding any single agent's organizational capacity. Augmenting one agent's capabilities or context cannot resolve this architectural mismatch; intelligence must instead be distributed across specialized agents and organized at the system level. We call this System Intelligence: an agent system's ability to organize and coordinate multiple intelligent components into a coherent, adaptive whole pursuing a shared objective. Achieving it requires more than adding agents; it demands explicit structures to organize work, coordinate heterogeneous agents, and maintain evolving execution states. We introduce Graph Engineering, an emerging paradigm for next-generation agent systems. Unlike prior paradigms that mainly optimize individual interactions or agent-level behavior, Graph Engineering constructs explicit, dynamic, evolving graph structures representing tasks, agents, and system states. These abstractions provide a unified foundation for organizing complex objectives, orchestrating heterogeneous agents, modeling system dynamics, and enabling scalable agent evolution. We systematically review the principles, methodologies, and applications of Graph Engineering for LLM agents. Related papers, open-source data, and projects are collected at https://github.com/DEEP-JLU/Awesome-Graph-Engineering.

With large pretrained models, existing methods have effectively improved instruction-based video editing. However, most of them rely on an in-place editing assumption. They align the edited video with the given source clip frame by frame over a fixed time span. This pattern fails for open-ended streams, e.g., restyling a live game or applying a camera move to an ongoing shot. In such cases, edits must extend to future frames as they arrive, rather than be applied to a static input clip. In this paper, we study this setting and name it infinite video editing: given a preceding segment and an edit request, a model must generate the next segment that continues the stream while applying the requested edit. This process repeats as an unbounded sequence of edit instructions arrives. This task brings two challenges: the edit must be a faithful continuation rather than a frame-wise rewrite, and generation quality must remain stable as edits accumulate. To address them, we first design a data-collection pipeline for infinite video editing. Based on the collected data, we propose InfinityEdit, a lightweight edit adapter that equips a streaming video generator with unbounded editing ability. The adapter contains three attention modules. History cross-attention guides the denoising frames using the input frames. Temporal causal self-attention keeps temporal cues flowing only from earlier frames to later ones. Edit cross-attention injects the edit request into generation. During inference, the adapter is activated only in the chunk where an edit request arrives. Subsequent chunks are generated by the original model with a reset anchor frame. This scheme applies the edit while preserving the original model's infinite generation ability. Extensive experiments show that InfinityEdit faithfully continues the stream under each edit, and stays stable over unbounded edit sequences.

Parallel reasoning improves the accuracy and robustness of large reasoning models by exploring multiple solution paths, but its computational cost grows with reasoning depth and branch count. Existing methods for managing these parallel paths typically rely on final-answer consensus, local token confidence, or isolated intermediate probes. However, these signals are often delayed, weakly tied to actual reasoning progress, or too noisy for dynamic, branch-level control. To address these limitations, we introduce ParaTempo, a training-free asynchronous parallel reasoning framework. ParaTempo is driven by temporal confidence, a branch-local measure of answer-space convergence. Each branch is periodically probed for a tentative answer probability distribution, and temporal confidence quantifies how sharply the recent intermediate probes concentrate on a dominant answer. Once sufficient evidence has accumulated, ParaTempo drives its entire control process from this single signal: low-confidence branches are pruned, branches that persistently commit to their dominant answer are retired early, freed computation is reallocated by forking new branches, and generation stops globally once the confidence-weighted vote concentrates. Without requiring synchronization among reasoning trajectories, ParaTempo adaptively allocates computation based on branch-level convergence. Experiments on challenging mathematical and scientific reasoning benchmarks show that ParaTempo reduces average latency by 21.8-32.2% and total token usage by 18.1-30.3% while maintaining competitive accuracy. Moreover, temporal confidence exhibits stronger temporal stability and predictive power for future branch convergence than token-level and instantaneous signals.

Recent omni-modal large language models (Omni-LLMs) show great potential as real-time video assistants, which continuously perceive environments and guide users to achieve specific goals. Unlike traditional passive video understanding, interactive assistants should actively combine visual states, user goals, and prior knowledge to provide effective help. Evaluating this is rather challenging, as the model's unpredictable response dynamically changes the user's subsequent actions, which static offline datasets cannot accommodate. To address this bottleneck, we introduce OmniAssistBench. To solve the issue of diverging interaction paths where the same user goal can be achieved through various methods, we provide models with predefined priors derived from the source video, requiring them to guide users along the exact same routes. Since real interaction videos are rare, we construct the dataset by reverse-engineering existing Internet videos. We deduce logical user goals and segment the videos into multi-turn clips to simulate continuous interactions. This rigorous pipeline required over 1000 expert person-hours to build the dataset. Results show that the proprietary Gemini-3-Pro reaches 66.4 out of the max point of 100, while the open-source Qwen3-Omni-Instruct achieves 51.2. Although current models generally understand user inputs, they frequently provide incorrect or incomplete answers. Specifically, they struggle with visual prompts (e.g., hand gestures), fail to maintain historical context during multi-turn interactions, and fail to delay response until the target event. Results indicate substantial room for improvement before models can become reliable assistants.

Hybrid-thinking multimodal large language models (MLLMs) allow a single model to alternate between deliberative thinking and latency-efficient non-thinking inference. Although these modes differ in reasoning budget, their delivered responses should satisfy the same user-facing standard. Correctness alone may not characterize this response quality; we therefore evaluate task accuracy and response-pattern failures as complementary outcomes. We study this gap through response-pattern alignment: whether thinking and non-thinking interfaces preserve acceptable final-response behavior. We introduce PatternEval, a failure-enriched diagnostic benchmark comprising 2,415 multimodal prompts spanning visual perception and grounding, structured image understanding, and multimodal knowledge reasoning. PatternEval tests four recurrent failures: chain-of-thought leakage, response repetition, logical contradiction, and performative reasoning. Response-pattern failures are widespread across models from different providers, with non-thinking inference exhibiting substantially higher failure rates and thereby creating systematic misalignment between thinking and non-thinking interfaces. Motivated by this diagnosis, we develop PatternRM, a response-level reward model, and PatternRL, which introduces pattern-specific penalties during reinforcement learning. Experiments on Qwen3-VL-4B and Qwen3-VL-8B show that incorporating pattern-specific penalties into reinforcement learning can mitigate cross-mode misalignment while incurring a marginal task performance trade-off. Together, PatternEval and PatternRL provide an evaluation-and-training framework for aligning user-visible response patterns across hybrid-thinking interfaces.

On-policy distillation (OPD) transfers teacher capabilities by supervising trajectories sampled from the student's own policy, yet its generalization behavior remains poorly understood, as most studies evaluate OPD on a single domain and on benchmarks close to the training data. We present a controlled study that varies one generalization factor at a time, from in-domain distribution shifts to cross-domain transfer and the multi-teacher setting. We find that OPD transfers a teacher's reasoning behavior rather than its answers to particular problems: training difficulty barely matters, and even problems the teacher never solves are useful. Transfer depends strongly on the origin relationship between teacher and student: same-origin pairs bring the student close to the teacher across languages, reasoning horizons, and even other domains, whereas cross-origin pairs mostly fit the trained distribution. This broad reach is a double-edged sword: since routing prompts to domain experts cannot confine each teacher's influence, combining them yields a mixture-dependent seesaw among their capabilities. These results clarify when OPD generalizes and offer a useful perspective for diagnosing multi-teacher OPD.

Real-world image search queries are multimodal and compositional: ``find this shirt in pink'' specifies an entity to retain, an attribute to modify, and context to ignore. Yet existing re-rankers either compress such multifaceted relevance into an opaque embedding or rely on free-form chain-of-thought that easily omits or hallucinates fine-grained constraints. Drawing on rubric- and checklist-based evaluation from NLP, we recast multimodal image re-ranking as a semantic constraint satisfaction problem and propose EviRank, which parses any query - text-only, image-only, or composed - into a unified evidence package: typed criteria across six semantic slots (e.g., entities, attributes, relations), each labelled required, forbidden, or ignorable. Re-ranking then reduces to evidence-conditioned verification, combining deterministic rubric scoring and evidence-grounded listwise comparison in a single training-free procedure. The explicit evidence can further serve as structured supervision for optionally distilling a lightweight student. Across five benchmarks spanning text-to-image, image-to-image, and composed image retrieval, EviRank achieves state-of-the-art performance, and the distilled student preserves over 90% of the teacher's capability at substantially lower cost.

Semantic vision encoders have become a central visual interface for multimodal understanding and semantic conditioning in image generation. However, their final tokens discard fine-grained visual details, leading to poor pixel reconstruction and limiting their use in reconstruction-sensitive tasks such as image generation and editing. In this work, we ask whether understanding, generation, and editing can be modeled in a single visual representation space built from a pretrained semantic ViT. We show that the frozen Transformer blocks of a semantic ViT are not intrinsically unable to preserve visual details. Instead, the original patch parameterization drives the representation toward semantic abstraction, making fine-grained information difficult to recover from the final tokens. Based on this observation, we introduce Patch Reparameterization, which preserves the original semantic pathway while adding a reconstruction-aware patch embedding that provides fine-grained visual information to the same frozen ViT blocks. The resulting unified representation preserves multimodal understanding while enabling high-fidelity image reconstruction and a favorable reconstruction--generation trade-off. We further scale this representation into UniSpace, an 8B Mixture-of-Transformer-Experts model that performs understanding, generation, and editing in the same visual space without a separate VAE pathway. System-level evaluations demonstrate practical text-to-image generation and instruction-based image editing, showing that a reparameterized pretrained ViT can serve as a unified visual interface for scalable multimodal modeling.

Training-free block-sparse attention can accelerate video transformers, but row-wise attention concentration does not by itself specify an executable sparse operator. Queries sharing a block route may have poorly overlapping supports, while retained attention mass alone does not determine the post-softmax error from skipped interactions. We show that partition geometry affects both pooled support and the predictability of the remaining residual from the sparse output. We introduce SparsePR, which combines Response-Coupled Partitioning with Probe-Fitted Residual Reconstruction. Sampled-query key responses form paired K/V groups, whose centroids induce query-response coordinates for shared routing. A small set of exact query rows then calibrates a call-specific affine correction from the sparse output within the output subspace observed in the probe residuals. Across four heterogeneous video generation and world models, SparsePR consistently reduces attention-reconstruction error. Ablations show that probe fitting accounts for most of this reduction, while response-coupled partitioning lowers hard-drop error and improves reconstruction under a finite probe budget. SparsePR preserves generation quality at 22.0-26.0% realized executed-pair density while achieving 1.48x-2.61x end-to-end speedups. Project page: https://pardistaghavi.github.io/SparsePR-website/

Agents learn to act through interaction with environments, yet the environments used for training are often manually constructed or synthesized around predefined tasks and benchmarks. This task-centric paradigm makes it difficult to scale environments that reflect realistic and evolving workflows where diverse tasks can naturally emerge from the underlying world. We introduce AgentMercury, a scalable framework for synthesizing executable environments from high-level business scenarios. Rather than constructing an environment for a specific task, AgentMercury first instantiates a persistent world with entities, services, tools, state, and executable cross-service invariants, from which diverse tasks and interaction trajectories can subsequently emerge. We construct 4,783 executable environments spanning 14 industries and 50 countries, and use them as training substrates for reinforcement learning. Despite being generated without targeting the evaluation benchmarks, policies trained on these business-oriented environments improve substantially on both enterprise workflows and out-of-domain benchmarks spanning reasoning, coding, scientific computing, and tool use. In our experiments, Qwen3.5-4B improves from 12.3 to 15.7 on EnterpriseOps-GYM and from 45.9 to 56.0 on AIME26 after training on AgentMercury environments. We further show that the construction process itself can be learned: fine-tuning Qwen3.5-35B-A3B on construction traces increases executable-world authoring success from 3.3% to 83.3% on held-out business scenarios. These results show that scenario-grounded environments can provide useful and generalizable learning signals beyond benchmark-specific training, while their construction can itself become a learnable capability.

Small language models are usually built like large ones and then squeezed onto a CPU afterwards. We did the opposite: we fixed the target first, one user, one token at a time, 4-bit weights, ordinary CPU, and chose the architecture to suit it. The result keeps full attention in only 6 of its 18 blocks. The other 12 use short convolutions whose memory is two timesteps wide no matter how long the conversation gets, so two thirds of the network never re-reads a growing cache. Trained from scratch on 59.9B tokens, the model scores 47.31 on a five-task benchmark against a bar of 42.20 that was fixed before training began. It beats GPT-2 124M, Pythia-160M, OPT-125M and GPT-neo-125M, all trained on three to six times more data, and exceeds MobileLLM-125M's published score despite that model seeing a trillion tokens. Validation bits-per-byte is 0.8685. To check the architecture rather than the training recipe, we trained a conventional all-attention model of the same size on the same data, and wrote down the winning condition before scoring either. The hybrid won the chosen quality metric by 0.81%, matched it on downstream tasks, produced a 6.3% smaller 4-bit file, and decoded 1.76x faster at 2048 tokens of context, 2.08x against an external model of similar size. In every measurement the speed advantage is near zero at an empty context and grows with length, which is what the mechanism predicts and what a merely leaner model would not show. A simple bandwidth calculation predicts only 1.17x, so memory volume alone does not explain the gap. We also report what did not work: an unmitigated 4-bit quality cost, roughly half the convolution channels ending up inert and impossible to remove, and a vocabulary larger than this model size warrants.

Deploying vision-language models (VLMs) on mobile devices is challenging due to their significant memory and compute requirements. We present a framework for quantizing VLMs for efficient inference on resource-constrained hardware. Our approach combines a quantization pipeline that uses the model itself to generate training data and does not require access to the training setup, with a novel 2.7-bit-per-parameter format supporting efficient execution on Arm CPUs. We validate our approach by compressing the Llama 3.2 11B Vision Instruct model to 3.7 GB with 8-bit activations, preserving strong performance on a set of standard visual question answering tasks.

Human-centric intelligence is evolving in the foundation-model era, with growing emphasis on scale, transferability, and general-purpose modeling. Yet it has not fully integrated with foundation models to achieve the comparable progress seen in them. More importantly, recent advances across this broad landscape remain fragmented across tasks, modalities, and research communities, leaving their intrinsic conceptual and methodological connections unclear. To bridge these divides and rethink human-centric intelligence in the foundation-model era, we introduce a full-spectrum human context taxonomy that integrates six interconnected levels by viewing humans as observable subjects through visual appearance and spatial geometry, as dynamic actors through kinematic dynamics and interaction modeling, and as situated agents through world simulation and embodied agency. We next present the methodological foundations of the field, covering human-centric data families, computational architecture paradigms, and representative training and inference optimization strategies. We then systematically review representative methods across these levels and organize the associated datasets, benchmarks, and evaluation metrics. We further discuss open challenges and promising research directions toward human-centric intelligence that is scalable, trustworthy, physically grounded, and deployable, aiming to provide a coherent framework and practical reference for advancing the field. Finally, we provide a systematically organized and continuously updated collection of human-centric AI literature and resources on our project page.

Simulating realistic user shopping behavior underpins offline evaluation and reinforcement learning in e-commerce scenarios. While recent LLM- and VLM-based simulators have made encouraging progress, reproducing a real browsing session remains difficult for two reasons. (i) Memory Challenge: a shopping session spans dozens of pages, yet existing agents either discard long-range observation histories, losing the evolving user state, or naively concatenate them, overwhelming the context window and even degrading simulation quality. (ii) Optimization Challenge: current user simulators are typically supervised to match each logged action via imitation or step-level rewards; the resulting sessions often display unrealistic patterns, such as over-exploration or excessive passivity, which per-step supervision can neither detect nor correct. To address the above challenges, we present RecVerse, a GUI-grounded simulation agent that perceives pages through screenshots and produces faithful multi-turn trajectories. For the memory challenge, RecVerse adopts a cognitive-inspired hierarchical memory: Working Memory for short-term focus, Episodic Memory for in-session traces, and Preference Memory for high-level intent, with memory updates treated as actions so that the agent adaptively learns when and what to memorize. For the optimization challenge, RecVerse is optimized with a trajectory-level RL objective that scores entire sessions, aligning both macro-level action-type distributions and micro-level shopping intent with real users. We further release USB (User Simulation Benchmark), an interactive e-commerce GUI trajectory dataset for multi-turn user simulation. Experiments show that RecVerse significantly outperforms existing baselines in both behavioral fidelity and intent consistency.

Bloomberg : Sources: smart ring maker Oura and some of its backers seek to raise up to $3B in a US IPO that could take place as soon as September and value it at over $16B —  Oura Health Oy, a maker of smart rings that track health, fitness and sleep, and some of its backers are seeking to raise …

Dan Primack / Axios : Source: robotics startup Generalist, which released its GEN-1 model to complete physical tasks in April, raised ~$200M led by 8VC, after raising $400M in June —  Generalist has quietly raised around $200 million in new funding, just two months after the robotics AI startup raised $400 million.

Colin Campbell / Axios : Canada-based logistics software company Descartes acquires Tai, a California-based transportation management system provider, for $100M in cash —  Logistics software giant Descartes has acquired transportation management system provider Tai for $100 million in cash, the companies announced today.

Grace Kay / The Information : At his first Cursor all-hands, Musk said Grok needs to catch up, AI will become impossible for humans to control, Anthropic is leading the AI race, and more —  Earlier this month, on the day SpaceX announced it had completed its $60 billion acquisition of Cursor, Elon Musk called …

Michael Kan / PCMag : Mercury Research: AMD's share of x86 client CPU shipments tops 30% for the first time, hitting 30.3% in Q2, up from 21.1% two years earlier, vs. Intel's 69.7% —  For the first time, AMD's share of the x86 client CPU market, covering both mobile and desktop, has crossed 30% …

Raphael Boyd / The Guardian : The UK backs Ofcom to take “tough action” under the OSA against social media sites failing to remove dangerous driving videos, after a crash killed seven people —  Downing Street condemns ‘utterly disgraceful’ clips and says sites are legally obliged to take them down

Cassandre Coyer / Bloomberg Law : Alabama AG Steve Marshall launches an investigation into OpenAI's security procedures following the Hugging Face breach in July —  Alabama Attorney General Steve Marshall launched an investigation into OpenAI's security procedures after one of its AI agents escaped a testing environment and hacked AI firm Hugging Face in July.

Financial Times : The UK becomes the first foreign nation to gain access to Ukrainian combat data used to train AI models to strike Russian targets, as part of an AI partnership —  Trove of combat imagery used to train AI models to identify and strike Russian targets  —  Britain will become the first foreign country …

Andrew Kreighbaum / Bloomberg : The Trump administration proposes a $103,265 fee on H-1B applicants hired within the US after a court blocked a more limited fee on workers from outside the US —  The Trump administration wants to add an expansive $100,000 fee on the H-1B program, applying it to individuals hired within …

Christopher Mims / Wall Street Journal : A look at startups like General Intuition working on large action models, aka world models, which are trained on videogames and simulations, to pilot robots —  Engineers and investors pile into world models, aka ‘large action models,’ to do for robotics what ChatGPT did for writing and coding

The Register : Nvidia says its Groq 3 LPX racks delivered 3,400 tokens per second in an Artificial Analysis benchmark running Gemma 4 31B with a 100,000-token input sequence —  Nvidia's $20 billion bet on Groq's LPU tech sure looks like it was a good one.  On Monday, the GPU giant offered the first glimpse …

Tom Rees / Bloomberg : The UK's Office for National Statistics: UK digital infrastructure investment crossed £11B in 2025, hitting dot-com boom levels, amid an AI data center surge —  Businesses pumped more than £11 billion ($15 billion) into Britain's digital infrastructure last year in the rush …

Tracy Qu / Wall Street Journal : Temu and Pinduoduo owner PDD reports Q2 revenue up 8.1% YoY to ~$16.72B, missing est. of ~$17.17B, and net profit down 12% YoY to ~$4.04B, above ~$3.63B est. —  Rising competition from livestreaming and social e-commerce is challenging PDD's market position

Michael Nuñez / VentureBeat : IBM says it is developing a dual-architecture processor for IBM Z and LinuxONE systems that can natively execute both IBM's z/OS workloads and Arm applications —  IBM is announcing today at the annual Hot Chips conference what may be the most consequential change to mainframe architecture in decades …

Mike Wheatley / SiliconANGLE : Nvidia says its inference accelerator Groq 3 LPX has entered full production and Nebius has signed on as the first customer, and SpaceX will deploy Vera CPUs —  Chipmaker Nvidia Corp. says its dedicated artificial intelligence inference accelerator Groq 3 LPX has now entered full production …

微软曾从 2013 年起向全世界的小型非营利组织免费提供 Microsoft 365 Business Premium，但在 2025 年初它宣布将从 2025 年 7 月起停止提供免费授权，转为提供折扣价付费订阅。6 月 11 日前如果没有转为付费的账号内相关数据将被删除。根据 Slate 的报道，没有注意到微软邮件通知的非营利组织遭到重创，17.1 万非营利组织储存在 OneDrive 中的数据被删除，全部丢失。一名经营着一家儿童医保组织的人士称，他翻遍了该组织的邮件存档和垃圾邮件，没有看到任何授权终止的通知。在 Reddit 和微软官方的技术社区论坛，很多非营利组织都表示他们没有收到提前通知就被删除了数据。还有很多人报告在数据删除后才注意到微软的警告邮件。微软没有解释为什么客户的数据无法恢复。

从 2009 年的 Wi-Fi 4 起，每一代 Wi-Fi 的一大卖点都是相比前一代数据速率提升多少。Wi-Fi 5 的最大数据速率十倍于 Wi-Fi 4，到 Wi-Fi 7 每频段理论最大吞吐量达到了 23Gbps，对于大部分用户的网速是绰绰有余了。也许是时候放慢速度了。正在开发中的 Wi-Fi 8 将与 Wi-Fi 7 维持基本相同的最大数据速率、支持相同数量的空间流，使用相同的 4096-QAM（4K-QAM）调制，工作在相同的频段，支持相同的 320MHz 信道带宽。Wi-Fi 8 将专注于提升可靠性，目标是在不同的信噪比 (SINR) 水平下吞吐量提升 25%；95% 的请求延迟降低 25%；MAC 协议数据单元丢失率 (MPDU) 降低 25%。Wi-Fi 8 标准预计将于 2028 年定稿。

Anthropic 的美国客户正使用其最强大 AI 工具的更廉价替代品，引发了外界对其高投入商业模式的质疑，而市场普遍预计 Anthropic 即将进行史上规模最大的 IPO。根据支付服务集团 Ramp 收集的 7 万家公司支出数据，Anthropic 规模最大、成本最高的模型 Fable 5 发布两个多月后，其支出占该公司各类工具总支出的比例仍仅约为 11%，目前已趋于稳定。这打破了企业用户默认选择最强大模型的惯例。分析师和投资者表示，这一变化主要是由于 Fable 5 价格高昂，以及旧款模型能满足大部分企业需求。如果这一转变持续下去，可能会彻底改变前沿 AI 实验室的商业模式。前沿 AI 实验室一直将数十亿美元的研发经费投入到训练规模越来越大、技术越来越复杂的模型上。Fable 的低需求和低市场接受度，加剧了Anthropic 在 IPO 前的不确定性。投资者预计 Anthropic 的 IPO 估值将达到 2 万亿美元或更高，它最早可能会在下个月上市。

Valve 由前微软员工 Gabe Newell 和 Mike Harrington 创办于 1996 年 8 月 24 日，该公司的第一款游戏是备受好评的《半条命》，它被认为对 FPS 这一游戏类别产生了深远影响。Harrington 于 2000 年离开 Valve。Valve 在 2003 年开发了 Steam，之后随《半条命2》的发布强制捆绑推送给玩家，此举曾引起广泛批评，但它最终成为 Valve 最主要的收入来源，它过于成功以至于部分导致 Valve 大幅减少了游戏新作的开发速度。Valve 的游戏开发主要集中在 2013 年前，它发布了一系列享有盛誉的游戏作品，包括《反恐精英》系列、《传送门》系列，《求生之路》系列和《Dota 2》。2020 年代之后的新作包括 VR 游戏《Half-Life: Alyx》，以及仍然在封闭测试的多人游戏《Deadlock》。Valve 还开发了一系列硬件产品，包括 Linux 游戏机 Steam Machine、掌机 Steam Deck 和 VR 头显 Valve Index。

今年 10 月在法国举行的国际计量大会将讨论用闰时取代闰秒。闰秒需要经常性修改，会导致混乱，而引入闰时后数百年内无需修改。协调世界时（UTC）基于用原子振动实现的高精度原子钟来确定，成为 1 秒长度的标准。不过由地球自转时间决定的 1 天长度会发生不规则变化，因此会产生偏差。地球的自转速度根据潮汐涨落以及地球内部地核运动而波动。地球的一天即地球自转一圈，大约需要 86,400 秒。闰秒的作用就是调整与原子钟的偏差，当偏差累积即将超过 0.9 秒时，就会在一天中增加1秒。1972 年引入闰秒以来已修改了 27 次。然而对于需要同步时间进行自动控制的金融交易、交通、电力等高端通信系统而言，哪怕是微小的误差也可能导致整个网络陷入混乱。2012 年增加闰秒时大型网络服务等发生故障。因此美国 IT 巨头 Google 和微软等方面为避免影响而引入各自的调整方法。各公司采用的“时间系统”各不相同，导致闰秒作为标准时间基准的地位动摇。

TikTok 与字节跳动同意支付 4 亿美元与美国司法部就儿童网络隐私争议达成和解。美国司法部 2024 年代表联邦贸易委员会（FTC）起诉 TikTok，指控 TikTok 允许数百万名 13 岁以下儿童在家长不知情或未同意的情况下创建账户，并设置障碍，让家长难以要求删除这些账户。诉讼指 TikTok 违反儿童网络隐私法，在未取得家长同意的情况下收集儿童个人资料。TikTok 与字节跳动同意达成一项 4 亿美元的和解协议，以解决美国司法部对 TikTok 的指控。

中国载人航天工程网周日发表一句话声明：“本着稳妥可靠、万无一失的原则，经综合研判，嫦娥七号任务不满足发射条件，不能在今年预定窗口实施。”官方没有解释原因。嫦娥七号原计划的发射窗口是 2026 年 8 月 24 日上午。下一个发射窗口可能是明年 2 月或 3 月。推迟发射的一个可能原因是台风娜拉的影响，娜拉正在海南文昌发射场附近海域徘徊，在发射窗口关闭前，天气状况不太可能好转。嫦娥七号将登陆月球南极，寻找水冰，探测器对登陆地区的日照条件要求较高。

摩托罗拉正与 Android 安全加固项目 GrapheneOS 合作，准备在 2027 年推出支持 GrapheneOS 的智能手机。GrapheneOS 目前只支持 Google 的 Pixel 智能手机，原因是 Pixel 使用的 Tensor SoC 提供了项目所需的安全特性，但不是所有高通手机芯片都提供了类似的安全特性。高通只有最昂贵的 8 系列芯片才具有类似安全功能。这意味着摩托罗拉只有高端旗舰型号才能支持 GrapheneOS，而此类型号价格不菲。GrapheneOS 项目透露其定价甚至比 Pixel 手机更贵。最新的 Pixel 11 系列起售价为 900 美元。 摩托罗拉未来的中低端型号可能会支持 GrapheneOS，但在短时间内不会发生。

在论文写作中使用 AI 的情况可能比以前认为的更普遍。根据发表在 arXiv 上的一篇预印本，对 PubMed Central 数据库收录的 2025 年全年发表的英文论文的分析显示，LLM 的使用比例高达 77%，而 2024 年这一比例为 52%。这一结论与 2025 年进行的一项调查结果一致。调查显示，71% 的研究人员表示在使用 AI 辅助写作——实际使用率可能高于承认的水平。2025 年 12 月发表的论文有近九成出现了 AI 辅助写作的痕迹。最新研究还发现，AI 辅助写作在摘要、引言和讨论部分比在方法和结果部分更为常见。

美国皮尤研究中心的数据科学家使用 AI 检测工具分析了过去五年的 50 万英文网页，估计其中 AI 撰写内容的比例。所有这些样本网页中，十分之一有明显 AI 创作痕迹。在 OpenAI 的 AI 工具 ChatGPT 于 2022 年底推出之后，AI 内容的比例在不断上升。2026 年 7 月的快照中，逾三分之一 ChatGPT 推出后发布的网页显示出 AI 创作的痕迹。在主要顶级域名中 AI 创作文本的分布并不均匀。ChatGPT 最初发布时 AI 语言模式在 .com、.org、.edu 和 .gov 等域名上的出现频率大致相同。但 2026 年的样本中十分之一的 .com 域名显示出 AI 创作的痕迹，是 .org 域名（4.6%）的两倍，.edu 或.gov域名（1%）的十倍。相比 2023 年，今天的网页：破折号的使用频率增加了一倍；牛津逗号的使用频率增加了 63%；AI 常用词如 delve、interplay 或 testament 的使用频率增加了一倍以上；it's not just X, it's Y 的负面/对比平行语言模式几乎是之前的三倍。

免费游戏并不真免费，游戏内虚拟物品的售价可能比一款典型 3A 游戏的定价高得多。免费游戏内的虚拟交易通常被称为微交易，然而很多微交易动辄上百美元，“微”之名其实名不副实。根据《Frontiers in Public Health》上的一项研究，研究人员基于 2308 名 10-19 岁奥地利学生（48.8% 为男性；平均年龄 14.3 岁）的数据，分析了游戏内消费的分布情况及其与游戏成瘾、赌博成瘾和社会经济地位的关联。他们发现，支出前 10% 的青少年贡献了 61.4% 的总消费额；重度消费者比轻度消费者更容易出现游戏成瘾和赌博成瘾；高消费额并非仅限于家庭经济条件较好的青少年。研究人员认为，免费游戏这种消费模式与赌博行业高度相似，但相比高度监管的赌博行业，游戏行业受到的监管要少得多。他们呼吁为保护青少年加强监管。

天文学家首次利用天然极端磁场，即对磁星信号的观测研究，为 90 年前德国物理学家海森堡作出的量子电动力学预言提供了迄今最有力的证据：看似空无一物的真空，在极端磁场中会改变光的传播方式，产生真空双折射效应。研究人员利用 NASA 成像 X射 线偏振探测器，对编号 1E 1547.0-5408 的射电磁星进行了观测，并结合国际空间站 X 射线望远镜以及澳大利亚帕克斯射电望远镜的数据展开分析。磁星是大质量恒星爆炸后留下的致密残骸，拥有宇宙已知最强的磁场。理论认为，当磁场足够强时，真空不再是无性质的虚空，而是表现出类似晶体的光学性质，使不同偏振方向的光以不同速度传播，从而改变 X 射线偏振特征。1E1547.0-5408 每两秒自转一周，是少数能持续发射射电波的磁星之一。2025 年 3 月和 4 月，团队将探测器指向这颗天体。结果显示，其 X 射线偏振度接近同类天体的 3倍，且偏振方向与恒星磁场方向完全一致，这一特征符合海森堡等人预测的真空双折射效应。

为期五天的世界人形机器人运动会于周六在北京开幕。运动会共设 51 个项目，包括 30 项体育竞技和 21 项场景化竞赛。超过 40% 的项目要求机器人完全自主运行。在 8 月 22 日的首日赛事中，两台机器人跑出了比人类百米世界纪录(9.58秒)保持者博尔特(Usain Bolt)更快的成绩，相比去年百米短跑仍然耗时 20 秒以上的机器人，可谓进步巨大。另一台机器人则在 400 米短跑中实现了 39.7 秒的成绩，超越了南非运动员 Wayde van Niekerk 43.03 秒的世界纪录。不过，在冲过终点线后，这些机器人的制动能力依然存在很大缺陷。现场画面显示，机器人纷纷撞向十几米开外的巨大软垫，然后跌倒在地，多台机器人翻倒后甚至出现了明显损坏。这次运动会还设置了在模拟工厂、餐厅、办公室及紧急情况场景下测试机器人性能的比赛项目。比如人形机器人能否在处理包装和仓储作业的同时，可靠地完成诸如线缆连接等精密任务？它们能否应对角度不当的线缆、刚好够不到的物体、发生位移的包裹等工厂中常见的困难？这些任务旨在评估它们在那些不那么引人注目的岗位上像人类一样工作的能力。

俄罗斯安全公司卡巴斯基的研究人员报告他们发现第一种针对汽车的 Android 恶意程序。恶意程序通过基于 Android 的兜风出行汽车主机（head unit）固件的内置更新程序传播，被认为与 MoYu Group 黑客组织有关，该组织与 BADBOX 僵尸网络有关联。卡巴斯基称它已经通知了兜风出行，对方表示已修复相关安全问题。这一汽车恶意程序传播案例类似廉价电视盒，攻击者旨在创建住宅代理僵尸网络，因此使用了相同的网络基础设施。

数百万棵垂柳和白杨树将北京装缀成一个绿色的大都市。然而根据《Science Advances》上发表的一项研究，柳树和杨树释放出的化合物是城市空气污染的重要来源。广州暨南大学的研究人员最初想要了解人类活动对臭氧污染的影响，结果意外发现城市植被是臭氧的重要来源。植物会释放出挥发性有机化合物，作为植物光合作用的副产品，被广泛种植的柳树和杨树会释放出大量的异戊二烯。研究团队发现，北京 35% 的树木会释放异戊二烯。研究人员在北京各地采集空气样本，测量挥发性有机化合物浓度，在城市各监测站收集臭氧数据。研究发现，在 2021 年 5 月至 7 月期间，植物排放的化合物约占北京总排放量的 10%，其余来自人类活动如汽车尾气和工业化学品。植物释放出的挥发性有机化合物与大气中的羟基反应生成过氧自由基，过氧自由基再与空气中的氮氧化物（NOx）反应生成化合物，这些化合物在阳光照射下会转化为臭氧。研究发现，植物排放的有机化合物占最终生成臭氧的化学物质的 52%，其中异戊二烯是主要贡献者。人类活动产生的挥发性有机化合物总排放量高于植物排放，但对生成臭氧的贡献远小于植物。研究人员还调查了 24 个特大城市种植的树种，发现澳大利亚悉尼和墨尔本所种植树的异戊二烯排放量预计会高于北京，悉尼有 65% 的树木会排放异戊二烯。