OrangeBot.AI Digest — 2026-07-25

Talk to any app on your Mac

The teleprompter that actually listens

The easiest way to deploy a managed agent.

Transform your web audio with real-time effects

The easiest way to sell on Facebook Marketplace

The SSH workspace you'll actually enjoy using

Clip living pieces of the web onto one canvas

Caps Lock keeps your Mac awake, even with the lid closed

Give Claude Code and Codex a voice

The AI database client for Mac

Your AI memory, everywhere you work

All your coding agents, synced everywhere, free forever

Cloud for your AI Agent's tools and functions

Your personal health companion, powered by ChatGPT

iPhone's Dynamic Island, brought to your Mac's notch

Deep research requires agents to find answers that jointly satisfy multiple constraints. Discovering such answers is costly, whereas verifying a candidate can often be decomposed into tractable constraint-wise checks. This discovery--verification asymmetry suggests that a research agent should do more than simply search longer: it should recursively improve its current answer by verifying intermediate results and using the partially verified state to guide subsequent refinement. We introduce AREX, a family of Recursively Self-Improving (RSI) deep research agents. AREX alternates between an inner research loop that gathers evidence and constructs a provisional answer, and an outer self-improvement loop that audits the answer constraint-wise, identifies unresolved claims, and launches targeted follow-up research. To sustain RSI over long horizons, AREX learns an autonomous context-update tool that compresses growing interaction history into a compact improvement state preserving verified evidence and unresolved constraints, without relying on an external model. We train AREX on verified synthetic tasks and high-quality trajectories through agentic mid-training and long-horizon reinforcement learning. To mitigate sparse final rewards during long horizon learning, we emphasize key steps where decisive evidence is acquired or erroneous research directions are corrected. We instantiate a dense 4B model and a 122B-A10B Mixture-of-Experts model. Across BrowseComp, WideSearch, DeepSearchQA, Humanity's Last Exam (HLE), and other reasoning and tool-use benchmarks, AREX substantially outperforms comparable-scale baselines and remains competitive with models using substantially more activated parameters.

Large language models are increasingly used in K-12 education, but existing benchmarks mainly test exam question answering rather than understanding how curriculum knowledge is structured and visually presented. We call this capability curriculum cognition. It covers prerequisite chains, concept taxonomies, experiment-concept links, pedagogical sequencing, and visual grounding. We introduce K12-KGraph, a curriculum-aligned knowledge graph extracted from official People's Education Press textbooks in mathematics, physics, chemistry, and biology across primary, middle, and high school. It contains nine node types and fourteen relation types covering curriculum structure and visual grounding. From this graph, we derive K12-Bench, a 23,640-question multi-select benchmark with five task families: Ground, Prereq, Neighbor, Evidence, and Locate. We also build K12-Train, a graph-guided supervised fine-tuning corpus of 7,335 samples, including 2,267 text-only QA pairs and 5,068 multimodal VQA pairs. On K12-Bench, Gemini-3-Flash achieves only 57 percent exact match and Gemma-4-31B-IT reaches 46 percent, with Prereq and Neighbor being the hardest tasks. Our training experiments show that domain-specific supervision can reduce this gap. Under a matched 2,300-sample budget, K12-Train-Text consistently outperforms equally sized subsets of eight mainstream instruction-tuning corpora on GaokaoBench and EduEval. For vision-language models, K12-Train-Full achieves the best overall results on Gaokao-MM, MDK12-medium, and K12Vista among all compared training configurations, despite using fewer samples than the full DataFlow and WizardLM baselines. It also surpasses both text-only and multimodal-only variants, showing that textual and visual supervision are complementary. We release the graph, benchmark, training data, and complete construction pipeline.

Embodied visual tracking (EVT) requires a mobile agent to continuously follow a specific target described in natural language using only onboard vision. While recent vision-language-action (VLA) policies unify target identification and trajectory planning, their chain-of-thought (CoT) reasoning often operates in abstract spatial latents that are difficult to supervise and weakly aligned with explicit image-space detections. To address this, we introduce ReferTrack, a referring-then-tracking paradigm that grounds EVT using a single forward-facing camera. Our model first selects the target from an indexed set of bounding boxes, then decodes tracking waypoints conditioned on this image-grounded decision. To preserve target motion cues over time, ReferTrack maintains a sliding-window queue of previously selected bounding boxes, injecting their geometric features into the visual history via temporal-viewpoint-bbox indicator (TVBI) tokens. We further enhance target identification by co-training on a custom Refer-QA dataset. On EVT-Bench, ReferTrack achieves state-of-the-art single-view performance with success rates of 89.4%, 73.3%, and 74.1% on the single-target, distracted, and ambiguity tracking splits, respectively -- matching or even surpassing several multi-camera baselines on identification-heavy tasks. Finally, real-world deployments on legged and humanoid robots validate its robust sim-to-real transfer capabilities. Code is available at https://github.com/MedlarTea/referTrack.

On-policy self-distillation (OPSD) is promising as it removes the external teacher required by on-policy distillation (OPD), yet it still needs asymmetric information between teacher and student to ensure that the self-teacher provides a stronger learning signal than the student. Existing methods create this asymmetry either through privileged answers or visual evidence. We ask whether both can be removed, yielding a simpler form of OPSD driven purely by input conditioning. For this purpose, we propose Visual Contrastive Self-Distillation, namely VCSD, which converts image-content removal into an on-policy self-distillation signal. At each student-generated response prefix, the EMA teacher produces two next-token distributions under the same prompt and prefix -- one conditioned on the original image and the other on a content-erased control. Their token-wise log-probability difference highlights candidates whose likelihood is specifically increased by the instance-level visual content. We use this contrast to sharpen the teacher's original-image distribution within its plausible support, and distill the resulting full-distribution target into the student. Using ViRL39K dataset, VCSD consistently outperforms matched OPSD across Qwen3-VL and Qwen3.5 models. For example, on Qwen3-VL, it improves the seven-benchmark aggregate from 62.27% rightarrow 67.04% at 2B, 71.30% rightarrow 73.16% at 4B, and 72.51% rightarrow 76.26% at 8B. Furthermore, VCSD requires no external teacher, privileged answers, visual evidence signals, reasoning traces, or additional inference-time cost.

Spatial intelligence is essential for agents to move from static semantic understanding toward interacting with the physical world. Many spatial tasks are grounded in continuous visual scenes, where locations, regions, and paths are more naturally expressed by pointing, marking, or drawing than by reporting precise coordinates or discrete textual symbols. Yet existing spatial reasoning benchmarks usually require coordinates, options, or text, creating an answer-interface mismatch for image-generation models. This makes it difficult to evaluate image-generation models under the same task semantics as text-output VLMs, despite their ability to externalize spatial judgments directly in pixel space. We propose ProVisE (Protocolized Visual Evaluation), a benchmark-agnostic framework that elicits protocol-constrained visual answers from image-generation models and parses them into structured predictions compatible with original metrics. ProVisE also includes an Agentic builder that constructs and validates task-specific protocols for new benchmarks. We further introduce SpatialGen-Bench, a curated diagnostic benchmark of 470 samples across 14 spatial subtasks, four capability levels, and diverse answer forms. We evaluate representative text-output VLMs and image-generation models in a unified setting and validate Agentic protocol construction on six external spatial benchmarks. Results show that image-generation models are competitive when spatial answers can be externalized directly in pixel space, while text-output VLMs retain a clear advantage in compositional spatial reasoning. These findings reveal complementary strengths of pixel-space expression and text-based reasoning and establish a metric-compatible testbed for studying spatial cognition in image-generation models.

Traditional agent development is split across prompt templates, tool schemas, callback code, and workflow graphs. We present NVIDIA Object-Oriented Agents (NOOA), a model-agnostic Python framework for building reliable AI agents. NOOA takes a simpler approach: an agent is a Python object. Its methods are the actions the model can take, fields are its state, docstrings are its prompts, and its type annotations are contracts. A method whose code body consists of "..." is completed at runtime by an LLM-driven agent loop, while methods with normal bodies remain standard deterministic Python. This gives developers and agents the same interface, so agent behavior can be tested, traced, refactored, and improved just like other software. This paper makes three contributions. (1) We present the agent-as-a-Python-object programming model and the design principles behind it. Where Python has existing abstractions, we adopt them directly. Agent-specific capabilities--context, events, state rendering, long-term memory, and validated LLM loops--are exposed through simple Pythonic APIs, so both developers and agents share one familiar programming model. (2) We identify six model-facing ideas that NOOA is, to our knowledge, the first to combine on a single surface: typed input/output, pass-by-reference over live objects, code as action, programmable loop engineering, explicit object state, and model-callable harness APIs for context and events. We find the community already converging on several of these ideas--often as experimental or partial features--and present the comparison to encourage further adoption. (3) We demonstrate that current models use this interface effectively, both in targeted capability tests and on agentic and reasoning benchmarks such as SWE-bench Verified and Terminal-Bench 2.0 and ARC-AGI-3.

We introduce Tencent WorkBuddy Bench, a multi-domain evaluation suite for coding agents; this report documents its construction methodology, scoring protocol, and a cross-model leaderboard. At its core is a unified evaluation framework for constructing and running distribution-informed coding-agent tasks across four work domains - Code, Web, Office, and Security. Rather than adapting public issue text, every task is reverse-engineered from a real commit, pull request, or business scenario and rewritten as a short, colloquial, role-played request, so that a task's prompt is not recoverable by web-searching the underlying issue, pull request, or commit thread. Because the dataset is released openly - task directories, environment images, evaluation harness, tests, and reference solutions - contamination resistance rests on this construction together with dataset versioning rather than on secrecy. The four subsets - repository-level engineering, front-end development, office and business workflows, and red-/blue-team security - probe complementary facets of real work, each with its own verification style. All are packaged in a uniform task-directory format and run, under a uniform and reproducible protocol, on two agent harnesses (CodeBuddy Code and Claude Code); the full open release makes the benchmark reproducible end to end and directly auditable, since any third party can re-run each task and inspect its content. Because each subset uses a different scoring instrument, scores are not comparable across subsets and the suite reports no suite-wide average. We report a cross-model leaderboard across several model families.

We introduce SANA-Video 2.0, a hybrid video diffusion transformer instantiated at 5B and 14B scales under a unified architecture. Designed to generate high-quality video up to 720p on a single GPU, SANA-Video 2.0 matches full-softmax video DiTs in quality while retaining the favorable long-sequence scaling of linear attention. To avoid quadratic attention throughout, Hybrid Linear-Softmax Attention combines gated linear attention for O(N)-dominated mixing with periodic gated-softmax anchors at a 3:1 ratio, restoring the full-rank token interactions that pure linear attention lacks. To propagate these refreshed representations across depth, Block Attention Residuals (AttnRes) route completed block summaries into later linear layers, enabling anchor-feature reuse and boosting deep-layer effective rank by ~12%. Through from-scratch training, SANA-Video 2.0 learns the complete hybrid directly rather than linearizing pretrained models, with reduced-resolution proxy studies establishing 25% softmax as the optimal quality-efficiency trade-off. With 40-step sampling, SANA-Video 2.0 achieves a VBench score of 84.30 in 13.2s at 480p on a single H100, remaining competitive with far larger softmax video DiTs at a fraction of the latency. Its compiled DiT forward pass is 3.2x faster than a matched full-softmax baseline at 720p/60s, a gap that expands with video duration. Furthermore, full-stack Sol-Engine optimization (kernel fusion, caching, and sparse attention) accelerates this hardware-friendly backbone by a further 3.58x, bringing the 5B pipeline to 13.06s at 720p/5s and making it 120x faster than Wan 2.2-A14B on one H100. Overall, our hybrid design recovers softmax-level expressiveness at substantially reduced cost, unlocking scalable long, high resolution video generation.

When a real-world scene is captured by a smartphone camera and viewed on its screen, the displayed image often differs noticeably from the original scene in color, brightness, and contrast. This gap persists despite substantial advances in both modern cameras and displays. A key reason is that most pipelines factor the high-dimensional capture-to-display process into two separately calibrated camera and display stages, and then connect them through low-dimensional color transforms, leading to information bottlenecks and inevitable error accumulation. To address this systemic challenge, we propose Color Pass-Through, an end-to-end learned framework that operates directly on captured images. Our key insight is to treat the camera and display as a coupled system rather than calibrating them in isolation. Coupling the camera and display yields two practical advantages: (1) it brings the entire real-world scenes to the display via end-to-end optimization, and (2) it allows efficient one-step calibration for each distinct observer via complete capture-to-display path. We validate Color Pass-Through using both digital and human observers. Compared with representative baselines, our method achieves an average gain of +2.0 points on a 5-point user study and more than 2x improvement on quantitative metrics, demonstrating improved reproduction of the perceived color of the original scene.

As LLMs become more capable, they are increasingly deployed as collaborative agents, taking on user-delegated tasks through iterative interaction. Yet genuine interaction is inherently dynamic: users rarely specify their intent upfront, instead disclosing, revising, and reshaping it as the conversation unfolds. Despite this, LLMs are still predominantly evaluated or trained in single-turn, fully-specified settings, leaving open a fundamental question: how well do LLMs track and act on user intent as it evolves over the course of a conversation? To study this, we introduce a framework that transforms static, single-turn tasks into dynamic multi-turn conversations in which the user's intent evolves across turns--incrementally revealed, revised, and at times redirected mid-conversation--while preserving each task's original evaluation protocol, enabling existing benchmarks to be reused as controlled testbeds without new annotation. Across multiple tasks, we surface a consistent phenomenon: strong static-setting performance does not transfer to the evolving-intent setting, with substantial drops across model families. Our findings point to a fundamental gap: today's LLMs do not yet faithfully track and act on the user's evolving intent, a capability invisible to static evaluation yet critical for future collaborative agents.

Understanding motion in video is a fundamental challenge for visual learning, as frame-to-frame change entangles two sources of dynamics: camera motion and object motion. This decomposition has remained underexplored in representation learning, partly because these factors are tightly coupled in natural videos and difficult to supervise separately. Yet recovering it is important for learning robust motion representations that separate meaningful object dynamics from camera-induced variation. We study whether such structured motion representations can be recovered from frozen features of a pretrained image vision transformer. We propose the Structured Dynamics Model (SDM), which explicitly separates the dominant source of temporal change from residual dynamics through future-feature prediction, rather than representing video change with a single entangled latent or with unstructured, spatially dense transition tokens. Training combines self-supervised learning on real video with weak supervision of scene dynamics on synthetic Kubric data. We evaluate SDM on ProbeMotion, a new evaluation suite spanning synthetic and real videos with camera motion, object motion, and combined dynamics. SDM outperforms backbone baselines using global CLS or average-pooled features, and compares favorably to strongly supervised representations such as VGGT on several probes, despite using substantially weaker supervision. These results suggest that pretrained image models can be readily repurposed into structured video-dynamics representations, providing a useful inductive bias for learning and analyzing latent video dynamics.

Multi-agent interactive world models should not only generate consistent observations, but also maintain world states that persist across agents and evolve across views. Existing autoregressive video diffusion pipelines carry forward observation history as conditioning context, which makes shared state difficult to maintain in multi-agent and multi-view settings. We present WorldWeaver (W^2), a streaming multi-agent video diffusion model that augments rollout with cross-agent world state registers: learnable tokens that store shared world information, track individual agent status, and are dynamically updated after each generated chunk. We ground these registers with supervision signals spanning individual agent status, global state views including bird's-eye views, and scene text. We further improve the architecture with a Mixture-of-Transformers design that uses separate weights for world state modeling and visual frame modeling. Extensive experiments in two-agent Minecraft video generation show that explicit world-state modeling improves logical consistency and generation quality.

Real-world agent learning is often constrained by costly environment interactions, such as running time-consuming experiments or obtaining human feedback. In-context learning offers a highly sample-efficient way for agents to learn from their own interaction histories, but its gains disappear once that experience is removed from the context. Separately, context distillation provides a mechanism for internalizing contextual information into model weights. However, applying it to agents' interaction histories without sacrificing environment sample efficiency remains underexplored. We term this problem Experience Distillation and develop an implementation that requires no further environment interaction beyond the collected experience. Experiments on 749 curated software-engineering tasks and six text-adventure games show that it retains at least 64.8\% of the gains from in-context learning across both domains, whereas direct supervised fine-tuning on the collected experience recovers only 3.8\%. Compared with classical reinforcement-learning baselines, in-context learning from trial-and-error experience followed by Experience Distillation matches their performance with at least \(9.6\times\) fewer environment samples.

We study on-policy distillation (OPD) for agentic tasks, where an LLM agent interacts with an environment over multiple turns and a student imitates a teacher over these multi-turn interaction histories. Fully online OPD is costly because each update requires fresh student rollouts through the environment and teacher queries at visited histories. We propose Replayed-Prefix On-Policy Distillation (ReOPD), an off-environment alternative that reuses pre-collected teacher trajectories as replayed prefixes: the student acts at selected steps, while the teacher provides dense per-step supervision without executing new environment interactions. We show that multi-turn OPD introduces a prefix trap: making histories more student-on-policy improves relevance to the student, but can query the teacher on histories where its target is unreliable. This creates a two-sided distribution shift between student occupancy and teacher reliability. ReOPD addresses this by treating multi-turn OPD as a reliability-aware prefix distribution design and implements it with a simple step-decaying sampling schedule that emphasizes early, lower-shift prefixes. Across mathematical reasoning with Python and search environments over multiple teacher and student model scales, ReOPD preserves or improves OPD-level accuracy, uses zero tool calls during student training, and is at least 4times faster per rollout than OPD. ReOPD therefore turns expensive agent-environment interaction into a reusable offline resource, enabling scalable distillation across tools, tasks, and environments.

Deploying navigation systems at scale requires a recipe that minimizes sensor assumptions, generalizes across robot embodiments, and trains efficiently. Yet, today's best systems depend on depth sensors, multi-camera rigs, or pre-built maps, limiting the hardware they support and increasing deployment cost. We introduce Robostral Navigate, an 8B vision-language model built around this scalability objective. The model consumes only a stream of monocular RGB images - the most ubiquitous sensor across robotic platforms and predicts waypoints by pointing to the next target location in the current camera view. Operating purely in image space, rather than robot-specific coordinates, makes the policy naturally robust to changes in camera intrinsics and scene scale, enabling deployment across wheeled, legged, and aerial robots without recalibration. We generate 2.4 million trajectories across 350k simulated scenes to reduce the reliance on real-world data collection and scale easily. We further introduce a prefix-caching training recipe that packs entire episodes into single training sequences, reducing training tokens by 22x and cutting training time from months to days. A tree-based attention mask prevents conditioning on previous ground-truth actions, encouraging visually grounded action prediction, and reinforcement learning is used to further improve exploration and recovery capabilities. On the Room-to-Room and Room-Across-Room in Continuous Environments (R2R-CE and RxR-CE) benchmarks, Robostral Navigate sets a new state of the art. On R2R-CE, it achieves a 77.4% success rate, surpassing the best monocular method by 10.5 points and the strongest depth- or multi-camera system by 5.3 points despite using only a single RGB camera. On RxR-CE, it reaches 75.1% success rate, outperforming all monocular baselines.

Reuters : China's market regulator says it had fined and confiscated ~$770M from Trip.com for abusing its dominant position in the domestic online hotel-booking market —  China's market regulator said on Saturday it had fined and confiscated a total of 5.2 billion yuan ($770 million) from Trip.com Group …

Ima Jackson-Obot / Financial Times : Several universities including Yale, Johns Hopkins, and the University of Waterloo have restricted or disabled their use of AI detectors over accuracy concerns —  Some institutions are overhauling assessment and trying to move away from the emphasis on surveillance

Pei Li / Bloomberg : Sources: DeepSeek told investors it is suspending its second funding round after remarks attributed to Liang Wenfeng on US-China AI competition went viral —  DeepSeek has told prospective investors in its second fundraising round that it's suspending the deal for now, people familiar with the matter said …

New York Times : Sources: OpenAI and Anthropic quietly lobby Washington regulators to restrict open-source AI models, even as Sam Altman publicly says he supports open source AI —  Anthropic and OpenAI are clashing with the rest of the tech industry over whether “open-source” models from China should be freely available or restricted.

Ian King / Bloomberg : SK Group Chair Chey Tae Won says Anthropic has asked SK Hynix for supplies to make its own chips, calling it remarkable that an AI developer has chip ambitions —  AI developer Anthropic PBC has asked SK Hynix Inc., one of the world's biggest manufacturers of memory chips …

Akriti Sharma / Bloomberg : As student protests spread across India, Narendra Modi turns to Reels, swapping his trademark cinematic videos for self-recorded messages aimed at Gen Z —  As student protests spread across India, Prime Minister Narendra Modi has turned to Meta Platforms Inc.'s Instagram Reels …

Charlie Campbell / Time : A profile of Hangzhou-based Unitree, which shipped 5,500 humanoid robots in 2025, accounting for 25%+ of the global market, as it prepares for a Shanghai IPO —  Spectators typically leave a UFC fight with ringing ears and a sheepish resolve to get back to the gym.  Wang Xingxing went straight to his laboratory.

Hooyeon Kim / Bloomberg : Samsung announces a $200B+ contract to make chips for Broadcom through 2030, focusing on its 2nm and below process technologies for Broadcom's products —  Samsung Electronics Co. won a contract worth more than $200 billion to make chips for Broadcom Inc., as the companies race to win a larger share of the AI infrastructure market.

Financial Times : A profile of Yang Zhilin, founder of Moonshot AI, which faced early doubts over revenue and model capabilities before its Kimi K3 delivered a “DeepSeek moment” —  Known as ‘Yang the genius’ by classmates, his new K3 model has closed the gap with leading US rivals

Lily Mae Lazarus / Fortune : Candid Health, which develops tech and AI agents to automate medical billing and claims processing workflows, raised a $120M Series D led by Sixth Street Growth —  Candid Health cofounder Doug Proctor spun his laptop around to show me a herd of wild horses grazing behind him in upstate New York during our first Zoom call.

Jaspreet Singh / Reuters : Verizon CEO Dan Schulman says Verizon will provide dark-fiber connectivity for Google's data centers in a $1B+ deal, with more such deals in the pipeline —  Verizon CEO Dan Schulman said on Friday the company has secured a deal with Google (GOOGL.O), valued at more than $1 billion …

Abhinaya Prabhu / Tech Funding News : Meshy, whose AI-powered tech lets users generate 3D assets from text and image prompts, raised a ~$400M Series B at a $1.5B valuation —  - Meshy raises nearly $400 million at a $1.5 billion valuation, which it says is the largest round any dedicated AI-3D company has raised.

Financial Times : Analysis: US tech companies have cut ~140K jobs YTD, or more than 33% of announced US layoffs in 2026, with ~50K cuts from Amazon, Oracle, Meta, and Microsoft —  Investment spree is reshaping Silicon Valley while wider US jobs market holds steady  —  US tech companies have shed nearly 140,000 jobs since …

Larry Dignan / Constellation Research : Progress Software agrees to acquire Domo's AI and data platform business for $400M; Domo will remain publicly listed and change its name after the deal closes —  Progress Software said it will acquire Domo's AI and data platform business for $400 million.  —  According to Progress Software …

Maria Deutscher / SiliconANGLE : Neocloud Fluidstack, which has partnered with Anthropic, announces that it raised an $830M Series A led by Situational Awareness at a $7.5B valuation in January —  Fluidstack Ltd., a startup helping Anthropic PBC build artificial intelligence data centers, has raised $830 million in funding.

Debian 项目正在讨论如何使用 LLM 以及是否允许使用 LLM。其中一个提议是明确禁止使用 LLM，但只限于 Debian 本身，上游项目以及上游安全补丁不包含在内。另一个提议是允许 AI 辅助贡献，但需要确保 AI 生成代码符合 Debian 的各项规定，贡献者需要承担责任和履行披露义务。开发者指出，LLM 存在版权和许可方面的问题，也不能确保质量和准确性，会增加社区审核者的负担，AI 公司在训练大模型时无视版权和许可抓取数据，伤害了整个自由软件社区，对无数用户依赖的网站发起了一波波拒绝服务攻击，它的训练也消耗大量资源。

美国司法部起诉了一位公民在入境搜查时提供了所谓的“胁迫密码”，输入该密码导致手机上的数据自动删除。这起案件被认为是美国首例“胁迫密码”案例。本案的被告是亚特兰大居民 Samuel Tunick，他的手机运行了安全加固的 Android 操作系统 GrapheneOS，它内置了一项安全功能，允许用户设置“胁迫密码”，如果输入的不是解锁密码而是“胁迫密码”，那么手机上的数据将会自动清除。本案引发了在入境时美国公民可以援引哪些宪法权利的疑问。美国政府长期坚称，在允许入境前边境处不属于美国领土。数字安全专家 Runa Sandvik 建议，为了避免被当局认为是故意销毁数据，公民可以在入境前先清空数据，入境后再重新下载。

在一群初创公司之后，科技巨头如微软、英伟达、思科、戴尔、IBM、甚至 OpenAI 等数十家公司和组织联署发表公开信，呼吁美国政府不要过早限制开放权重模型，认为此举会扼杀竞争或导致本土创新放缓。公开信称，开放权重模型能促进竞争，确保 AI 技术的好处能广泛共享，避免其掌握在少数人手中，而闭源模型并不意味着绝对安全。没有署名的知名 AI 公司包括了 Google、Anthropic 以及 xAI/SpaceX，其中 SpaceX CEO 马斯克通过社交媒体口头表达了对公开信的支持，而 Anthropic 明显是站在支持限制开放权重模型的一边，该公司正投入 4000 万美元游说特朗普政府加强监管 AI 模型。

IRGC 声称使用巡航导弹袭击并摧毁了亚马逊 AWS 位于巴林的数据中心。亚马逊和美国军方未证实这一说法。如果属实，这是亚马逊巴林数据中心第三次遭到袭击，根据 AWS 自己的公告，巴林数据中心自 2026 年 5 月以来就处于离线状态。

印度政府的网络犯罪协调中心命令 GitHub 移除 Jack Dorsey 的去中心化蓝牙聊天应用 Bitchat。前 Twitter CEO Dorsey 在 X/Twitter 上公布了收到的通知，通知的日期是 7 月 23 日，声称该应用能在网络受限的情况下进行通信，能被反国家分子、恐怖组织、有组织犯罪集团和网络犯罪分子滥用，他们试图逃避合法监测，继续在法律限制下进行通信。印度最近爆发了被称为“蟑螂人民党（Cockroach Janta Party)”的抗议活动。

Google 账号加入了可选的人脸识别登录。用户向 Google 提供自拍视频，随后就可通过人脸识别登录。在人脸验证过程中 Google 可能会要求用户以特定方式转动头部，此举旨在防止深度伪造之类的身份冒用，确保用户当前确实在摄像头前。Google 需要保存用户的自拍视频以将其用于未来的登录验证。搜索巨人表示会对视频进行加密，且仅用于登录不会用于其它用途。如果用户改变主意，可以从 Google 帐户中删除相关视频。

当你的计算机接上 LG 显示器，它会通过 Windows Update 自动安装应用 LG Monitor App Installer，安装之后该应用会不时弹出广告，其中最常见的广告是迈克菲杀毒软件。其它还有 LG 自家产品广告如 LG Switch、LG Calibration Studio、LG Dual Controller 和 LG Channels。此举在消费者中间引发了强烈反应，以至于微软都出来做出回应。微软 Windows 和设备执行副总裁 Pavan Davuluri 称，他们联系了 LG 团队，LG 已经同意立即在其应用中禁止迈克菲广告弹窗。除此之外一切照旧。

近两百家美国科技初创公司组成 Little Tech Association 联盟，联署致函美国总统特朗普、商务部长 Howard Lutnick 等政府高官，呼吁政府不要禁止使用来自中国的开放权重模型。此前由于中国 AI 模型在性能上逼近美国的前沿 AI 模型 Anthropic 的 Fable 5 和 OpenAI 的 GPT-5.6 Sol，美国政府官员表示在考虑封禁中国的 AI 模型。Little Tech Association 的联署信指出，此举会削弱美国初创公司，且并不能阻止开放权重模型扩散。而美国要保持其领先地位需要有世界领先的开放权重模型，确保美国公司能访问其它地方的开放权重模型。

根据发表在 BMJ 上的一项研究，GLP-1 减肥药相比其它糖尿病药物有更高的脱发风险。研究人员指出，快速减肥和减少卡路里摄入是公认的脱发诱因。研究人员称，多数 GLP-1相关的脱发病例都是可逆的，因为毛囊保持完整。研究人员对比了 GLP-1 和其它两种糖尿病药物 SGLT-2 和 DPP-4，结果显示 GLP-1 的脱发风险比 SGLT-2 高 37%，比 DPP-4 高 68%。研究人员称，脱发通常不会造成身体伤害，但可能会产生心理社会影响，如影响自尊、生活质量和治疗依从性。研究人员认为需要开展进一步研究去深入理解之间的关联。

根据发表在《The Lancet Public Health》上的一项研究，1990-2023 年全球出生时的预期寿命从 64.6 岁增至 73.8 岁，而预期健康寿命则从 55.9 岁增至 63.1 岁。预期寿命和健康寿命不同步，预期寿命增幅超过了健康寿命的增幅，这意味着人类不健康时间持续更久。2023 年全球平均有 14.5% 的寿命处于不健康状况，高于 1990 年的 13.6%，不健康时间从 1990 年的 8.8 年增至 2023 年的 10.7 年。不健康的负担在各国和地区之间存在显著差异，通常寿命最长的地方不健康时间也最长。2023 年美国不健康寿命最长为 14 年，其次是澳大利亚（13.9年）和加拿大（13.7年）。

2025 年 3 月六岁女孩美美（化名）在上海新华医院接受了首例针对大脑的基因编辑手术，七天后她死于手术相关的严重免疫反应。美美在四岁时被幼儿园老师注意到存在智力发育问题，老师建议其父母去医院检查下。她之后被诊断患有罕见的 Snijders Blok-Campeau 综合征，全世界至今仅有 237 人确诊。该病是由 CHD3 基因突变引起的，碱基 C 被错误转成 T。该病的症状差异很大，中度到重度患者可能需要尽快手术，但美美的症状是轻度的，是否需要手术专家对此有疑问。她的父母则为此忧心忡忡，通过微信群听说了上海交大的仇子龙教授。仇子龙曾是联署批评贺建奎的胚胎基因编辑手术的中国科学家之一，谴责其无视医学伦理。美美的父母联络了仇教授，之后筹资近 600 万人民币资助仇团队研发针对个人的基因编辑疗法。在猕猴治疗取得突破之后，新华医院伦理委员会批准了首例人体实验。手术使用了比 CRISPR 更精确的碱基编辑技术，通过 AAV9 病毒载体将编辑工具注入脊髓液。三天后美美开始发烧，这是病毒注射后的常见反应，但她的病情没有好转，她一直没有排尿——这是肾脏受到严重损伤的信号，她的血小板也下降到了危险的水平。她的父母事后说，手术的知情同意书淡化了死亡的风险，没有明确指出这些症状会导致死亡。伦理学家称首次人体实验须始终提及死亡。美美送进了 ICU，之后医生通知其父母死亡的消息。医院伦理委员会召开紧急会议，确认死亡与手术相关。仇的团队在《自然》上发表了相关论文，删除了人体实验失败的信息。美美的父母要求仇子龙撤回论文，他们担心这会误导孩子有类似情况的家庭和研究人员。新华医院因监管问题被罚了约 2.4 万元，其他人没有受到惩罚。

人类曾经历过一个语言多样性的“黄金时代”——仅仅在数千年前，全球使用的语言可能多达数万种。然而几个世纪以来，随着少数几种强势语言在全球范围内的传播，这种多样性一直受到侵蚀。如今近半数语言处于濒危状态，且每年至少有四种语言消亡。由于文字记录只能追溯到大约 6000 年前，在人类历史的大部分时间里，语言多样性的长期模式很大程度上仍不为人知。为重建过去 1 万 2000 年间世界语言的历史，研究人员用来自 171 个狩猎-采集社会的民族志数据作为早期人类群体的替代指标研发了一个人口种群模型。该模型假设，随着农业和技术进步使人口数得以增长，越来越多的人开始使用同一种语言，因而导致语言多样性的增长速度滞后于人口增长的速度。研究结果估计，1 万 2000 年前的全球语言数量可能比现在的语言数量少，语言数量随着人口的增长而增加，并在 1000 至 3000 年前达到顶峰；当时全球存在的语言可能有数万种。

7 月 23 日举行的国际数学家大会公布了 2026 年菲尔兹奖得主，四名得主的名字与此前泄漏的名单完全一致。菲尔兹奖被誉为数学界的诺贝尔奖，这是首次有两名中国数学家同时得奖。邓煜在公理化物理的希尔伯特第六问题上做出重大贡献，王虹解决了开放问题三维空间内的挂谷集合猜想，John Pardon（白杰文）解决了 Gromov 的纽结理论问题，Jacob Tsimerman 在 André-Oort 猜想等问题上做出了重大贡献。

Google 因搜索和应用商店服务违反在线竞争法，被欧盟处以总计 8.9 亿欧元的罚款。欧洲委员会表示，Google 违反了 Digital Markets Act（DMA），在搜索结果中优先展示自家服务如购物和酒店而非竞争对手的服务。Google 还通过阻止应用开发者引导消费者前往更便宜的网站或替代应用商店购买而违反了 DMA。Google 因搜索相关违规被罚款 4.6 亿欧元，因应用商店违规被罚款 4.3 亿欧元。欧盟委员会命令 Google 以“公平且无歧视的方式”对待在搜索结果中出现的第三方服务，允许应用开发者在 Google 应用商店之外提供优惠。欧盟委员会指出，Google 已开始测试调整其搜索结果中自家服务的展示方式，称这些变化“在合规方面取得了实质性进展”。

对地球逾 3380 种候鸟的评估报告显示，由于栖息地丧失、有害入侵物种、与建筑物碰撞、狩猎、宠物贸易、有毒杀虫剂等等威胁，逾半数候鸟种群数量出现下降。候鸟不仅仅是非凡的旅行者。它们传粉植物、传播种子、控制农业害虫、在生态系统之间运输养分，并通过吸引观鸟者和生态旅游支持当地经济。它们的数量减少不仅意味着生物多样性的丧失，还预示着支持野生动植物和人类的生态系统正逐步崩溃。研究人员呼吁各国政府采取行动保护候鸟。