OrangeBot.AI Digest — 2026-09-03

A free reader that answers your questions and reads aloud

Build agents that heal themselves.

The Best AI Browser Build both for You and your Agents.

Control Every AI Request

Ask your computer anything you saw or heard

One terminal for you and your AI agent

AI Powered Canvas for planning visual projects

The only app that owns your whole menu bar, in and out

Turns coding agent sessions into better rules and skills

Fix grammar in any Mac app with free on-device AI

Give everyone their own AI workforce in one platform

Quick Look code previews with 700+ editor themes

Recast motion with your characters, locations & products

Claude Cowork for high-volume GTM workflows

Turn text, pics, video, + 3D into camera-controlled HD video

Autonomous agents are beginning to carry out machine-learning (ML) research end to end. These agents combine a model backbone with a harness for planning, execution, memory, and verification, but this architecture still leaves domain-specific know-how outside the agent. We call this missing layer operational knowledge, the know-how that separates knowing a method from making it work. That knowledge is not absent from the field. It appears in repositories and papers, but in forms written for human readers and too large to load during a task. Once distilled into compact, verified skills, this knowledge can be reused across tasks rather than rediscovered during each run. We present DisCo, a skill-powered research agent that creates skills and uses them during research. Its distillation runs in two complementary forms: task-agnostic, condensing the field's widely used repositories into reusable skills, and task-oriented, producing the skills a concrete task calls for. The former, applied across the open ecosystem, yields the AREX-Skill Library, with 5,000+ verified skills distilled from 1,000 widely used ML repositories and organized into 20 areas and 178 capability families. With the GPT-5.5 backbone, research harness, and downstream execution budget held fixed, the skill-equipped research agent scores 134.3% higher on MLE-bench, 34.4% higher on PaperBench, 9.2% higher on FrontierCS, and 14.0% higher on PassNet than the same agent without skills. These gains come from adding distilled operating context under that fixed setup.

As agents move from research prototypes to deployed tools, their capability increasingly depends on model-external execution infrastructure, commonly termed the agent harness. Changing this harness while holding model weights fixed can substantially alter task performance. Current agent evaluations typically report downstream performance under a chosen harness, leaving a model's ability to develop the harness itself comparatively underexplored. We introduce HarnessDev, a benchmark that shifts the unit of evaluation from task outputs to runnable infrastructure. HarnessDev covers two stages. In Creation, the agent starts from a minimal seed and a small number of cases, then builds a complete execution system. In Evolution, it starts from its own created harness and iteratively revises it using downstream execution feedback, with the goal of improving benchmark performance. We then evaluate each constructed harness on capability (task success on held-out benchmarks) and efficiency (execution-token cost). The reported Creation results cover six creator LLMs, four domains, and five downstream benchmarks totaling 2,207 unique downstream instances, with hidden evaluation tasks withheld from development. We find that generated harnesses remain substantially behind mature human-engineered references on code and on search and research, while matching or exceeding the selected references on writing and machine-learning experimentation, with large variation in execution cost. Evolution produces some performance gains, but they are unstable and transfer only partially to held-out tasks. Experiments with a fixed runtime model further show that the gains depend strongly on the model executing the harness, indicating limited transfer across models.

Many important forms of human learning begin with a vague goal, such as "become a better physicist" or "improve at research." Learners must interpret the goal, identify capability gaps, decide how to learn, and determine whether they have actually improved. In contrast, existing work on LLM self-evolution typically begins with tasks and evaluation metrics specified by humans, reducing self-evolution to optimizing an explicit objective rather than deciding what and how to learn. We introduce ASPIRE, a benchmark for vague-goal-driven self-evolution. ASPIRE provides only a natural-language capability goal while downstream evaluation tasks remain hidden. The agent must operationalize the goal by choosing data and update methods, constructing training and validation signals, and deciding when to evaluate. ASPIRE supports both model-weight and agent-harness evolution in a unified interactive environment and evaluates the resulting systems on a hidden, expert-authored set of 520 items spanning six goals. Our experiments show that vague goals redirect search effort toward goal interpretation. Current agents routinely complete training and harness-editing loops, but weight-level gains remain sparse and unstable, and the strongest evolved harness remains below the engineered Qwen-Agent reference. Agents often train on mismatched data and trust narrow self-evaluations, so local gains fail to transfer to hidden evaluation and continued search and training can erase earlier improvements.

We introduce SolarWM, a fully open foundation for building interactive video world models from data preparation through long-horizon inference. Training across heterogeneous data sources and video backbones is challenging: datasets differ in temporal scale, camera geometry, visual quality, motion, and captioning styles, while video generators use distinct representations and architectures. Naive data mixing and model-specific implementations therefore produce inconsistent supervision and make results difficult to reproduce and compare. SolarWM addresses this coupling with a reconfigurable multi-source data engine and a backbone-native adaptation framework. The engine converts 1.43 million canonical clips from 10 datasets into a unified, frame-aligned contract covering visual observations, metric camera geometry, captions, quality metadata, selection decisions, and provenance, while decoupling source processing from mixture construction. Under shared camera-conditioning, training, and inference interfaces, we instantiate four 5B--33B models based on Wan2.2, LTX-2.5, and MiniMax-H3 while preserving their native representations and objectives. A unified three-stage recipe combines bidirectional adaptation, teacher-forced autoregressive initialization, and distribution matching distillation. The resulting causal models enable real-time interaction over rollouts ranging from minutes to hours after being trained on only 5s sequences. By releasing the resulting data, pipeline, recipes, weights, and framework, SolarWM provides a reproducible and extensible foundation for interactive world-model research.

Evaluating LLM agents is essential for guiding their development, yet it has grown prohibitively expensive: a single pass of a frontier model over an agentic benchmark can cost hundreds to thousands of dollars, a price paid repeatedly across iterative development cycles. Prior efforts, centered on benchmark distillation, reduce the number of evaluation tasks but leave the cost of executing each retained task untouched. In this work, we introduce early outcome prediction, a complementary axis of efficiency that instead cuts cost within each task. Our key insight is that an agent's final outcome is often evident from its intermediate behavior well before execution completes. We instantiate this idea in EarlyEval, a lightweight framework that trains a pair of LightGBM success and failure classifiers over behavioral, textual, and reference-solution features, and halts an agent run the moment either classifier crosses a calibrated confidence threshold, adding negligible per-step overhead. Across three benchmarks, SWE-bench Verified, TerminalBench, and Toolathlon, EarlyEval can eliminate 13%-26% of agent steps and up to 44.1% input tokens and 29.4% output tokens at 89%-97% prediction accuracy, while perturbing per-agent resolve rates by only one to two percentage points on average.

Retrieval is the first stage of modern search and advertising systems, selecting a candidate set from a large item universe for downstream ranking and auction. Recent work increasingly leverages LLMs to improve retrieval through query expansion, data synthesis, and retrieval-feedback training. However, the generative component is typically used for query-side augmentation, while final matching is still delegated to a downstream retriever. We introduce CoGR, a retrieval framework that instead trains LLMs to directly construct retrieval representations on both query and item sides. Each generator produces a compact set of keywords, which are matched directly through an inverted index, preserving compatibility with existing keyword-based retrieval infrastructure. CoGR uses a two-stage training pipeline. Supervised fine-tuning first establishes an aligned keyword space, after which co-evolving reinforcement learning alternately optimizes the query- and item-side generators with GRPO against the opposite side's frozen index. Both sides optimize the same query-to-item retrieval F_1 objective: the query side receives retrieval F_1 directly, while the item side receives a counterfactual marginal reward measuring the change in query-side F_1 caused by its generated keywords. Across 10 representative sparse, dense, and generative baselines, CoGR achieves the best performance on both an internal APP Marketplace dataset and the public WANDS benchmark, improving F_1 over the strongest baseline by 10.9% and 36.1%, respectively. Further analysis shows stable co-evolution and increasingly aligned query--item keyword spaces over training.

Language models spend most of their attention on a small fraction of context, yet they read the entire KV cache to find the few tokens that matter. If the user asks about a previous detail in a 1M-token conversation, global attention layers must scan the full context to generate each token of the reply. A prominent approach mitigates this cost by pre-selecting relevant tokens via lightweight proxy scores, but this extrinsic scoring still incurs O(N) per step. We take an intrinsic approach motivated by the simple question: wouldn't the model already know which parts of the context are relevant? To this end, we introduce Declarative Attention (DA), a protocol that elicits the model to declare where it needs to attend within its chain-of-thought, partitioning generation into three modes: <global> (full context), <focus> (a specific region), and <local> (recent output only). The inference engine parses these declarations like tool calls and skips most of the KV cache read. Under zero-shot evaluation across 15 long-context tasks, DA on off-the-shelf models (Gemma-4-31B, Qwen-3.6-27B) significantly reduces total attended tokens during decoding (52.0%, 31.1%) with modest accuracy drops (1.27pp, 2.75pp) that shrink with model scale. DA unlocks a new axis of sparse attention, with further potential under training-based methods that future work can explore.

Text-rich visual inputs require models that can read, retrieve, and compress language directly in pixel space, yet existing pixel-text encoders struggle with fixed resolution pretraining, visual shortcut learning, weak visual grounding, and multilingual visual text understanding. In this work, we investigate the fundamental design principles required for robust visual text representation learning. Through systematic controlled ablations, we identify four critical components: variable image resolutions and rendered font sizes provide spatial proxies for high-resolution document generalization; natural image-text pairs are indispensable for grounding and prevent text-only collapse; layout-aware rendering helps prevent pixel-level shortcuts; and a two-stage multilingual curriculum enables effective cross-lingual alignment. By integrating these principles into a scalable training recipe, we train Pixel Linguist II, a native-resolution vision encoder trained with on-the-fly rendering, unified contrastive grounding, and a multilingual curriculum over 280M training examples. Pixel Linguist II sets new state-of-the-art results on English, cross-lingual, and multilingual Visual STS and ViDoRe, while also enabling better MLLM downstream evaluation. Notably, Pixel Linguist II remains robust under 80\% visual token compression, showing great promise for optical context compression. Our code and resources are available at https://github.com/Pixel-Linguist/Pixel-Linguist-II.

Knowledge-Based Visual Question Answering (KB-VQA) relies on retrieving external information to answer queries involving long-tail entities. However, existing retrieval pipelines predominantly employ CLIP-style dual encoders, which prioritize surface-level visual similarity over entity-level semantic alignment. This paradigm often fails when semantically identical concepts exhibit large visual variations or when distinct entities appear visually similar. To address this, we propose KBMR, the first MLLM-based embedding retriever tailored for KB-VQA. Leveraging the robust autoregressive capabilities of MLLMs, KBMR maps images into a semantic space that better preserves concept identity. To tackle the challenge of noisy supervision in Wikipedia-scale retrieval, we introduce an MLLM-based semantic discriminator that generates continuous entity-consistency weights. These weights guide a novel continuous semantic distillation objective, enabling effective hard negative sampling and soft supervision beyond rigid binary labels. Extensive experiments demonstrate that KBMR significantly outperforms CLIP baselines, yielding up to a 14.7% improvement in retrieval Recall@1 and a 9.4% gain in end-to-end VQA accuracy. Code is available at https://github.com/realHarryX/KBMR.

Large language models (LLMs) increasingly interact with external environments and accumulate substantial behavioral experience, yet existing agent benchmarks largely evaluate them as fixed policies. It therefore remains unclear whether an agent can actively test its behavior, judge the resulting experience, and use that experience to improve future decisions. We introduce S\textsuperscript{3Gym}, an interactive benchmark for evaluating LLM self-improvement through three coupled capabilities: Self-Testing, Self-Judging, and Self-Improvement. S^3Gym separates permissive exploration from strict held-out evaluation and instantiates this protocol in seven text-based games with executable environment verifiers. We evaluate three pathways for incorporating interaction experience: direct History ICL, score-conditioned Summary Memory, and parameter Training. Our experiments reveal that self-improvement is neither automatic nor uniform. Context-level experience improves performance for several model--game pairs, but the most effective pathway depends strongly on the task structure: summaries are beneficial when experience can be compressed into reusable strategic rules, yet often underperform raw history when success depends on precise, state-contingent information. Parameter training produces substantial gains on some tasks, but also exhibits unstable improvement and severe negative transfer on others. These findings show that recognizing successful actions is insufficient; agents must also transform feedback into executable and transferable policies. S^3Gym provides a unified framework for diagnosing this process and identifying the bottlenecks that prevent agents from translating interaction experience into reliable self-improvement.

Multimodal models often build on architectures designed for generative vision-language modeling, typically combining separately pretrained vision encoders with causal language models. Visual document retrievers such as ColPali repurpose these models as encoders, carrying over the parameter and compute overhead of a VLM for a non-generative task. We introduce NeoMME, a family of 260M and 800M-parameter Multimodal and Multilingual bidirectional Encoders that process multilingual text and raw image patches in a single bidirectional Transformer encoder. Both models are pretrained from scratch with a masked discrete-diffusion text objective, conditioned on visible image patches for multimodal examples. Both support a 16,384-token context, enough to encode up to two standard 4K UHD images. To demonstrate its downstream capabilities, we fine-tune NeoMME with jointly trained dense and late-interaction heads. On the ViDoRe v3 benchmark, the resulting NeoMME-Retriever 260M outperforms all evaluated models strictly below 800M parameters with 0.523 nDCG@10, while NeoMME-Retriever 800M reaches 0.556. At a matched 2048x2048 image input size on an NVIDIA L40S, NeoMME-260M encodes pages with about 2x the throughput of ColModernVBERT. Hierarchical token pooling and asymmetric quantization compress late-interaction multimodal document embeddings by 255x while preserving over 95% of baseline nDCG@10. We contribute NeoMME to Hugging Face Transformers and release the pretrained backbone and retrieval-compatible checkpoints under Apache 2.0 at https://hf.co/collections/Hcompany/neomme.

Compact token sequences are essential for efficient 3D generation. However, existing 3D tokenizers typically organize latent representations either over spatial regions or as fixed-size sets of global tokens, both suffering sharp reconstruction degradation when compressed to extremely low token budgets. In this paper, we present ZipTok3D, a 3D tokenizer designed for high-fidelity reconstruction from extremely short token sequences. Its key idea is to organize object geometry into progressively informative global-token prefixes and unfold these compact representations through iterative decoding. Specifically, nested dropout randomly truncates the latent sequence after encoding during training and requires each retained prefix to reconstruct the complete object, thereby prioritizing essential geometric information in the leading tokens. The decoder then repeatedly applies a parameter-shared Transformer block to recover fine-grained geometry from each prefix without a separate generative sampling stage. With the same token dimension, ZipTok3D achieves reconstruction quality comparable to the 32-token COD-VAE baseline using only one token on ShapeNet and four on TRELLIS, yielding 32times and 8times shorter token sequences, respectively.

An image may be worth a thousand words, but most captioning models describe it in only a few. Modern vision-language models produce fluent high-level captions, yet routinely miss the attributes, counts, textures, materials, and spatial relations that make an image visually specific. Recent multi-stage systems recover some of these details through generation, decomposition, verification, and rewriting, but they do so at the expense of substantially higher inference latency. We propose SimLoss, a reference-free embedding-space objective for single-pass fine-grained image captioning. SimLoss trains a vision-language model to align its projected hidden-state representation with a frozen image embedding through an InfoNCE contrastive loss, supplying a dense visual supervision signal before any text is decoded, and requiring neither human-written fine-grained captions nor pseudo-captions from a multi-stage pipeline. We instantiate it as SimLoss FFT, which backpropagates through a locally available embedding model, and SimLoss GRPO, which treats that model as a black-box reward. Compared with single-pass, multi-stage verification, reward-optimized, and perception-aware baselines, the fully differentiable fine-tuning variant, SimLoss FFT, achieves the highest precision while nearly matching the F1 score of the multi-stage method, all while retaining single-pass inference and running roughly 20 times faster than the multi-stage pipeline. The reward-based variant SimLoss GRPO attains the strongest recall. Together, these results show that embedding-space supervision can recover the quality of multi-stage verification at the latency of a single-pass captioner.

Reinforcement learning with verifiable rewards (RLVR) has emerged as a powerful paradigm for large language model (LLM) post-training, but its reliance on coarse outcome rewards leads to limited guidance on intermediate reasoning processes. Existing approaches such as process reward modeling and on-policy distillation introduce additional constraints, such as reliance on a specialized reward model or assuming identical reasoning patterns between teacher and student. Nevertheless, we observe that once a reasoning process first goes wrong, evaluating the subsequent reasoning provides limited additional information, as it is already conditioned on an invalid prefix. Therefore, we propose Cliff, a reward shaping strategy that utilizes an off-the-shelf LLM as a teacher to identify the first mistake in each rollout. As a result, the rollout is naturally decomposed into two parts: a correct prefix and an incorrect suffix. Cliff then converts this signal into token-level advantages, assigning positive advantages for the correct prefix and negative feedback afterward. Experiments across 12 different scenarios demonstrate that Cliff consistently improves reasoning performance, outperforming on-policy distillation by 15% and standard GRPO by 7%, even with teachers of modest capability. Furthermore, we analyse the role of ``ground truth'' in Cliff and investigate its training dynamics. These results establish Cliff as a simple, general and effective approach for improving RLVR with richer, fine-grained supervision.

Sampled-token on-policy distillation (OPD) efficiently transfers capabilities from teacher to student using student-generated tokens, requiring teacher probabilities only for sampled tokens. Yet it frequently suffers from diversity distillation failure: the student's pass@1 improves while its pass@k plateaus, failing to inherit the teacher's diversity. To explain this, we introduce First-Order Local Entropy Influence, a signed first-order proxy that decouples each update's entropy effect into the teacher--student log-probability gap and the student's local probability structure, and empirically links entropy contraction to negative-influence positions. Motivated by this, we propose Influence-Directed Adaptive On-Policy Distillation (IDA-OPD): rather than relying on costly full-vocabulary Forward-KL objectives, it preserves entropy-expanding updates while replacing entropy-contracting ones with divergence-adaptive advantage shrinkage, using only the teacher's sampled-token log-probability. Experiments on reasoning-oriented distillation show IDA-OPD consistently improves pass@k, inheriting the teacher's diversity through distillation, matches the strongest teacher-informed methods at strictly lower cost, and broadly maintains vanilla OPD's pass@1, all without full-vocabulary teacher information.

Axios : Sources: PayPal has paused plans to sell its VC portfolio for $900M-$1B after bids came in lower than hoped, months after deciding to wind down PayPal Ventures —  PayPal has paused plans to sell its venture capital portfolio for around $900 million to $1 billion, Axios Pro has learned from multiple sources.

Subrat Patnaik / Bloomberg : Oura files for a US IPO, reporting a $924.3M net loss on $1.21B in revenue for the nine months ended June 30, vs. a $182.8M net loss on $697.6M a year earlier —  Health and fitness ring-maker Oura Inc. filed for an initial public offering, showing soaring revenue as well as increasing losses.

Annie Palmer / CNBC : Adobe names Customer Experience Orchestration president Anil Chakravarthy as CEO, effective December 1, replacing Shantanu Narayen, who becomes executive chair —  Adobe on Thursday named Anil Chakravarthy as its next CEO, succeeding Shantanu Narayen, who announced he would step down earlier this year.

Sam Sabin / Axios : OpenAI commits $1B in subsidized model access, training, support, and partnerships to a new initiative aimed at protecting essential services around the world —  - The $1 billion will come in the form of expanded subsidized access to models, training, technical support and partnerships for participating organizations.

Bloomberg : Sources: Crusoe signed a ~$13B, five-year deal with Jane Street to provide GPUs and other infrastructure for AI training and inference via its cloud platform —  Crusoe, a data center upstart with contracts to supply AI computing power to the likes of Meta Platforms Inc. and Oracle Corp. …

Greg Kamradt / ARC Prize : GPT-6 Astra scores 62.7% on ARC-AGI-3 with the standard harness and 99.9% with a new provider adapter harness; Claude Opus 5 scored 30.2%, and GPT-5.6 Sol 7.8% —  Summary  — GPT-6 Astra scores 62.7% for $26K on ARC-AGI-3 Semi-Private with our Standard harness, and 99.9% for $19K with a Provider Adapter harness.

Maggie Eastland / Bloomberg : DOD's Emil Michael says Anthropic is still a designated supply chain risk, a day after Howard Lutnick said it resolved its issues with the Trump administration —  US Commerce Secretary Howard Lutnick said Anthropic PBC has resolved its long-running issues with the Trump administration but a top Pentagon official begs to differ.

Politico : A White House official says Zuckerberg opposed a US AI regulator in a call with Trump last month; another source says he didn't ask Trump to change his stance —  Meta CEO Mark Zuckerberg expressed concerns about plans for a national AI regulator in a previously unreported phone call …

Michael Nuñez / VentureBeat : Microsoft AI debuts MAI-Transcribe-2, a speech recognition model that it says beats Gemini 3.5 Transcribe and GPT-Transcribe, at $0.10/audio hour through 2026 —  Then it priced the thing at 10 cents per hour of audio.  —  That figure deserves a pause. … Thursday's early-bird price cuts that by roughly 72%.

Ina Fried / Axios : OpenAI says Astra was built on its largest-ever training run, using more than 100,000 GPUs at its Stargate site in Texas —  OpenAI on Thursday released GPT-6 Astra, which president Greg Brockman called a “generational leap” and said could eventually be seen as the arrival of artificial general intelligence, or AGI.

Maxwell Zeff / Wired : OpenAI calls GPT-6 Astra the “world's best computer use model”; in tests, it booked DMV appointments and searched job listings faster than the average person —  OpenAI leaders think the company's next generation model, which excels at computer use and coding, may mark a major milestone in AI development.

The Deep View : OpenAI prices GPT-6 Astra at $10/1M input tokens and $50/1M output tokens, matching Anthropic's pricing for Claude Fable 5.1 —  OpenAI's now infamous model that was part of the Hugging Face breach is about to be loose in the world, but with new safeguards.

Hayden Field / The Verge : OpenAI launches GPT-6 Astra, initially for customers in its Daybreak program; Greg Brockman says it is a “generational leap” and “we are now in the AGI era” —  OpenAI's next big model is here: GPT-6 Astra.  The company calls it a “generational leap in capability” …

The Information : Sources: Mira Murati's Thinking Machines Lab is in talks to raise $1B+ at a $40B+ pre-money valuation, lower than the $50B+ valuation it sought last year —  Thinking Machines Lab, the AI developer led by ex-OpenAI chief technology officer Mira Murati, is in talks to raise at least $1 billion …

Bloomberg : Humain launches humain-m3, an Arabic-language model developed with MiniMax, amid controversy among US allies over sovereign AI built with Chinese models —  Saudi Arabia's artificial intelligence company Humain has built a national AI platform based on an open model from Chinese lab MiniMax Group Inc. …

根据发表在《Cerebral Cortex》期刊上的一项研究，科学家招募了 18-74 岁的成年人，最后的参与者包括 17 名青年人（18-30 岁）、21 名中年人（50-60 岁）和 23 名老年人（61-74 岁）。每一位参与者都躺在 MRI 扫描仪中完成了一项多步骤的记忆任务。参与者首先接受静息扫描，闭上眼睛保持静止不动；接着在学习阶段观看一张脸与一物体或一场景配对出现，被要求想象此人互动的情景以帮助记住配对信息；最后进行记忆测试。结果显示年龄是记忆表现的重要预测因素。年轻参与者正确识别配对的频率远高于中年或老年参与者，而中年和老年人的表现则相近。老年人表现出更多的混淆错误。老年人的大脑可能并非只是记忆力较弱，他们的大脑会记住过多错误的信息，记忆被扭曲混淆了。

纽约市长 Zohran Mamdani 周三宣布了一项为期一年的新政策，禁止从幼儿园到八年级的公立学校学生使用生成式 AI。纽约市同时还限制低年级学生的屏幕使用时间，为高中生开设 AI 素养课程。禁令将于 2026-2027 学年生效，将影响近 60 万名学生，占全市公立学校学生总数的 2/3。Mamdani 表示，“这项禁令体现了我们对未来负责的承诺，我们将拥抱新技术，但前提是它必须服务于我们的学生。”新政策引入了新的屏幕时间指导方针，包括限制低年级学生一对一使用电子设备，建议 3-5 年级学生每日使用时间上限为 30 分钟，6-8 年级学生每日使用时间上限为 45 分钟。

在任天堂之后，另外两大游戏机制造商微软和索尼都表示他们无法律义务将美国关税退款退给消费者。去年美国在全球大规模征收关税，任天堂等公司据此提高了相关产品的售价。今年 2 月美国最高法院裁决征收关税违法，任天堂等公司随后起诉美国政府要求退还关税。玩家们随后起诉了任天堂、微软和索尼等公司，想要获得部分退还的关税。任天堂表示它无法律义务将美国关税退款退给消费者。索尼和微软都持相同的立场。索尼表示它预计将会获得 5.08 亿美元的关税退款，大部分退款将交给游戏部门。索尼律师本周在一份请求法庭驳回诉讼的动议中表示：“为自愿购买的消费品支付公平的市场价格，不构成法律上可认定的实际损害。”微软律师 8 月 21 日在类似案件中提出了相似的观点。

音频编辑器项目 Audacity 释出了 v4.0。主要变化包括：基于 Qt6 重构了 UI，支持原生高 DPI 渲染；能将 UI 布局保存为 Workspaces；工具栏和面板可移动、停靠、浮动、显示或隐藏；改进了音频片段的处理；新增 .aup4 项目格式；等等。Audacity 4.0 与 Audacity 3.x 不完全兼容，部分 Audacity 3 系列的功能尚未移植到 4.x 版本。

尼泊尔和西藏的泥石流灾害发生迄今过去 8 天，累计死亡人数已超过 1200 人，还有近 5000 人仍然下落不明。尼泊尔外交部长卡纳尔（Shisir Khanal）上周接受该国媒体 Kantipur TV 访问，称这场灾难不只是一般的洪水，其规模之大犹如海啸。他认为，冰川快速融化是全球暖化与气候变迁的后果，但气候变迁的始作俑者并不是尼泊尔。“我们在为这场不是我们制造出来的全球危机，付出沉重代价。”卡纳尔指出，世界上最大的碳排放国是中国，美国是第二名，第三是印度，“这些国家负有历史责任，应该补偿像尼泊尔这样的脆弱国家。”尼泊尔的气候倡议者拉兹姆（Tashi Lhazom）认为，气候变迁的背景下，尼泊尔属于最脆弱的国家，却也是最重视气候和环境议题的国家之一。尼泊尔官员表示已向联合国申请“损失与损害”资金（loss and damage）的紧急援助。国际气候倡议组织也纷纷呼吁工业大国应更积极协助。

LibreOffice 26.8 于 8 月 26 日发布，根据官网的统计数据，在发布一周内其下载量达到了 1,031,162 次。负责 LibreOffice 项目的基金会 The Document Foundation 表示这是至今最高的首周下载次数数据。大多数 Linux 用户是通过 Snap Store、Flathub 等软件商店渠道进行更新，因此实际数字会更高。为什么这一版本会如此大获成功，也许是在 AI 时代它可能是少数没有引入任何 AI 功能的常用办公软件。微软的 MS Office、苹果的 iWork 甚至开源的 ONLYOFFICE 都在不断集成 AI 功能。

全球最大暗物质探测器 LUX-ZEPLIN（LZ）实验团队于 9 月 1 日在日本山形县天童市举办的 TeV 粒子天体物理大会上宣布探测器记录到一次不同寻常的高能闪光，它可能是穿行于银河系的大质量暗物质粒子与原子核发生碰撞所产生的。如果这一观测结果能得到更多数据的支持，或许意味着人们终于发现了暗物质。这种不可见的物质被认为维系着星系，防止其分裂。它同时也将证实，暗物质由质量远大于质子的粒子构成，即弱相互作用大质量粒子（WIMP）。研究人员分析了 2023 年 3 月-2024 年 4 月共计 220 天的观测数据。他们重点研究高能粒子撞击氙核产生的罕见闪光，这种撞击会使后者高速反冲。这个“超额”事件是一次反冲信号，在探测器中沉积了 248 千电子伏特的能量。LZ 合作组估算，能造成这种反冲的暗物质粒子的质量至少相当于 200 吉电子伏特，可能在 1000 吉电子伏特左右（1 吉电子伏特约等于一个质子的质量）。 LZ 合作组已经有效排除了放射性本底的干扰。

土星北极地区自 1980 年代航海家号发现以来便稳定存在数十年的巨大六角形气流，始终是太阳系中最神秘的气象奇观之一。过去科学家一直不确定这种近乎完美的几何多边形是否仅是北极的独家专利，一个科学研究团队近期在土星的南极，首度发现了另一个更加庞大的十边形气体结构。它最早由业余天文爱好者在 2024 年拍摄的影像中捕捉到蛛丝马迹，随后研究团队利用哈勃望远镜的高解析度数据进行深入分析与极地投影，不仅证实了其清晰的十边形边界，更追踪出它其实早在 2023 年就已然形成。这个十边形以南纬 60 度为中心，深深植入土星大气层中，顺着一道时速高达 420 公里的强劲向东喷射气流，以相对悠闲的速度缓慢推移。它的出现一举打破了六角形的孤立神话，证实土星大气层的特殊条件确实具备在两半球极区孕育各种多边形波的独特潜力。尽管两者皆属于大气波现象，但这个新诞生的十边形与北极六角形却非简单的镜像对称，十边形所处的纬度较低，且似乎不如稳定存在逾44年的六角形那般顽强。研究人员推测，十边形北侧一颗宽达 4,000 公里的高压涡旋红斑可能在其中扮演了强行驱动气流震荡的角色，但目前的模拟仍无法完全重现其完美的十边形结构。

最大 PC 游戏 Mod 分享托管平台 Nexus Mods 收购了 SteamDB。SteamDB 创始人退出的理由与 2025 年 6 月 Nexus Mods 创始人 Robin Scott 将网站出售给游戏公司 Chosen 的理由类似——筋疲力尽。SteamDB 的两位创始人 Pavel 和 Marlamin 于 2013 年创办了记录 Steam 游戏商店所有上架游戏相关信息的网站，和 Nexus Mods 一样，一开始是一个爱好性质的项目，但之后逐渐成为最大的 Steam 游戏信息平台，Marlamin 后来退出，新冠疫情之后 AI 时代来临，互联网已经不是以前的互联网，Pavel 表示筋疲力尽了，因此过去几个月一直在寻找人来接手，最后选择了 Nexus Mods 的母公司 Chosen，Chosen 承诺无意改变 SteamDB 的任何功能，也不会将其设为付费内容，计划将部分功能整合到 Nexus Mods 中。

NASA 授予 Blue Origin 一份价值高达 7 亿美元的合同，用于建造和运营一个专门的火星通信网络，以支持当前和未来的火星探测任务。Blue Origin 预计于 2028 年底交付火星飞行器，预计 2030 年投入运行。Blue Origin 将负责设计、开发、集成、发射和运营该网络，作为 NASA 更广泛的太空通信和导航基础设施的一部分。该架构将包括一艘环绕火星运行的高性能通信飞行器，用于传输科学数据、图像、导航信息以及为在火星表面及周围运行的飞行器提供的关键任务通信。

研究人员开发出名为 PathStAR 的深度学习模型，将 40 种人体组织的 2.5 万余张活检图像分割成 3000 多万个小图像区域，从中提取反映组织结构的特征，以分析不同年龄阶段组织结构的变化。卵巢组织的分析显示，卵巢呈现出明显的“双相结构衰老”特征：35-40 岁期间衰老速度明显加快，这一时期恰逢女性生育能力下降；55-60 岁期间再次出现衰老加速，与绝经期相对应。研究人员将 PathStAR 应用于 15 种样本量较大且覆盖全年龄段的组织，发现不同组织的结构衰老主要呈现三种模式。第一种是早期衰老型，血管系统属于这一类，其结构衰老在30-39 岁期间明显加快，之后衰老速度下降。第二种是晚期衰老型，子宫就属于这一类。它在成年早期相对稳定，直到 50-55 岁才进入衰老加速期。第三种也是最常见的模式是双相衰老型，即组织在一生中出现两个明显的衰老加速阶段，共有 9 种组织呈现这一特征。进一步分析发现，各组织在结构衰老加速期具有相似的分子特征：与炎症相关的基因表达增加，而参与能量产生、细胞生长和细胞质量控制的基因活性下降。

欧洲粒子物理研究所（CERN）计划年底前将其加速器控制基础设施中的逾 2200 台工业计算机和嵌入式系统全部迁移到 Debian 13。CERN 的计算机长期使用的是 RHEL/CentOS，此前还维护过基于 RHEL 的企业级发行版 Scientific Linux，2016 年迁移到 CentOS，在 CentOS 停止维护转为滚动更新发行版 CentOS Stream 之后考虑过该发行版。CERN 表示，Red Hat 默认采用“-march=x86-64-v2”编译器标志是压垮骆驼的最后一根稻草，因为这意味着强制淘汰旧硬件。

Uber 宣布裁员 3300 人，大约占全球员工总数的 10%。与其它科技公司不同的是，Uber 不是以 AI 的借口裁员的，而是旨在精简管理层和降低成本。这是 Uber 2020 年 5 月以来最大规模的一轮裁员。CEO Dara Khosrowshahi 在一封内部邮件中宣布了裁员决定。该公司计划裁掉 20% 的管理人员，将向 CEO 汇报层级达到或超过七层的员工人数减少 20%，将只有一到两名成员的团队数量减少 50%，合并工程、科学和配送部门，整合餐饮、零售和直销部门的配送业务，取消远程办公岗位——Uber 目前仅允许不到 1% 的员工远程办公。

全世界正在快速发展太阳能这一廉价电力来源，但各地发展并不均衡。以欧洲为例，除了罗马尼亚，其它欧洲国家几乎没有增长，原因是欧洲大陆电网的局限性。发电高峰期间电网拥堵，储能不足，导致能源浪费、负电价，甚至出现电网故障和灾难性停电风险，欧洲各国正竞相增加电网储能容量以解决这些问题。罗马尼亚太阳能装机量在 2025 年增长了 45%，该国在太阳能建设早期就对电池储能进行了投资。除此之外，巴基斯坦的屋顶太阳能和电池安装速度超过了其他任何市场，巴西、智利、萨尔瓦多、摩洛哥、肯尼亚和纳米比亚在清洁能源转型方面都已超越了美国。

根据发表在《Frontiers in Public Health》期刊上的一项研究，全世界每年有近五成农民因农药中毒。研究发现每年约有 4.02-4.33 亿农民和农场工人因意外急性农药中毒。全球农业人口约为 9.34 亿，这意味着每年约有 46% 的农民农药中毒。中毒人数最多的地区是南亚，其次是东南亚和东非。中毒率最高的国家是西非布基纳法索，该国近 84 %的农民和农场工人曾遭受农药中毒。研究还发现，每年约有 11,000 人死于农药中毒，其中近 60% 发生在印度。研究指出，农药的使用量仍在持续增长，2023 年的使用量达到 380 万吨，比 1990 年翻了一番。