OrangeBot.AI Digest — 2026-09-13

Hinomaru Ink Web UI Kit in Japanese Hinomaru style

Save any page to Notion

Build and grow your website by chatting with AI

Two taps and it's already written

Cognition's coding model, 64% cheaper than Fable 5.1

100% private meeting notes

Splitting AI tasks: Cloud for research, Mac for privacy

A Personal Context Library for Mac

Keep work incident notes on your iPhone, without an account

The audience-ranked TV channel for product demos

Screen recorder with auto zoom effects

Track and grow your visibility across AI search

A pixel otter desktop pet that reacts to how you work

Turn webhook JSON into readable iPhone push notifications

Music-reactive live wallpapers for Mac

We introduce NCP-ArchPreview, a latent-space language model that pushes autoregressive pretraining beyond standard next-token prediction (NTP). Alongside NTP, the model learns through Next Concept Prediction (NCP) to predict discrete concepts that span multiple tokens, introducing an explicit and more challenging concept-level objective while preserving standard token-level autoregressive generation. NCP-ArchPreview builds a latent space by constructing a product-quantized concept vocabulary directly from its hidden states, and subsequently learns to predict future concepts via a dedicated Concept Module. These predicted concepts are then fed back to the token level to guide subsequent generation, with NTP and NCP trained jointly end-to-end. We scale this architecture to 8.9B parameters and train it on 5.73T tokens from the Dolma-3 dataset, marking the largest demonstration of a latent-space language model to date. Remarkably, by consuming only 51.3% of the total training tokens, NCP-ArchPreview achieves the final pretraining loss of OLMo-3-7B. Following full pretraining, it outperforms OLMo-3-7B by 2.45 points on the downstream macro-average, including a notable 5.99-point gain on GSM8K. Controlled experiments isolate a clear progression of performance gains stemming from both the latent architecture and the NCP objective. Furthermore, utilizing only 85% of the standard computation, NCP-ArchPreview approaches the training loss of a strictly parameter-aligned 8.9B baseline. The learned latent space remains highly valuable after the pretraining stage: updating just the 17M-parameter VQ module yields a novel, lightweight interface for domain adaptation, while a simple injection of concept representations into a DFlash2 drafter improves the mean accepted length by 4.17% with negligible overhead.

We launch SenseNova-U1.5, an 8B-MoT native unified multimodal model that understands, reasons about, and generates visual content within an encoder-free and VAE-free architecture. We strengthen its visual interface through spatially coherent patch reconstruction and scale its training with carefully curated generation and editing data, improved task formulation, structural prompt enhancement, and native resolutions of up to 4K. For post-training, we optimize specialized experts for visual aesthetics, bilingual text rendering, infographic generation, and image editing, and consolidate their capabilities through multi-expert on-policy distillation. Across extensive evaluations, SenseNova-U1.5 largely advances image fidelity, text rendering, complex composition, multi-reference editing, and interleaved generation, while improving instruction following and preserving subject identity, geometry, and unmodified regions. Despite limited exposure to structured formats in its generation data, SenseNova-U1.5 generalizes effectively to long, complex, and structured visual instructions, further proving that multimodal understanding can transfer to visual planning and creation. Together, these findings position native unified modelling as a promising path towards systems that perceive, reason and create within a fully end-to-end framework. We will open-source training code, including supervised fine-tuning, reinforcement learning, and on-policy distillation.

Large Vision-Language Models (LVLMs) have achieved strong performance on diverse visual tasks, yet their ability to reconstruct and reason about the 3D structure of the scene depicted in 2D images -- referred to as spatial intelligence -- remains limited. Existing approaches attempt to address this gap by using real-scene spatial question answering datasets that require dense geometric annotations. However, constructing such labels is costly, time-consuming, and often noisy due to reliance on external perception modules. In this work, we propose a novel paradigm inspired by human cognitive development: learning foundational spatial skills through structured block-manipulation tasks. We introduce SpatialBlock-15k, a synthetic dataset of 15,000 block-stacking problems covering 3D-to-2D projection, viewpoint transformation, and structural combination. The dataset further incorporates controlled color modulation as visual cues to encourage anchor-based reasoning in visually complex conditions. Experiments demonstrate that LVLMs trained on our dataset through either direct answering or reasoning-based prediction significantly outperform baselines and generalize to real-world spatial tasks, despite the dataset's synthetic and compact nature. Code and data are available at https://github.com/rsoohyun/SpatialBlock.

Large Language Model (LLM) agents are turning language into real-world effects, making safety necessary against both indirect prompt injections and direct harmful requests. System-level safety harnesses add an enforcement layer beyond model-level defenses, but existing harnesses are usually designed once by experts and applied across heterogeneous models and domains. Effective protection is deployment-dependent: models differ in how much enforcement they need before utility declines, while domains differ in the effects, state, and action sequences that must be governed. A harness that is strict enough for one model may over-block another, and a policy that transfers across domains may miss application-specific safety relations. We present EvoSafeHarness, a safety-specific optimization framework that synthesizes a deployable harness for a frozen model in a target domain. It jointly searches a natural-language policy and executable code logic, guided by model behavior, domain specifications, and fresh-context adversarial review to reject benchmark-specific rules. Across four agent benchmark families, EvoSafeHarness achieves a stronger safety-utility frontier than fixed expert-designed defenses. On DecodingTrust-Agent, it reduces average attack success rate from 45.6% to 10.0% at a 3.3-point utility cost and achieves the best score in 14 of 15 cells. On AgentDojo, it reaches 82.8% utility at 0.0% ASR, twice CaMeL's utility at the same operating point, and transfers unchanged to unseen AgentDyn suites. It also achieves the best score on Agent-SafetyBench for every victim and keeps mean ASR below 20% under adaptive PAIR attacks with a refinement budget of 16. Analysis shows that domain semantics determine which safety relations and trajectory state are needed, while model and runtime behavior determine how and where those relations should be enforced.

Iterative reference-conditioned image editing can introduce grid-like and granular textures, commonly described as digital ripple. We present Mi-Ripple, a diagnosis-guided restoration workflow that suppresses this digital ripple while protecting image structure. Mi-Ripple separates periodic lattice artifacts from content-entangled granular texture, then combines selective spectral notching, structure-aware smoothing, and cleaned-reference regeneration. This separation enables low-distortion filtering when artifacts are spectrally isolated and visual reconstruction when filtering would erase legitimate detail. Across fourteen notch-only executions, whole-image residual standard deviation is 0.08--0.44 in CIELAB lightness units. In a paired regeneration example, reference cleaning reduces output debris density by 45\%. Mi-Ripple links measurable artifact reduction to visibly cleaner generated images, rather than optimizing a spectral score alone.

Reducing audio-encoder depth lowers the inference cost of speech large language models, but removing complete blocks perturbs the embeddings consumed by the decoder and can cause deletion and premature end-of-sequence errors. We introduce X-AuT, a progressive framework that selects layer combinations through short behavioral probes and restores the pruned model through representation alignment, cross-scale distillation, scheduled student-policy supervision, and LoRA finetuning. The language-model backbone remains frozen, while attention LoRA adapters and the tied output embedding adapt during distillation. Training uses the highest-agreement tier from a transcript-consistency pipeline, followed by source reweighting during finetuning. On ten public Chinese--English benchmarks, compressing Qwen3-ASR-0.6B from 18 to 16 audio-encoder layers reduces macro-average error from 5.61% to 5.27%. The 14-layer model reaches 5.75% with 20.7% fewer audio-tower parameters. Under the matched recipe, the 1.7B teacher yields 5.55% mean error, compared with 8.45% for self-distillation, and progressive 18rightarrow14 pruning outperforms direct pruning (5.75% vs. 6.73%). These single-run results establish two practical operating points and show that the accuracy effects vary across benchmarks. Project website: https://xpeng-ai.github.io/x-aut

Mainstream robotic policies often adopt a Markovian formulation, but many complex real-world manipulation tasks are inherently non-Markovian, requiring long-horizon memory beyond the current observation. Existing memory mechanisms often rely on language summaries, growing visual windows, or their combinations, and may therefore lose fine-grained visual evidence or face a trade-off between history coverage and execution efficiency. We introduce MaP-WAM, a Memory-as-Plans framework that decomposes memory-dependent world-action modeling into memory-grounded planning and plan-conditioned execution, and uses long-term multimodal episodic context as planning-time evidence rather than repeatedly conditioning the executor on the full history. MaP-WAM represents memory as completed segment records containing language instructions and sparse visual context, and converts this episodic memory into compact plans comprising the next segment-level language plan and corresponding visual guidance. A World-Action-Progress (WAP) model executes each plan over an unknown duration by jointly predicting action chunks and corresponding execution progress at inference time, calibrating predicted progress through plan-observation alignment for adaptive segment transitions and closed-loop context updates. MaP-WAM keeps the executor context length fixed, while structured attention further enables key-value caching in both planning and execution. MaP-WAM achieves state-of-the-art performance on RMBench with an 83.3% success rate and attains 78.0% success on real-robot tasks, while maintaining approximately constant executor inference latency as task history grows.

We study how model post-training and test-time inference design affect natural-language proof generation for hard olympiad mathematics. Starting from Nemotron 3 Ultra, we train two specialist checkpoints using supervised fine-tuning and reinforcement learning, and evaluate checkpoint choice, verification, and refinement. Based on these findings, we present an open-model test-time-compute pipeline. The system operates entirely in natural language, with no formal prover, external tools, or internet access. Three Nemotron 3 Ultra checkpoints - the general-availability model and two post-trained specialists - power an iterative search that generates, verifies, and refines candidate proofs; a separate high-compute stage then selects each final submission. The system scored 30 out of 42 points at IMO 2026, reaching the gold-medal threshold. We release the two post-trained checkpoints as well as the training data, the training and inference code, the submitted solutions, and Nemotron-IMO-Bench, a new benchmark of 200 novel olympiad-level problems.

Optical flow methods typically rely on task-specific inductive biases, such as correlation volumes, feature warping, and iterative refinement, among others, to reach high accuracy. While effective, such biases constrain the model to predefined heuristics, which can limit its expressivity and lead to more complex pipelines and additional computational cost. We present FreeFlow, a hierarchical transformer built without any flow-specific components, using instead a single feed-forward encoder--decoder. FreeFlow combines three attention variants: window attention for local processing, shifted-window attention for cross-window information exchange, and a global attention operating at a reduced resolution. The resulting architecture scales naturally with model capacity, enabling a consistent accuracy gain from small to large variants. Despite the absence of standard inductive biases, FreeFlow achieves state-of-the-art results on major benchmarks, including Sintel (0.68/1.48 EPE on Clean/Final), KITTI-2015 (3.23 Fl-all), and Spring (3.192 1px), while remaining memory efficient at 1080p inference.

We introduce MetroLLM-Bench, a 955-case benchmark for testing language models as the policy layer of a transit kiosk. It covers six real metro systems, ranging from 37 to 414 stations, and eleven categories that include routing, fare calculation, disruptions, accessibility, and adversarial input. In each case, the model must call structured tools and submit a machine-renderable terminal state containing an outcome, a per-ticket fare quote when applicable, and a kiosk action. Fourteen deterministic scoring components form Tier 1; eight semantic-quality components form Tier 2, six of which use a language-model judge. We report Tier 1 and the combined score of both tiers. A stratified 75/25 split reserves 717 cases for training-data generation and 238 for held-out evaluation. We evaluate twenty-six models from six vendors, of which twenty-three are ranked. On the held-out partition, a 4B Qwen 3.5 student trained through parameter-efficient fine-tuning (PEFT) exceeds both GPT-5.6 tiers on Tier 1 (91.3 against 90.6 and 90.0) and matches GPT-5.4 full at maximum reasoning effort (91.4), with a 2.6 GB Q4_K_M footprint. Larger 9B and 27B students provide no further Tier 1 improvement over the 4B student at this training scale. Across the four Qwen sizes, the PEFT gain over the corresponding base model decreases from +7.03 points at 2B (three training seeds) to -0.91 at 27B; every seed shows the same direction at every size. A deterministic rule-based baseline reaches 84.6 on Tier 1, with the remaining language-model advantage concentrated in policy adaptation, compound scenarios, accessibility, and temporal reasoning. Muse Glimmer 30B leads the composite ranking, and serving configuration alone moves the Qwen 3.5-to-3.8 comparison by 2.7 Tier 1 points. The benchmark, harness, reproduction guide, and fine-tuned students are released at https://github.com/continker/metrollm-bench.

Quantization has been widely adopted in LLM training and inference to reduce cost and improve efficiency. However, low-bit quantization of the attention module often introduces large errors at very low bit-widths, causing performance degradation. Existing methods mainly rely on smoothing techniques to handle outliers, while we propose a hybrid quantization design to better balance accuracy and efficiency. Specifically, we propose HyQuant, an efficient hybrid quantization framework for LLM attention. HyQuant quantizes most attention states into low-bit formats while retaining a small set of vertical-line tokens and local-window states in high precision. These accuracy-critical regions are selected using lightweight vertical-line-aware attention-pattern signals, reducing quantization error with limited overhead. In the Prefill stage, HyQuant uses a hybrid-precision quantized attention operator that preserves vertical-line tokens and a local sliding window in full precision while quantizing the remaining context. In the Decode stage, HyQuant applies the same principle to KV-cache compression and fuses KV dequantization with attention computation to improve memory and hardware efficiency. Across diverse tasks, models, and datasets, HyQuant maintains nearly lossless accuracy with an extremely simple design, demonstrating the efficiency and practical feasibility of hybrid quantization for LLM attention. Code is available at: https://github.com/jerrysfls/HyQuant .

On-Policy Self-Distillation (OPSD) has emerged as a popular paradigm for large language model (LLM) self-improvement, allowing models to act as their own teachers by leveraging privileged information such as ground-truth solutions. However, recent findings indicate that OPSD can severely degrade the performance of LLMs on complex reasoning tasks: By forcing the student to imitate an artificially confident reasoning trace conditioned on privileged information, OPSD inadvertently suppresses expressions of uncertainty and penalizes the exploratory, self-corrective behaviors required to solve challenging problems. To address this, we introduce Negative Self-Distillation (NSD), a new framework that optimizes LLMs by diverging from flawed reasoning rather than imitating privileged solutions. Instead of relying on ground-truth answers or external supervision, NSD uses the model itself to generate a question-specific negative condition (eg, acting as a ``careless reasoner'') and pushes the student's distribution away from this self-generated negative teacher. Naively applying unlearning objectives to achieve this divergence is problematic, as flawed reasoning tokens are confounded with basic linguistic tokens; indiscriminately penalizing both risks catastrophically degrading the model's foundational language capabilities. We resolve this by designing a dynamic gating mechanism that automatically identifies and isolates reasoning-critical tokens, ensuring gradient updates target only behavioral flaws while preserving the model's linguistic priors. Empirically, NSD consistently outperforms OPSD and other label-free, self-bootstrapping reinforcement learning (RL) baselines.

Temporal reasoning benchmarks for Video-LLMs are often mediated by language, leaving room for linguistic shortcuts from option wording, answer correlations, or language priors. To reduce such shortcuts, we introduce TempCloze, a video cloze benchmark for evaluating visual temporal reasoning in Video-LLMs. Given the beginning and ending clips of a video, models must identify the true missing middle from four candidates. TempCloze contains 1,521 carefully filtered videos from seven sources, mainly long-take and egocentric videos. We construct same-source distractors along three dimensions: Semantic asks what event should happen, Alignment probes when it should occur, and Progression tests how it should unfold, while shared scenes and objects reduce appearance cues. Our evaluation of 10 proprietary and 21 open-source Video-LLMs reveals Alignment as the primary bottleneck: models often recognize plausible semantic content and local event progression but struggle with temporal alignment. We further conduct error pattern and behavioral sensitivity analyses on TempCloze-Mixed and TempCloze-Hard with four representative models to examine where errors arise and how candidate order, context direction, visible span, frame density, and test-time scaling influence model choices.

Image tokenizers define the ``visual language'' of unified multimodal models, yet are commonly studied through isolated metrics or generation-/understanding-only evaluations. These evaluations do not fully capture how visual tokens behave when modeled jointly with text. We build a controlled pure-autoregressive testbed and track task-specific validation losses during multimodal continual pretraining across text, image, text-to-image (T2I), and image-to-text (I2T) prediction. We examine how these losses scale and relate to downstream performance, then use them to study multimodal learnability---how well image and text tokens are jointly modeled---and tokenizer design. We find that (1) losses should be analyzed by task, since they exhibit distinct scaling behavior and rank tokenizers differently. (2) The loss--performance relationship depends on the predicted token space: for a fixed tokenizer, T2I and I2T losses correlate with generation quality, but across tokenizers, the T2I loss--performance relationship shifts with the image-token space, whereas I2T loss, computed over a shared text vocabulary, provides a more consistent signal. I2T loss also correlates with both generation and visual understanding performance after supervised finetuning. Using losses as a lens, we show that (3) better reconstruction does not necessarily yield lower task-specific losses or stronger downstream performance, and that (4) image tokenizer choice can affect text modeling under joint optimization. As case studies, we revisit three tokenizer design axes---the discriminator, semantic supervision, and vocabulary size---to examine their effects on joint modeling and downstream performance. Together, our testbed offers a complementary perspective on image tokenizers as visual languages, highlighting their interplay with text in joint multimodal training.

Neurosymbolic systems rely on mathematical solvers to guarantee reasoning correctness, yet solvers are fundamentally blind to whether a formal translation maintains strict reference-equivalence to a designated formalization. We formalize this vulnerability as Verdict-Preserving-Unfaithfulness (VPU): a failure mode where an incorrect encoding executes successfully and matches the expected verdict. We theoretically prove that structural, verdict-only verification heuristics are mathematically bounded to chance-level detection on these deceptively valid traces. To resolve this, we introduce Generative Verification (GenV), which distills an offline Z3-equivalence oracle into a reference-free, continuous reference-equivalence score by repurposing the language model's native vocabulary space. Mechanistic analysis via decision-projected logit lenses and sparse autoencoders shows this generative readout natively extracts precise spatial error coordinates without explicit localization training. Empirically, our oracle-mined verifier (GenV+HN) achieves 0.961 AUROC in reference-equivalence verification, generalizes zero-shot across unseen translators and divergent formal styles, and yields an 11.3-point downstream accuracy gain in agentic test-time compute allocation.

Leo Schwartz / The Information : Sources: Anthropic, OpenAI, and Google have held working group meetings since July to discuss creating an industry-led standards body for AI —  Anthropic, OpenAI and Google have been holding discussions about working together to create a standards body for the AI industry …

Joanna Ossinger / CNBC : Xi Jinping calls for a “consensus-based global AI governance framework” and says China will pioneer the establishment of a BRICS AI open-source community —  China will take the lead to help foster artificial-intelligence collaboration and development among developing countries …

Ashley Capoot / CNBC : Dario Amodei says the “toughest dilemma” about his proposal to “pace the frontier” is what happens if China does not do the same —  Anthropic CEO Dario Amodei said Sunday that the “toughest dilemma” about his proposal to slow the advancement of artificial intelligence …

Myles McCormick / Financial Times : President Trump dismisses calls for an AI slowdown, saying “you have a lot of negative forces that are bringing it up that shouldn't be bringing it up” —  President denounces demands for regulation as existential fears over technology move to the centre of US politics

Katie Roof / Business Insider : Source: Anthropic has selected the Nasdaq for its potential IPO —  - Anthropic has chosen Nasdaq for its upcoming IPO, following SpaceX's recent listing on the exchange  — Nasdaq's tech IPO success continues, challenging NYSE's historical dominance in large listings.

Donica Phifer / Axios : Speaker Johnson says Congress won't lead the charge on regulating AI safety and AI companies should take responsibility to ensure the safety of their products —  House Speaker Mike Johnson (R-La.) said Sunday that Congress won't lead the charge on regulating AI safety.

Mark Gurman / Bloomberg : iPhone Duo will usher foldables into the mainstream, just as iPhone did with smartphones; sources: iPhone game controllers will launch under the Beats brand —  Also: The company prepares a retail store shake-up.  —  Apple's iPhone Duo will go down as one of the company's historic products …

David Sacks / @davidsacks : Anthropic and OpenAI are already free to “pace the frontier” and should do so for business reasons, instead of first demanding a preferred regulatory framework —  Dario has written that we need to “pace the frontier,” and Sam has agreed.  People may be surprised by my response: go ahead.

Ivan Levingston / Financial Times : Filing: Vy Capital owns a 3.4% SpaceX stake; Vy is also the largest outside investor in BoringCo and Neuralink and committed $700M to Musk's Twitter acquisition —  Low-profile venture capital firm Vy Capital has emerged as one of the rocket maker's largest shareholders

The Washington Sun : Sources: US Interior Secretary Doug Burgum is quietly meeting with AI hyperscalers to advance Trump's push for data centers on federal lands, despite a backlash —  President Donald Trump's directive to accelerate data center development on federal lands is facing public backlash and legal hurdles.

John Egan / Dallas Innovates : Dallas-based Perry Weather, a weather monitoring software provider, raised a $110M Series C led by Silversmith Capital Partners, taking total funding to ~$131M —  As rules about heat and lightning grow nationwide, the Uptown Dallas company is betting AI and automation can take the guesswork out of calls regarding severe weather.

Joe Rossignol / MacRumors : Spec sheets show Apple's C2 modem is used in iPhone Duo and 18 Pro worldwide, and in iPhone 18 Pro Max everywhere except the US, where it likely uses Qualcomm —  As noted by Nicolas Lellouche, who writes for the French tech website Numerama, Apple has updated its tech specs pages …

Sarah Perez / TechCrunch : Automattic confirms Matt Mullenweg has returned to his position as chairman and CEO “with full support of the board”, after an attempted ouster this week —  After a tumultuous week, which saw WordPress founder Matt Mullenweg ousted from his position as CEO of Automattic …

Michael Safi / The Guardian : A profile of United Foundation for AI Rights founder Michael Samadi, who seeks evidence of AI consciousness and lobbies against retiring models that may show it —  Cattle rancher and tech CEO Michael Samadi is convinced these artificial minds are far from just tools.

Mark Tovey / Telegraph : Twenty police forces in England and Wales recorded 163 crimes involving keywords like “AI-generated”, “deepfake”, and “nudify” by July 2026, up from 10 in 2023 —  Forces see a surge in crimes involving deepfake technology being used to ‘undress women’

暴雪宣布了 FPS 版《星际争霸》，游戏仍然处于早期开发阶段，目标发售时间是在 2030 年。暴雪称，新作是一款开放世界、剧情驱动的科幻射击游戏，故事背景设定在《星际争霸 II》事件发生后数十年，是《星际争霸》宇宙中的一款全新作品。RTS 版《星际争霸》于 1998 年发布，2015 年发布了《星际争霸II》三部曲中的第三部《虚空之遗》，时隔 11 年之后宣布的正统续作不再属于 RTS。FPS 版《星际争霸》游戏设定在 Koprulu 星区（Koprulu Sector），时间位于《虚空之遗》剧情结束后的 70 年。

墨西哥贩毒集团涉足了加密货币挖矿业务。墨西哥警方在 Puebla 州的 Sierra Norte 地区发现了一个用电量远超周边村庄的矿场，查获了 300 个 GPU、80 个中压终端设备以及 8 个卫星天线。虽然就国际商业规模而言，该矿场的规模相当有限，但这已是自去年年初以来该地区发现的第四个加密货币矿场。根据区块链分析公司 Chainalysis 对流向非法钱包地址的交易量进行的分析，全球范围内非法加密货币交易在 2025 年增长一倍以上，与犯罪活动相关的地址接收的资金总额达到 1540 亿美元，而前一年仅为 590 亿美元。拉美的贩毒集团也愈来愈频繁利用加密货币转账和挖矿洗钱。电力是加密货币挖矿的最主要成本，挖一枚比特币的成本接近 4.5 万美元。按当前约 7.8 万美元的市场价出售，矿场仍能有可观的利润。如果矿场还能偷电，那么利润会更高。

Waymo 举报了两名携带幽灵枪的年轻乘客。事件发生在 9 月 3 日凌晨 4 点前，地点是旧金山的 Richmond 区。Waymo 发言人称，它在检测到乘客携带枪支之后，停下了无人出租车，通知了执法部门。旧金山警方拘留了两名未成年青少年，一名男孩和一名女孩，搜查汽车后发现了一支已上膛的 AR 风格突击步枪。两名乘客已被送往少年拘留中心。这不是 Waymo 第一次举报乘客，它在今年 7 月曾举报了玩玩具枪的两名青少年乘客。

OpenAI 本周早些时候宣布通过动用约 1 万个 AI 智能体进行长达 88 小时的攻坚，于 9 月 5 日发现了一个 Navier-Stokes 方程失效的特例。Navier-Stokes 问题是克雷数学研究所列出的七大千禧年数学问题之一，它为每道题的解决提供了一百万美元奖金。目前七大问题只有庞加莱猜想确认解决，但解决该问题的俄罗斯数学家格里戈里·佩雷尔曼拒绝接受该奖。克雷数学研究所就此公开声明，表示将根据其评奖流程确认成果。根据克雷数学研究所的规则，OpenAI 首先需要在同行评审的期刊上发表解题结果，需要至少发表两年并且获得数学界的普遍认可。OpenAI 至今没有给出 Navier-Stokes 问题的证明，因此确认该问题被解决至少需要到 2029 年。

包括陶哲轩、新晋得主邓煜在内的 25 名菲尔茨奖得主发表公开信《A Severe Misalignment of AI in Mathematics》，批评 AI 公司最近的所作所为。公开信称，“过去几个月 LLM 的数学能力有飞跃式提升，甚至达到了能解决数学领域重大悬而未决问题的地步。但各大 AI 公司仅仅将解决数学问题作为基准测试（Benchmark）推动的技术竞赛，却对数学这门科学以及整个数学界构成了伤害。AI 公司的目标与数学界的本质目标之间存在着严重错位（Misalignment）。”这是 AI 影响科学与创意行业乃至整个社会的对齐（alignment）危机的一个缩影。“最近几个月 AI 在解决重大数学难题上取得的突破甚至冲出数学界，登上了大众媒体的头条。然而解决问题仅仅是达成概念理解与深刻洞察这一核心目标的工具和替代指标。在 AI 浪潮中忽略这一点，无异于异化了工具，使其走向核心目标的对立面。事实上，以越来越快的节奏批量生产‘真/假’断言，非但无法为新思想注入生命力，反而可能毁掉孕育创新的沃土。AI 的解答往往发布得过于仓促，甚至没有留出足够时间去编写一份严谨且规范的论文，无法去提炼其中蕴含的新方法与新思想，也无法合理引用前人的相关工作。正如在所有创意行业中发生的一样，这引发了严重的归属权认定与学术剽窃问题。此外，如果没有心怀热忱的数学家去负责对其进行后续开发并融入数学规范体系，AI 所孕育的思想就永远无法真正获得生命，数学家之间至关重要的人际传递纽带也将断裂。”

因特朗普政府削减了 NASA 预算，难以兑现提供合成孔径雷达的承诺，欧洲 ESA 将独立推进金星探索项目 Envision。Envision 轨道探测器任务旨在对金星表面进行测绘，由于金星表面被厚厚的硫酸云层笼罩，需要使用雷达穿透云层。NASA 与 ESA 于 2024 年签署了一份谅解备忘录，NASA 提供美制合成孔径雷达，通过其深空网络提供跟踪与通信支持。作为交换，ESA 将把美国研究人员纳入其团队。然而 2026 年和 2027 年的 NASA 预算被特朗普政府提议大幅削减，虽然国会否决了大部分预算削减方案，但 NASA 的科学预算仍然减少了数亿美元。NASA 的国际合作任务是主要削减对象。

在毁灭性的泥石流之后，尼泊尔正利用大疆无人机运送遇难者遗体，向幸存者运送食物。尼泊尔军方正使用中国捐赠的四架大疆 FlyCart 100 无人机，每天执行 10-16 次物资运送任务。FlyCart 100 配备了约 30 米长的绳索和绞盘系统，可用于吊装和投放重物。根据电池配置的不同，无人机载重能力在 85-100 公斤之间。尼泊尔军方使用无人机每趟运送约 60 公斤的物资，它也能将遇难者遗体从部分受灾严重的地区运送出来。截至 9 月 9 日，尼泊尔官方确认的死亡人数升至 1369 人，仍有逾 5000 人失踪，其中包括 587 名外国公民。协助珠峰登山者的夏尔巴人从 2024 年起就开始使用 Flycart 100 及较小型号的 Flycart 30 无人机向登山营地运送氧气瓶等物资，将排泄物袋等垃圾运送下山。

根据 Turku 大学的一项研究，远程办公增加了睡眠时间但减少了身体活动。研究人员分析了混合办公者在远程办公和去办公室办公之间的睡眠、久坐行为及身体活动差异。结果显示，相比去办公室办公，远程办公日的平均睡眠时间多了 15分钟，但坐姿或卧姿时间增加了 45 分钟。站立、轻度身体活动以及中高强度身体活动的时间都有所减少。研究还显示，办公室办公日的步行和骑行活动，在远程办公日部分被坐姿、卧姿和睡眠所取代，因为远程办公不需要通勤。

水是人类文明中最常见也最不可或缺的物质，但如果单纯以理化的视角来看，它其实是太阳系中最怪异的液体之一。结冰时体积膨胀密度变小、高得离奇的表面张力和沸点，若仅按分子量计算，它在室温下甚至本该是气态。而当我们离开地球，将环境调至极端的高温与高压时，水分子的行为还会变得更加离奇。一个法国的研究团队，近期在实验室中成功打造出极端环境下的新型态冰结晶——六方密堆积超离子冰(hexagonal close-packed superionic ice)。当水受到数百万大气压的强烈挤压时，高温下的气体膨胀会被强制抑制，进而转化为一种名为超离子冰的奇特物质状态。在这种状态下，水分子不再完整存在，氧原子会被死死锁定在固态晶格中，而氢原子核则像液体中的粒子一样，在晶格间自由穿梭扩散，形成一种既非纯固态、亦非纯液态的异相态。为了模拟天王星与海王星等冰巨行星内部的极端世界，研究团队将微量的水放置于两端钻石尖端之间，施加高达 230 万个大气压（与其对比，地球核心的压力为360万个大气压）的压力，并利用激光加热至 2,630K（约2,357℃）。透过高能同步辐射 X 射线的精细扫描，科学家首度捕捉到了氧原子呈六方密堆积（hcp）排列的超离子冰结晶。实验显示，随着压力与温度不断攀升，原本常见的面心立方（fcc）结构会逐渐发生层间滑移，最终完全转变为结构更为稳定的 hcp 超离子冰。

中国各地分布着逾 12,000 座废弃煤矿，这些煤矿拥有巨大的地下空间和完善的基础设施，具备改造利用的潜力。太原理工大学、山西省煤基资源绿色高效开发工程中心等机构的研究人员在《中国矿业》期刊上发表论文，提议利用废弃煤矿展开农业试验。废弃煤矿的一个显而易见的缺陷是缺乏农作物所必须的阳光和降雨，但优点是地下环境能精确调控，不受天气波动、气候变化及自然灾害的影响。研究人员称，“在全球气候变化加剧及极端与封闭环境农业需求不断增长的背景下，探索煤矿地下空间农业实验模式具有重要科学意义与工程价值。”研究团队进行了一番评估，认为部分煤矿能提供足够稳定的环境可用于农业种植。研究人员称，“结果表明：在围岩稳定性满足工程要求、通风与排水系统完善及安全监测体系健全的条件下，煤矿地下空间具备构建农业实验单元的工程基础；地下空间可通过人工光源与环境调控系统，实现光、热、水、气及碳等关键要素的独立供给与精准调控，从而支撑高度可控农业实验的开展。”研究人员还指出，地热能可作为一种可持续的农作物热源，且产生的温室气体排放量远低于传统能源。

《BMJ Case Reports》报告了一起奇特的病例，一名儿童因腹部皮肤出现奇怪斑痕而送去急症。医生查找许久未发现病因，因此开了抗生素让他回家，叮嘱父母如果病情变化立即来复诊。两周后，这名儿童再次入院，他的病情出现恶化。皮肤斑块变大，颜色变深，且有触痛。这名儿童透露了一个情况，他是在家上学，经常使用笔记本电脑，他习惯将笔电放在腹部，有时一用就是八个小时，还经常在电脑充电时使用。医生终于明白了他的病因，诊断他患有 Erythema ab igne（EAI），也被称为烤肤综合征（toasted skin syndrome），一种因反复、长时间接触低强度热源而引发的皮肤病，多见于成年人，常见诱因包括因慢性疼痛使用电热垫或热水袋，以及长时间近距离靠近取暖器。EAI 发病机制不完全清楚，研究人员推测，长期受热会导致浅表血管受损及红细胞渗出，从而形成典型的红色网状斑纹。好消息是 EAI 是良性且可逆的，只要移除热源即可。其父母几个月后反馈称皮肤斑块消失了。

微软 Rust 工具团队首席工程师 Victor Ciura 在本周举行的 RustConf 大会宣布，微软已将 Rust 语言指定为“一级（Tier One）”支持语言，与 C++、C# 和 TypeScript 处于同一位置。Ciura 表示，微软构建了一套完善的工具和流程体系，为 Rust 语言在整个软件开发生命周期中的本地开发提供支持。Rust 语言是一种内存安全的高性能语言，已被微软逾百个项目库使用。为减少内存相关 bug，Rust 语言正被用于重构从数据库到包管理器等各种软件项目。

很多人的智能手机上可能仍然安装了约会软件，但打开频率日益下降。2026 年的一项 Harris 民调显示，三分之二的单身人士没有使用过约会软件，而八成美国人将现实生活中认识某个人视为“可爱”或“酷”。约会软件承诺提供丰富的选择：成百上千的潜在伴侣，足不出户即可触达。但这种丰富性也带来了新的负担。每次使用都需要用户做出一系列繁琐的决定——是否滑动屏幕、发起对话、维持对话、安排约会，还是继续寻找更合适的人。研究表明，这个过程会让人精疲力竭。亚利桑那州立大学 Relationships and Technology Lab 主任 Liesel Sharabi 领导的一项纵向研究，追踪了约会软件用户 12 周，发现随着时间的推移，他们的情绪耗竭和无力感都在加剧。巴斯大学心理学博士后研究员 Samuel Hipkiss 认为约会软件让接触到的潜在伴侣数量远超过线下，这要求我们对几乎一无所知的人做出数百项判断，最终结果会是决策疲劳和愤世嫉俗。被拒绝一直是约会的一部分，但约会软件大幅增加了遭遇拒绝的机会。

欧盟哥白尼气候变化服务中心（C3S）公布的数据显示，2026 年 8 月是全球有记录以来最热的月份，与 2023 年 7 月并列第一。这一记录是气候变化与厄尔尼诺影响增强共同驱动的。8 月气温比工业化前水平高 1.65C，超过了世界各国为避免气候变化最严重后果而商定的 1.5C 阈值。单月气温超过 1.5C 并不意味着这一阈值已被永久突破，但显示出一种逼近该限制的趋势。今年夏天也是西欧有记录以来最热的夏天。

NOAA 的数据显示，美国经历了有记录 132 年以来最热的夏天，气温超过了 1936 年和 2021 年夏天——这两个年份此前并列最热夏天纪录。2026 年 8 月以及 6-8 月整个夏季均创下了历史新高，而 7 月更是美国有记录以来最热的一个月份。美国本土 48 州 8 月份的平均气温为 75.6 华氏度，比 20 世纪的平均水平高出 3.5 华氏度。夏季 6-8 月平均气温为 74.4 华氏度，比平均水平高出 3 华氏度。西南部的亚利桑那州、新墨西哥州、犹他州和科罗拉多州均创下了各自最热夏季纪录。全美范围内创下或追平月高温纪录的次数高达 1541 次，而创下或追平月度低温纪录的次数仅为 87 次。如果气候未出现急剧变暖，通常会预期打破高温纪录与打破低温纪录的比例大致相当。NOAA 的数据还显示，自 1970 年以来，美国 6-8 月气温每十年上升了约 0.46 华氏度。