OrangeBot.AI Digest — 2026-09-29

Let your AI agents remember what you've seen

We Run The Numbers. You Run The Show.

Meeting notetaker that installs and runs inside Claude Code

Independent auditing of AI agents to uncover misalignment

Find out who'd pay before you build more

500+ AI models available via a single API

Intelligent clipboard that knows what you're about to paste

Anthropic's second model in the Claude 5.5 family

AI agents purpose-built for managers

Create, optimize & report on LinkedIn Ads from any AI tool

AI that turn your ideas into real investments

Private, self-hosted smart home, no YAML required

Put Codex or Claude Code on your own cloud machine

Floating Markdown notes for Mac. Pay once, own your files

Putting your most recurring tools in the right-click menu

We find that language models can transfer capabilities through task-unrelated text. Post-training typically improves language models using task-specific data. Prior work on subliminal learning shows that information about these updates can pass through unrelated generations, but has largely focused on traits or preferences using extensive teacher outputs. We introduce Active Taskless Distillation (ATD), which achieves capability transfer using only a single word from the teacher per prompt. ATD probes the behavioral shadow of post-training by selecting prompts where the teacher and student's shared public ancestor is nearly indifferent between two ordinary words. A student initialized from this ancestor learns solely from the resulting prompt-word pairs, without target-task examples, teacher logits, or teacher parameters. In the primary coding experiment with Qwen2.5-1.5B, 5,664nses yield a 5.34 pp gain on HumanEval+ over an exact nuisance-matched control thadisrupts prompt-resperiments showtransfer in scientific knowledge, commonsense reasoning, and reading comprehensins across additional model generations, sizes, and families. Functional analyses show that the learned sid composable, andthat its strength tracks the teacher's update strength.

Reinforcement learning can turn one language model into several specialists, each excellent at a single skill such as mathematics, coding or following instructions, but users need one model with all of these skills. Multi-teacher on-policy distillation (MOPD) merges them by letting the specialists teach one student: the student answers each prompt, and the specialist for that prompt's domain gives feedback on every token. This routing decides which specialist teaches, but not how strongly its feedback moves the shared student. In Qwen3.5 models at three sizes, we find that MOPD's student does not beat one taught by the best single specialist and gains little of the mathematics specialist's advantage. The feedback is unbalanced: instruction-following feedback is several times more spread out than mathematics feedback and dominates the student's updates. We propose Domain-Normalized MOPD (DN-MOPD), which keeps the routing and rescales each domain's feedback by its measured spread. On six public benchmarks, DN-MOPD improves the average score over MOPD at every size, across three random seeds and under two answer-length limits, and recovers most of the lost mathematics gain. Controls with fixed domain weights show that the gain comes mainly from turning down instruction-following feedback rather than turning up mathematics alone, and that fixed weights close to those DN-MOPD measures perform comparably. Combining specialists therefore requires deciding not only which one teaches, but also how strongly its feedback counts.

Visual backbones have evolved from Convolutional Neural Networks (CNNs) with local aggregation to Vision Transformers (ViTs) with global interactions, State-Space Models (SSMs) with input-dependent state transitions, and Test-Time Training (TTT) layers that adapt an inner learner while processing an image. Across this progression, visual computation has become increasingly adaptive to each input, yet the rules governing that adaptation remain largely prescribed by the trained backbone. We introduce VisionHOPE, the first generic visual backbone formulated as a self-modifying learning system, in which what the model remembers and how it learns co-evolve within an image. Building on the self-referential construction of Nested Learning (NL), VisionHOPE realizes this co-evolution through five coupled memories that store content, generate key and value representations, and govern learning rate and retention. These memories evolve jointly as visual context accumulates along each scan. However, directly applying the unconstrained self-referential update to a visual backbone leads to instability. We therefore derive a stability-matched step-size control scheme that combines a soft cap on self-referential injection with a spectral clamp on the retained memory transition, and prove that the resulting memory dynamics are non-expansive along each scan. For two-dimensional feature maps, we adapt NL's chunk formulation by aligning chunks with image rows and columns across four directional scans. The proposed VisionHOPE achieves competitive results on ImageNet-1K, COCO, and ADE20K, establishing self-modifying learning systems as a practical foundation for general-purpose visual backbones. The code is available at https://github.com/PSRben/VisionHOPE.

Reinforcement learning (RL) for code agents often uses executable tests to provide binary rewards. With these rewards, Group Relative Policy Optimization (GRPO) assigns identical advantages to test-passing trajectories within each rollout group, overlooking differences in implementation quality and adherence to task requirements. This leaves the policy without a learning signal that favors clean, targeted implementations over those containing unnecessary or out-of-scope changes. We introduce GAGAR, a framework for quality-aware credit redistribution in code agent RL. Built on dynamic sampling that retains groups containing both passing and failing trajectories, GAGAR places all trajectories from each group in a shared workspace, where an SFT-trained agentic grader jointly inspects them and ranks the test-passing candidates. Based on this ranking, we downweight lower-ranked trajectories and proportionally rescale the advantages of all test-passing trajectories to restore their original sum. This sum-preserving redistribution retains the relative weights established by quality-based downweighting while shifting credit toward higher-quality implementations. We evaluate GAGAR at industrial scale using pre-RL SFT checkpoints of MiMo-V2.6-Flash (310B total parameters) and MiMo-V2.6-Pro (1.02T total parameters). Controlled code-only Flash experiments show improved code agent performance, reduced trajectory-length growth, and more stable training. We further apply GAGAR in large-scale mixed-task RL with both Flash and Pro. Our results support combining test-based verification with groupwise agentic grading to improve the quality and stability of code agent RL.

Most modern multimodal large language models (MLLMs) build on a pretrained visual encoder that provides a strong visual prior. Encoder-free MLLMs instead learn visual representations directly from raw pixels, offering a simple and unified architecture, but their scaling behavior has not been systematically characterized. To fill this gap, we compare scaling laws for encoder-free and encoder-based MLLMs and report three main findings: (1) Removing the visual encoder shifts the compute-optimal allocation for the multimodal objective toward larger models, while leaving that for text nearly unchanged. (2) The two architectures exhibit nearly overlapping loss--compute frontiers on the text objective, but diverge on the multimodal objective: encoder-free models underperform at small scales yet are predicted to catch up at around 10^{22} FLOPs, well within practical pretraining budgets. (3) Without a visual encoder, the language model learns to take over its role via vision-specific adaptation: bidirectional interactions among visual tokens become increasingly beneficial as training compute grows, visual processing shifts toward earlier layers, and expert routing for visual tokens becomes more concentrated. Overall, our results indicate that the advantage of the visual prior provided by a pretrained encoder diminishes with scale, positioning encoder-free architectures as a promising direction for multimodal pretraining.

Unified multimodal models can both look at and render images, so in principle they can repair their own generations: diagnose what an image gets wrong, revise it, observe the result, and diagnose again. Whether a revision helps is known only after it is rendered, so the reflection text and the image generation must be learned jointly, over the whole loop. Supervised fine-tuning (SFT) on reflection trajectories gives a cold start but does not find the high-success repair paths, and naive RL that optimizes only the renderer or only one head leaves most of the gain untapped. We introduce UMM-Reflection, which applies reinforcement learning (RL) to complete reflection trajectories inside one unified model: sibling trajectories share one initial image, so the group-relative advantage compares reflection strategies, and one trajectory-level advantage updates both the reflection tokens and the flow-based revisions, avoiding the combinatorial blow-up of per-round credit assignment. Unlike single-round editing or pipelines with an external critic, credit flows across rounds and to both roles of the same model, and no verifier is needed at inference. On BAGEL, UMM-Reflection improves GenEval by 12.05 points over SFT, and the gains transfer to WISE (+10.97), OneIG-Bench (+3.48), and T2I-CompBench++ (+4.63), none of which is used in training.

A central goal in robotics is to enable manipulation across changing tasks and environments. Vision-language-action (VLA) models provide broad manipulation capabilities but can struggle when execution requires diagnosing failures and adapting behavior. Strong agents can discover effective interventions through interaction with these policies. We propose Recursive Harness Distillation to accumulate this experience as reusable guidance across agents. A strong agent distills its experience into a playbook for a light agent, then recursively refines the playbook using the light agent's execution feedback. The resulting playbook enables agents to reuse accumulated intervention knowledge in new task instances without updating model parameters. In real-world manipulation, the harness improves success from 37.3% to 64.0%. On SimplerEnv Bridge, the light agent with the playbook achieves 66.7% success, compared with 41.7% for the GR00T-only baseline, and outperforms the strong agent without a playbook. The same playbook also benefits the strong agent, which reaches 79.2% success. These results demonstrate the feasibility of harness distillation for robotics: intervention experience can be accumulated, refined through execution, and reused across agents to improve manipulation.

Vision-language (VL) pretraining using paired chest X-ray (CXR) images and radiology reports has shown strong potential for medical image understanding. However, existing methods often remain dependent on task-specific finetuning because radiology reports are lengthy, clinically dense, and difficult to align with simple zero-shot prompts. Recent sentence-level approaches partially address this limitation using clinical phrases extracted by large language models (LLMs), but they largely overlook the intrinsic characteristics of radiology discourse. In particular, limited positive-pair diversity constrains further gains, while clinically equivalent sentences frequently recur across patients, creating false negatives in contrastive learning. To address these issues, we propose SentZero, an enhanced sentence-centric VL pretraining framework for zero-shot, multi-task CXR analysis. SentZero introduces LLM-based abstract-level sentence structuring and mapping to expand positive-pair diversity, together with an additional loss term to mitigate false negatives. We further introduce sentence-conditioned residual modulation of visual embeddings, enabling visual features to adapt to the semantic characteristics of each input sentence. Across diverse downstream tasks and datasets, SentZero improves zero-shot generalization and outperforms prior multi-task zero-shot methods.

Large language model (LLM) agents increasingly undertake extreme-long (xlong) horizon tasks, where a single execution can span hours, hundreds of model--environment interactions, and nearly 1M tokens per rollout. Applying online reinforcement learning (RL) to such executions poses two fundamental challenges: (1) severe execution variance and prolonged rollout delays cause massive GPU idling; and (2) complex non-linear branching generates massive trajectory redundancy, crippling training efficiency. To address these, we presents QwenGyre, an end-to-end framework for xlong-horizon online RL. QwenGyre elastically reallocates GPUs between rollout and training without interrupting live executions, while its trajectory processor reconstructs branching histories, scores partial progress, and deduplicates redundant paths to bound training costs. Scaled to our flagship model, Qwen~3.8 2.4T, with 700K tokens per rollout, QwenGyre yields a 6.0% absolute gain on NL2RepoBench (52.5% to 58.5%) in 48 steps. Across our evaluations on diverse domains of training datasets, QwenGyre delivers up to 1.85times and 1.78times speedups over Colocate and Async, respectively.

Few-step autoregressive video diffusion generates a long video by splitting the video into temporal chunks and generating chunk-by-chunk, each through a short sequence of denoising stages. To memorize chunks that are already generated, previous methods reconstruct a clean or less-noisy key--value (KV) cache by additional forwards to build the cache without advancing an output latent. However, every denoising forward itself already computes the in-flight KV of the current chunk. We introduce FlashForward, which directly reuses this cache to avoid the heavy cache-update-only model forwards. After the current chunk completes one denoising stage, its stage-specific cache is already available for the next chunk. Assigning one GPU to each stage therefore lets different chunks occupy different stages concurrently. This early availability has a quality cost: the resulting stage-matched history is noisy, causing appearance and motion drift among chunks. To complement it, FlashForward produces sparse auxiliary clean anchor latents before the corresponding region is generated so the generation trajectories can be stabilized by this two-sided conditioning. The two memories operate at different temporal scales: sparse clean anchor KV supplies coarse, long-range two-sided structural guidance, while dense stage-matched history preserves fine, recent evolution. With up to four GPUs, FlashForward runs 1.16--1.69times faster than HiAR and 1.42--2.92times faster than Self-Forcing for 16 FPS videos of 20 seconds or longer across 1.3B and 14B backbone scales at 480p and 720p. On VBench, for the 1.3B model at 480p, it achieves higher scores and remains stable at longer durations, demonstrating that FlashForward generates high-quality and temporally consistent videos across durations of 20s, 35s and 65s at a much faster generation speed.

Fast matrix multiplication algorithms keep the product fixed and search for a cheaper way to evaluate it. We instead ask whether a Transformer's learned projections can use a different, cheaper product altogether. Building on an associative-algebra construction that replaces ordinary matrix multiplication with a sparser interaction table over the same weight blocks, we construct a family with quadratic arithmetic in the matrix dimension when the physical block size remains fixed, and derive finite-shape constraints for GPU execution. The construction is provably optimal for its bilinear rank by the Alder--Strassen bound and can be realized as row-typed rectangular projections compatible with causal masking and KV-cached decoding. We provide an empirical test of this approach by training two approximately 110M-parameter decoder-only Transformer LMs from the same recipe and 12.3B-token budget, differing only in their feed-forward layer: one uses ordinary dense matrix multiplication and the other uses the associative-algebra product. Across four prompt domains, the algebraic model achieves a 6.2--7.8\% increase in end-to-end generation throughput, while obtaining lower scores on all three reported downstream metrics. We treat these results as a feasibility and trainability check for the proposed approach at small scale, leaving further investigation to future work.

Coding-agent evaluation has progressed from resolving individual issues to carrying out long-horizon development, yet task completion is still largely assessed within a single codebase. In software ecosystems, many features and bug fixes require coordinated changes across multiple repositories. We introduce WideSWE to evaluate coding agents on such cross-repository tasks. Mining and reviewing changes across 103 software ecosystems yields 120 real-world tasks, balanced between 60 bug fixes and 60 features. We derive prompts from related issues and pull requests. We systematically review and adapt hidden tests to support diverse correct implementations while preserving required behavior and regression checks. Across seven agent configurations, full task success ranges from 10.83% to 42.50%, with the configuration pairing Codex CLI with GPT-5.6-sol achieving the highest rate. Trajectories show agents failing to identify necessary changes, recognizing changes but leaving them unfinished, or modifying the required repositories without fully satisfying the request. To examine whether working on one repository at a time can alleviate these difficulties, we compare it with joint execution under identical prompts. Independent execution mainly recovers omitted work and is less effective at correcting previously attempted but unsuccessful implementations. Joint execution can use information from related repositories to guide implementation and verification. Code is available at https://github.com/ZJU-ACES-ISE/WideSWE.

Tool-based image editing (image retouching) is commonly formulated with autoregressive multimodal large language models (MLLMs) that sequentially generate reasoning, tool selections, and parameter values. In this work, we present a novel approach to tool-based image editing by framing the task as a flow matching problem. We introduce FlowTool, a framework that directly models the distribution of high-quality tool parameters conditioned on the input image and user instruction using conditional rectified flow. FlowTool combines a vision-language model backbone for multimodal understanding with a Diffusion Transformer parameter generator that transforms Gaussian noise into an editing plan. We train FlowTool with a two-stage supervised flow-matching curriculum, followed by reward-based post-training. Across MMArt-Bench, FlowTool-Eval, ArtEdit-Bench, and MIT-Adobe5K, FlowTool achieves significantly stronger reference-based performance than specialized MLLM editing agents and proprietary MLLMs, while remaining competitive with proprietary models under reference-free evaluation. Moreover, FlowTool significantly improves inference efficiency, reducing latency by at least 50times while requiring nearly 2times less memory than the compared baselines. These results demonstrate that tool-based image editing can be effectively modeled as conditional generation over structured continuous editing parameters, without autoregressive reasoning.

Diffusion Transformers (DiTs) have established themselves as a scalable backbone for high-fidelity image synthesis. However, unlike U-Net based diffusion models that rely on rigid, hand-crafted skip connections, DiTs predominantly use a uniform residual stream that integrates all preceding layers as a monolithic state. In this work, we rethink residual connections in diffusion transformers and propose to transform them from passive summation into an active retrieval mechanism optimized for image denoising. First, we conduct a systematic analysis of DiT's internal representation, revealing a latent preference for early-layer feature reuse and symmetric layer guidance. Motivated by this, we introduce a structured connectivity design that explicitly integrates local residual connections with long-range pathways. Instead of static skip connections or dense all-layer routing, our method enables each transformer block to selectively ``attend'' to critical earlier representations, dynamically retrieving spatial and semantic cues through direct, differentiable cross-depth paths. Experiments show that our adaptive connectivity leads to faster convergence, with up to 1.73times fewer training iterations, and significant gains in FID and visual quality with less than 0.1% additional parameters, further improving a strong REPA-XL/2 model from 5.9 to 4.34 FID without guidance and reaching 1.39 FID with classifier-free guidance. Our findings suggest that adaptive cross-layer connectivity is a critical yet underexplored factor in diffusion transformers, and that incorporating structured information pathways provides a simple and effective direction for improving scalable generative models.

Frontier general-purpose systems are rapidly expanding beyond visual understanding into capabilities traditionally handled by dedicated computer-vision models. As these capabilities expand, a central question for the computer-vision community is how far this reach extends, and what remains hard. We evaluate GPT-6 Astra alongside five frontier general-purpose AI systems across 34 capabilities and 55 benchmarks spanning nine areas of computer vision. We compare their performance with dedicated models and humans where suitable references are available. Astra demonstrates broad visual capability, with substantial gains over other frontier systems in visual and spatial reasoning and several forms of structured prediction. Across the state-of-the-art systems, a consistent pattern emerges. Capabilities involving semantic interpretation, reasoning, and object-centric prediction increasingly approach or reach available reference levels. In contrast, larger gaps remain when tasks require metric geometric accuracy, faithful reconstruction, temporally consistent dense prediction, or specialized fine-grained visual knowledge. Additional reasoning and specialist tools close selected gaps, but their benefits vary across capabilities. These results map a changing landscape of computer vision in which increasingly sophisticated visual tasks are accessible through a general-purpose interface, while precise and fidelity-sensitive perception remains an important frontier.

Michael Shepard / Bloomberg : President Trump signs an executive order requiring federal government agencies to refer to AI as “Super Intelligence” or “SI” in all official communications —  President Donald Trump sees artificial intelligence as suffering from a branding problem …

Reuters : IPO filing: Anthropic routed 47% of its sales, or ~$2.16B, in 2025 through cloud partners Amazon and Google; analysis: it paid ~$351M back in distribution fees —  Anthropic's IPO prospectus shows how much it depends on a small group of customers and tech giants, highlighting key risks …

Axios : Mark Zuckerberg says the voluntary “accord” on AI standards that tech executives agreed to includes internal risk reviews and external auditors —  President Trump said Tuesday that top AI executives agreed to voluntary standards aimed at increasing industry oversight.

Samantha Subin / CNBC : Trump says he signed a “morally binding” AI doc with AI leaders; Mike Johnson calls it a statement of principles that are “voluntary on behalf of the industry” —  President Donald Trump on Tuesday said he signed a “morally binding” AI document with tech leaders following …

Bloomberg : After lunch with AI leaders, Trump rejects new federal AI rules, says “we automatically have regulation” via DOJ and FBI but “self-regulation is very important” —  President Donald Trump shot down the idea of new federal regulations on artificial intelligence …

Dannie Herzberg / Baseten : Baseten partners with OpenAI to let OpenAI enterprise customers use existing commitments for open models served by Baseten within Codex or via the Responses API —  Baseten is partnering with OpenAI to serve open models natively via Codex and the Responses API.  —  Authors  —  TL;DR

Elizabeth Dias / New York Times : Interviews with Anthropic co-founder Christopher Olah and 20 religious and philosophical leaders about their summits to investigate consciousness in Claude —  “To be clear,” Christopher Olah told me, “we don't know if A.I. models are conscious.  I don't know.  I'm genuinely uncertain.

Bloomberg : Sources: some potential Oura IPO investors decided to hold off, citing its ~$15B target valuation and poor post-listing performance of peers like Fitbit —  Some investors approached about potentially buying shares in Oura Inc.'s now-delayed initial public offering decided not to go forward …

Maximilian Schreiner / The Decoder : OpenAI prices GPT-6.1 Sol at $2/1M input and $10/1M output tokens, the same as GPT-6 Sol and Claude Sonnet 5.5, and says it performs well on safety tests —  Key Points … Ask about this article...  OpenAI is betting on cost efficiency over peak performance.

Sarah Perez / TechCrunch : OpenAI expands ChatGPT plugins with interactive panels and file viewers, launches a Plugin Creator tool, and adds MCP Events support for plugin automations —  OpenAI isn't only coming after productivity software with ChatGPT's new Pages, Slides and Space features — the AI lab is now targeting …

Dean Takahashi / GamesBeat : General Intuition, which trains AI agents in spatial reasoning via gameplay footage, raised $220M at a $6.2B valuation, for $650M+ in total funding —  General Intuition has raised another $220 million at a $6.2 billion valuation as the company starts making models available for training AI.

Paul Sawers / The New Stack : OpenAI says Plus and Pro subscribers can use their existing plan allowance with 16 launch partners, including Vercel and Devin, via Sign in with ChatGPT —  OpenAI has announced that it's making it easier for ChatGPT subscribers to use their allowance inside third-party AI and coding platforms such as Amp …

Frederic Lardinois / The New Stack : OpenAI unveils Decisions API, which returns predefined answers with confidence scores, available in limited preview with no pricing details —  With the sudden rise of TypeSafe's Jev, decision models have become incredibly popular, so it's not surprising that OpenAI is also announcing …

Kris Holt / Engadget : OpenAI announces ChatGPT Space, a collaborative hub for shared pages and files, available on Pro, Business, and Enterprise plans; it replaces Library —  ChatGPT is coming to Slack and Microsoft Teams too.  —  OpenAI unveiled several work-related tools at its Dev Day event today, including a hub designed for collaboration.

Bloomberg : Sources: OpenAI aims to raise at least $30B at a pre-money valuation of around $1.4T, meant to be a bridge round to provide it with capital in place of an IPO —  OpenAI aims to raise at least $30 billion from investors in a new round of funding, according to people familiar with the matter …

根据本周发表在《The Lancet Oncology》期刊上的一项研究，全世界八分之一癌症病例由感染引起。研究发现，2024 年感染性病原体相关的新癌症病例有 230 万例，占到了总新癌症病例的 12%。其中幽门螺杆菌会增加胃癌风险，这种病菌导致了 76 万例新癌症病例，数量最多，主要集中在东亚地区。其次是人乳头瘤病毒（HPV）导致了近 75 万例新癌症病例，撒哈拉以南非洲地区最多，这种病毒与宫颈癌、与肛门癌、外阴癌、阴道癌、阴茎癌等相关。乙肝和丙肝病毒则与肝癌病例相关。Epstein-Barr 病毒导致了 26 万例癌症病例，该病毒会导致鼻咽癌、胃癌以及霍奇金淋巴瘤。

意大利最大银行 Intesa Sanpaolo 旗下私人银行业务部门 Fideuram 的总裁 Paolo Molesini 今年 2 月成为一起组合利用 WhatsApp 假信息和 Deepfake 语音的诈骗行动目标，导致逾 1 亿美元被骗走，他本人则在 3 月份辞职。Molesini 首先是收到了看起来是母公司 CEO Carlo Messina 发来的 WhatsApp 信息，要求他帮助处理一笔巨额海外交易，接着他收到了一家大型律师事务所高级合伙人的电话。但打电话的不是真人，而是骗子利用 AI 模型模仿该律师的声音（即 Deepfake）。这通电话让 Molesini 相信之前的信息确实是其上司 Messina 发送的。Fideuram 随后向中国大陆以及香港的账号转了 9500 万欧元（约 1.08 亿美元）。在意大利、葡萄牙和中国有关部门及银行机构的协助下，Fideuram 追回了 5300 万欧元，其余款项则被通过海外复杂账号网络兑换成了加密货币。

继三星、海力士之后，全球第三大 DRAM 厂美光（Micron）在台湾的两个工厂，罢工行动也蠢蠢欲动。9 月 24 日，工会人数 2100 人，占全厂七成的美光桃园工会决定依据劳资争议处理法，10 月 1 日起进行为期一周的罢工投票。一旦支持罢工者超过工会会员数一半，就取得罢工权。有 1 万 2 千人的美光台中厂，台中工会也将于10 月 22 日和资方再次协商，若破局，也可能举行罢工投票。一位台中厂员工透露，原本工会成员维持在 400 人左右。但自三星和海力士宣布和员工达成协议，决定扩大对员工发放红利后，工会成员数目迅猛成长，现在台中厂工会成员已经超过一万人！等于超过 8 成员工参加工会，比例比桃园还高。触发美光员工举行罢工投票的关键在于分润制度。2025 年美光的净利已成长至 815 亿美元，但员工平均得到的分红只有 2.3 个月的月薪。比较其他半导体公司的分润后，工会认为美光的红利分配方式并不公平。美光工会认为， 目前三星半导体部门的营业利益 12% 分给员工，海力士是 10%。美光只有 4.4%。

Mozilla 释出了 Firefox 157。主要变化包括：新的 Nova UI，现代化浏览器界面边框、侧边栏及菜单外观；在兼容设备上支持 WebRTC 视频通话 AV1 硬件解码，此前它一直依赖于软件解码；修复 HDR 视频问题；改进了更改视频播放速度时的音画同步；移除内置搜索引擎 Amazon；等等。

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 5 月它宣布将从 2025 年 7 月起停止提供免费授权，转为提供折扣价付费订阅。今年早些时候没有转为付费的账号内相关数据都被删除了。根据一封发送给受影响客户的邮件，微软承认“由于错误”它在数据保留和导出期限结束前就删除了剩余数据，且尝试恢复数据的努力均告失败，“我们已经研究了恢复方案，遗憾的是，数据无法恢复。”微软甚至无法确定具体丢失了哪些数据。软件巨人对此向受影响客户表示歉意。微软没有披露受影响客户的数量，根据此前的报道，有 17.1 万非营利组织受到影响。

切开的香蕉片会在短时间内变成棕褐色。一种利用基因编辑技术修改过的新香蕉品种将能大幅延长香蕉保持新鲜的时间。英国农业生物技术公司 Tropi CEO Gilad Gershon 表示，新香蕉能延缓一到两天变黑，如果冷藏保存则保鲜期还会更长。Tropic 利用 CRISPR-Cas9 基因编辑技术，对香蕉的一个特定基因的所有三个拷贝进行了修改，减少了导致其变黑的多酚氧化酶（polyphenol oxidase）的产生。Tropic 表示，这种不易变黑的香蕉有望使整个供应链中的食物浪费及二氧化碳当量排放量减少 25% 以上。该公司称：“仅在香蕉出口市场，这一举措每年就能减少超过 900 万吨的二氧化碳排放。”新香蕉品种已在日本、巴西和菲律宾等 10 个国家获得监管批准，已在拉丁美洲投入种植。

AMD 宣布收购李飞飞联合创办的 AI 初创公司 World Labs，交易总额约 82 亿美元，采用全股票形式支付。 World Labs 成立于 2024 年，由李飞飞与 Justin Johnson、Christoph Lassner 和 Ben Mildenhall 等人联合创办，专注于开发世界模型，但目前还没有产品问世，它先后获得了超过 12 亿美元的融资，投资者包括了 AMD 公司。交易完成后，李飞飞将加入 AMD，担任执行副总裁兼首席科学家，直接向 AMD 董事长兼 CEO 苏姿丰汇报。

运行 Android 操作系统的 Googlebooks 新笔电即将上市，那么运行 ChromeOS 的旧设备将何去何从？根据 Google 支持页面“What the Googlebook announcement means for your ChromeOS devices”，搜索巨人声明：它将向 ChromeOS 设备提供定期更新和安全补丁至 2034 年年中；对于近期购买的合格设备，如果 10 年支持生命周期一直持续到 2034 年之后，那么 Google 承诺协助用户过渡到 Googlebook OS。相关措辞意味着 Google 计划到 2034 年停止支持 ChromeOS。Googlebooks 和 ChromeOS 设备一样提供十年的软件支持。

微软每次释出 Windows 11 例行安全更新，都会给 Windows 11 用户带来一系列新问题。9 月份的安全更新带来了远程桌面、USB 音频、Linux/WSL 应用、企业域、文件历史、Explorer.exe 以及 AMD 显卡相关的新故障。微软仍然在支持上一代的操作系统 Windows 10。Windows 10 的安全更新是否也会带来新问题？对比的结果并不出人意料：Windows 10 更新 bug 远少于 Windows 11。下一次重新安装操作系统 Windows 用户也许应该考虑更稳定的 Windows 10。

中国冰川主要分布在昆仑山、天山、念青唐古拉山、喜马拉雅山、喀喇昆仑山五大山系，新疆、西藏两地冰川面积与冰储量占全国总量九成左右。第三次中国冰川编目显示，截至 2020 年前后，全国冰川面积约 4.6 万平方公里，冰川共计约 6.9 万条。但对比 1960 年代基准数据，冰川总面积已经减少约 26%，有约 7000 条小冰川彻底消失；仅 2008-2020 十余年间，冰川面积又缩减约 6%，冰川进入快速退缩通道。天山乌鲁木齐河源 1 号冰川是观测时间最长的参照冰川，1959 年观测之初，冰川面积 1.95 平方公里，1993 年分裂为东西两支，2023 年总面积仅剩 1.496 平方公里。冰川被称为西部干旱区的“固体水库”，对下游河流径流起到天然调节作用。很多人简单认为冰川融化，河水就会源源不断增多，但科研观测揭示，这只是短期现象。随着冰储量持续耗损，当冰川体量下降到阈值之下，融水补给能力就会逐步衰减，流域会经历从“增水”向“减水”的转折。对于新疆、河西走廊及中亚内陆流域而言，冰川萎缩叠加降水波动，未来水资源不确定性将显著上升，农业灌溉、高寒生态系统都会承受压力。

SpaceX 于 9 月 28 日 8:15 a.m. EDT 在德州的 Starbase 发射了其重型火箭 Starship，执行第 14 次飞行任务，也是首次轨道发射，将 26 颗 Starlink V3 卫星送到轨道上。这次发射仍然是测试飞行，SpaceX 未尝试回收火箭，上面级 Ship 41 执行了减速点火，溅落在北太平洋海上。Starship 设计能将 150 吨有效载荷送到近地轨道上，将 100 吨有效载荷送到地球同步转移轨道上。Starlink V3 重 1.9 吨，相比下 V2 只有 575 kg。

Julia Van Etten 博士热衷于坐在沙发上用一台 300 美元的显微镜观察水滴，寻找其中的生命痕迹。2021 年她还是罗格斯大学的一名研究生时在位于北卡罗来纳州的家中度假。由于连日阴雨她被困在室内，母亲知道什么能让她振作起来，建议去室外采集些样本。她不抱什么希望，随便找个了公路旁的码头收集水样。但当她对其进行观察时，她震惊的发现了一种光合变形虫类原生生物 Paulinella。地球上所有能进行光合作用的植物都有色素体，色素体的起源可以追溯到 15 亿年前的一次事件，一个单细胞生物与一个光合细菌发生了融合，将其转化为一种产生能量的结构单元——科学家称之为细胞器。生物学家利用基因分析技术在 20 年前发现这种融合事件在 Paulinella 身上也发生过，但时间要晚得多，发生在一亿年前。对 Paulinella 研究有助于理解 15 亿年前导致地球遍布绿色植物的事件。Paulinella 属的第一个物种是德国生物学家 Robert Lauterborn 于 1894 年 12 月 24 日在莱茵河一处死水河湾的沉积物中发现的。他以继母 Pauline 的名字将其命名为 Paulinella 属。Van Etten 博士等人识别出了两种新的 Paulinella 物种——Paulinella marae sp. nov 和 Paulinella murrayi sp. nov。

沿海居民不仅需要担心海平面上升，还需要担心地面沉降。研究人员分析了卫星雷达测量的逾 1.9 亿垂直地面运动数据点，发现 2007-2020 年间美国沿海地区普遍存在地面沉降。全美 71.5% 的沿海地区沉降速度每年逾 0.1 厘米，墨西哥湾沿海沉降速度最快。43% 的美国海岸线每年沉降 0.2厘米或以上，23.3% 每年沉降至少 0.3 厘米。西海岸部分地区则出现地面上升。研究指出导致沿海地区地面沉降的原因包括自然地质过程如三角洲和湿地等区域软沉积物的自然压实，过度抽取地下水等。有 6710 万人生活在低度沉降区，1400 万人生活在中度沉降区，320 万人生活在高度沉降区。

你是否有过这样的经历，明知截止时间越来越近，却忍不住频繁刷手机、告诉自己“明天再说”？这种情况被称为拖延。一直以来，拖延被贴上懒惰、自律不足或意志力薄弱的标签，但这种看似影响学习工作以及身心健康的行为，为什么普遍存在且在进化过程中被保留下来？中国科学院心理研究所等团队证实，拖延与冲动这两种看似相反的行为，存在部分共同的生物学基础。有理论认为，拖延其实是非计划冲动性（无预案的即时冲动）的进化副产物。在远古时代，人类在野外生存，必须对眼前的机遇或危险做出快速反应。这种即时反应的本能有助于更好地生存。而现代社会的学习和工作，大多要求人们提前做好长期规划、延迟满足并持续执行目标。当偏好即时奖赏的本能遇到需要长期投入的环境时，拖延行为便容易产生。但这一假说一直缺乏系统性的研究证据。对双生子行为的分析显示，非计划冲动性存在中等遗传倾向，拖延的遗传倾向较高，且二者存在中等水平的遗传相关。整合既往研究数据后，这种遗传关联仍然稳健。研究还发现，拖延本身存在独特的遗传影响因素。因此，拖延不能被视作冲动性的完全副产物，个人经历、生活环境及具体任务情境，同样会对拖延行为产生重要作用。

澳门海关公布，近日积极打击水客活动，加强口岸关检执法，并运用科技手段严厉打击偷运物品进出本澳的不法行为。海关仅于一周内便破获 12 宗走私案，货物总市值达 151 万，当中包括 359 件 Apple AirPods 耳机及 2 张 GeForce RTX 5090 显卡等。12 名涉案人士年龄介乎 18 至 53 岁，当中包括 7 名内地旅客、3 名澳门居民及 2 名香港居民。澳门海关已根据《对外贸易法》对涉案人士作出起诉，相关违法行为一经证实，可被科处最高 10 万澳元罚款，所缉获的货物亦会宣告归澳门特别行政区所有。