OrangeBot.AI Digest — 2026-06-29

Are you in AI answers? Find out and fix it in minutes

Bookkeeping assistant that runs receipt workflows end-to-end

Convert leads into customers with AI conversations

AI to turn vertical UGC into widescreen ads

Lightweight PDF & EPUB reader in your browser

Social media growth agents that build your momentum

Stop re-explaining your project to AI coding agents

Run the best open-weights models in Cline

Automate ad budgets across Google, LinkedIn & Meta ads

A fast, beautiful, and native FTP/SFTP client for macOS

System stats and translation on your Mac's notch

Add WebMCP-native AI chat to any Frontend

"One AI agent across your browser, tools, and messages "

100+ AI models, one interface. ECO friendly.

Your local semantic search app

Video generation models have emerged as a promising paradigm for embodied world simulation. However, both general-domain video generators and robot-specific data fine-tuned models can still produce physically implausible manipulations, including discontinuous motion trajectories and inconsistent robot-object interactions, which limits their reliability as world simulators. Through extensive experiments, we find that such physical instability mainly arises from two factors: deformation of moving objects and implausible spatio-temporal correlations among interacting entities, particularly during contact. Building on this observation, we propose PhysisForcing, a scalable training framework that strengthens physical consistency by focusing supervision on physics-informative regions through joint optimization of pixel-level and semantic-level features. The framework consists of a pixel-level trajectory alignment loss, which supervises DiT features using reference point trajectories, and a semantic-level relational alignment loss, which aligns DiT features with inter-region relations extracted from a frozen video understanding encoder. Extensive experiments on R-Bench, PAI-Bench, and EZS-Bench show that PhysisForcing consistently improves embodied video generation over strong baselines, improving the Wan2.2-I2V-A14B and Cosmos3-Nano base models on R-Bench by 22.3\% and 9.2\% (7.1\% and 3.7\% over vanilla finetuning), with the Cosmos3-Nano variant attaining the best overall score. Beyond generation, as a world model under the WorldArena action-planner protocol it raises the closed-loop success rate from 16.0\% to 24.0\% and further improves downstream policy success, indicating that physically aligned video models yield stronger representations for robotic manipulation.

We study whether we can learn novel manipulation skills from human actions to a bi-manual robot with parallel grippers. Human action data is cheap, abundant, and diverse, making it one of the most promising resources for scaling up robot learning. Yet transferring skills from humans to robots remains hard: most prior work treats humans as just another bi-manual 6DoF embodiment, where hand-pose estimates are noisy and the contact patterns of human fingers differ fundamentally from those of a parallel gripper. We argue that learning rotation-inclusive action signals from human data is therefore sub-optimal, and instead propose a bridging action representation: the relative wrist translation within the initial head-camera frame, an action space shared by humans and robots. To handle the potential absence of certain action components in different embodiments, we build a π_0-like vision-language-action model with interleaved action tokens and attention masking. On a suite of novel bi-manual manipulation tasks, our bridging action transfers human manipulation knowledge to robots far more effectively than noisy 6DoF human actions and scales with the amount of human data.

We introduce an axiomatic evaluation framework for latent thought representations in LLMs, comprising metrics that are independent of downstream benchmark scores and reveal representational failures that benchmark accuracy masks. Existing evaluations conflate representation quality with model capacity. Therefore, failures cannot be attributed to the representation rather than to the model that processes it. We formalize four functional axioms (Causality, Minimality, Separability, and Stability) and define a quantitative measure for each, computed directly on the representation independently of downstream accuracy. We audit open-weight LLMs across 23 reasoning tasks (e.g., Spatial Reasoning, Factual QA). We find that no candidate satisfies all four axioms simultaneously, that the representations distinguish task type reliably but cannot distinguish between two questions within the same task, and that the representations encode little information beyond what is already present in the input embedding. The failure is consistent across dense, reasoning-distilled, and RL-trained model families, indicating that the gap is structural rather than a property of model size or training procedure.

We present Qwen-Image-2.0-RL, a post-training pipeline that applies reinforcement learning from human feedback (RLHF) and on-policy distillation (OPD) to improve both the visual quality and instruction-following capability of the Qwen-Image-2.0 diffusion model. To provide reliable reward signals, we construct task-specific composite reward models by fine-tuning vision-language models with a pointwise scoring paradigm and chain-of-thought reasoning. For text-to-image generation, the reward models cover alignment, aesthetics, and portrait fidelity dimensions. For image editing tasks, the reward system addresses instruction-following accuracy and face identity preservation. Building on this reward system, we develop a scalable GRPO-based RL training framework, incorporating a hybrid classifier-free guidance (CFG) strategy to preserve pre-trained knowledge, prompt curation via intra-group reward range filtering, and per-category reward weight calibration. To merge the task-specialized RL policies for T2I and editing, we propose on-policy distillation as the final training stage, which consolidates multiple teachers into a single student model through trajectory-level velocity matching. Extensive evaluation shows that Qwen-Image-2.0-RL achieves 57.84 overall score on Qwen-Image-Bench (+2.61 over the base model), Elo ratings of 1193 in text-to-image arena (+78) and 1349 in image edit arena (+93), demonstrating consistent gains in aesthetic quality, prompt adherence, and editing accuracy.

Language models (LMs) represent tokens using embedding matrices that scale linearly with the vocabulary size. To constrain the parameter footprint, prior work proposes hashing many tokens into a single vector within encoder-only models. While this offers parameter efficiency, many-to-one collisions prevent its use in causal LMs. In this paper, we propose MultiHashFormer, a new framework that allows hash-based autoregression. Each token is represented as a unique hash signature, a short sequence of discrete hash IDs, generated by multiple independent hash functions. A Hash Encoder compresses this signature into a single latent vector for processing by a Transformer decoder. Then, a Hash Decoder generates the hash signature of the next token, which is then mapped back to text. We evaluate our approach at the 100M, 1B and 3B parameter scales, demonstrating that MultiHashFormer consistently outperforms standard Transformer LMs across multiple benchmarks. Furthermore, we show that our model handles multilingual vocabulary expansion with a constant parameter footprint without any modifications.

Vision-language models (VLMs) are increasingly deployed in consumer, medical, financial, and enterprise applications. This broad deployment expands the safety surface: risks can arise from multimodal question answering, assistant responses, and cross-modal composition, while moderation policies may vary across products, regions, and deployment stages. Most existing guardrails either rely on fixed taxonomies or target only a narrow set of interaction settings, which limits their adaptability when safety rules change at deployment time. We present SingGuard, a policy-adaptive multimodal guardrail model family for safety assessment in multimodal conversations. SingGuard treats the active policy as a runtime input: given natural-language rules, it checks the target content against the active policy rule by rule and predicts both the safety label and the triggered rule. To balance efficiency and interpretability, SingGuard supports fast, hybrid, and slow inference regimes along a fast-to-slow reasoning spectrum, ranging from direct safety judgments to policy-grounded deliberation. We further optimize this behavior with fast--slow decoupled reinforcement learning. We also introduce SingGuard-Bench, a multimodal guardrail benchmark with 56{,}340 examples spanning 80+ fine-grained risk types across multimodal QA, adversarial attack, and dynamic-rule evaluation settings, including cross-modal joint-risk cases where each modality is harmless in isolation but their composition implies unsafe intent. Across six benchmark families (35 datasets), SingGuard achieves state-of-the-art average F1 in every family. Dynamic-rule evaluation further shows improved policy-following accuracy from 0.6465 to 0.7415 under runtime policy shifts. Our code is available at https://github.com/inclusionAI/Sing-Guard.

Text detoxification, the automated detection and mitigation of abusive and harmful content, is essential for ensuring the safety of online communities and protecting users. However, low resource languages such as Tatar have received little research attention. In this paper we present Tatoxa, a novel state-of-the-art system for text detoxification in the Tatar language. Comparative experiments show that the proposed approach outperforms existing open source and proprietary commercial LLMs on key quality metrics. We also introduce a new dataset for text detoxification in Tatar, designed for fine tuning and evaluation in low resource settings. Finally, cross lingual transfer experiments indicate that transfer from other languages, including the culturally close Russian, performs significantly worse than training on native Tatar data even when a large Russian corpus is available.

Multi-agent systems (MAS) built on large language models (LLMs) provide a promising framework for solving complex tasks through role specialization and structured interaction. However, their performance is often limited by miscoordination and, more fundamentally, the lack of fine-grained credit assignment across agents. Existing approaches typically rely on coarse-grained feedback, making it difficult to identify which agents or interaction steps are responsible for errors. We propose Gradient-Based Connections (GBC), an approach for fine-grained attribution and optimization of multi-agent systems. GBC models a MAS as a computational graph and introduces gradient-based connection weights to quantify the influence of each agent's output on downstream agents at the token level. By constructing an attribution graph and propagating task-specific loss signals backward, our method enables precise identification of error sources and targeted prompt optimization. We further develop AgentChord, an efficient implementation that leverages prefix-based gradient computation. Experiments on MultiWOZ and τ-bench show that GBC improves multi-agent performance and outperforms strong single-agent and multi-agent baselines, and higher attribution quality is associated with greater optimization effectiveness. Code is available at: https://github.com/yxc-cyber/AgentChord.

Knowledge-based Visual Question Answering (KB-VQA) requires models to combine image understanding with external knowledge. Most prior methods use a fixed retrieve-then-generate pipeline with a pre-selected retriever and a static top-k setting, which is not adaptive during reasoning. We propose ProMSA, a progressive multimodal search agent for KB-VQA. Given an image-question pair, the agent iteratively chooses image search, text search, or stop, under explicit tool-call budgets and with deduplication to avoid redundant retrieval. For training, we first use rejection-sampling SFT to learn valid tool-use formats, then optimize the agent with TN-GSPO, a sequence-level RL objective that normalizes updates by both generation length and tool-interaction depth. Experiments on E-VQA and InfoSeek show consistent gains over strong RAG and agent baselines, and improved retrieval and end-to-end accuracy. The code is available at https://github.com/DingWu1021/Promsa.

Training and evaluating robot policies in the real world is costly and difficult to scale. We introduce SimFoundry, a modular and automated system for zero-shot real-to-sim scene construction from a video. SimFoundry generates sim-ready digital twins and supports object, scene, and task editing, enabling the automated generation of diverse digital cousins: affordance-preserving variations of reconstructed real-world scenes. Policies trained on SimFoundry data transfer zero-shot to challenging real tasks involving multi-step manipulation, articulated object interaction, and bimanual interaction, and its digital cousins (variations of the original scene, objects, and tasks) facilitate generalization to new real-world conditions. Across 7 manipulation tasks and 5 policy architectures, SimFoundry simulation evaluations strongly predict real-world performance, with mean Pearson correlation 0.911 and mean maximum ranking violation 0.018. When evaluating sim-trained policies zero-shot in the real world, policies trained with object, scene, and task cousins in simulation show average task success rate improvements of 17%, 21%, and 40%, respectively. Additional details at https://research.nvidia.com/labs/gear/simfoundry/ .

Pixel-space continuous-token autoregressive (AR) generation directly models images as sequences of raw pixel patches, avoiding discrete tokenization or a separately pretrained tokenizer. However, it faces coupled challenges: high-dimensional patch generation causes large single-step errors, and teacher-forced training creates a train--inference gap that makes these errors accumulate across AR steps. Existing fixes such as x-prediction and input noise injection only partially mitigate these issues. Exact rollout training better matches inference-time conditions, but is impractical due to prohibitively slow sequential sampling. We propose Parallel Rollout Approximation (PRA), a scalable framework that addresses both challenges jointly. PRA generates low-dimensional intermediate states instead of high-dimensional pixel patches, then maps them back to pixel-space tokens with a pixel decoder, preserving a pixel-in, pixel-out AR interface. It also constructs inference-like pixel inputs through the same intermediate-state-to-pixel path used at inference, independently across positions, approximating the pixel-feedback interface encountered during inference-time rollout while retaining parallel teacher-forced training. On class-conditional ImageNet-1K generation at 256times256 resolution, PRA-S with 135M parameters achieves an FID of 2.58, surpassing the previous billion-scale pixel-space AR result of 3.60. Scaling to PRA-L with 511M parameters further improves FID to 1.94, establishing a new state of the art among pixel-space AR models. Beyond generation, PRA achieves higher ImageNet classification probing accuracy than other AR and diffusion baselines, suggesting its potential for unified pixel-space image generation and understanding.

Voice agents face a fundamental tension: the reasoning, retrieval, and tool use that make foundation models capable are iterative and slow, while conversational interaction demands responses on a millisecond timescale. Smaller, real-time models meet the latency bar but cannot match foundation models on complex tasks, leaving current voice agents to trade away either responsiveness or capability. We introduce conversational infill, where a small talker model both immediately generates contextually grounded responses to hide the latency of an external reasoner model and fluently integrates streamed reasoner knowledge into its responses during inference. We curate a 290,571-example synthetic dataset spanning six domains and demonstrate that this task is learnable across seven widely used small language models ranging from 135M to 1.7B parameters. Our system implementation, ConvFill, sustains millisecond-level time-to-first-response while closing the accuracy gap to within 6.3% of the corresponding frontier reasoner performance. In a live user study (n=18) with talker deployments running on an Apple M2 SoC, participants rank ConvFill on par with frontier models overall, prefer it for retrieval-heavy tasks, and rate it significantly more responsive. These results show that conversational infill unlocks a new point on the latency-capability Pareto frontier, offering a practical path toward voice agents that are both responsive and highly capable. Code, models, and datasets are available at https://github.com/vysri/conversational-infill.

I describe my solution to the LeHome Challenge 2026, an ICRA 2026 competition on bimanual garment folding. The system placed 1st of 62 teams in the online (simulation) round and 2nd in the real-world final. It improves a vision-language-action (VLA) policy with a reinforcement-learning loop. The policy is its own value function: the same network that predicts actions also predicts success, progress, and a few task-relevant future quantities, and those predictions drive advantage estimation, live failure detection, and candidate selection. The work mostly recombines existing RL ideas with engineering and optimization contributions that can be used together as one recipe or individually: AWR + RECAP combined for flow-matching VLA; an asynchronous distributed training / rollout pipeline through HuggingFace Hub; inference-time hyperparameters optimization via Thompson sampling; a sim-to-real recipe with camera-alignment tooling, heavy augmentation and DAgger-like HIL data collection.

Web-agent benchmarks overwhelmingly measure depth -- pinning one obscure answer behind a chain of constraints -- while breadth, exhaustively enumerating a closed set and filling each item's attributes, is barely evaluated, especially outside English. Breadth is also hard to build: certifying that a gold set is complete and every cell correct is far costlier than checking a single answer. I introduce Ko-WideSearch, a Korean breadth-search benchmark built by an automated synthesize-and-verify pipeline. Each task names a set-parent entity -- a TV season, a dynasty, a league, an administrative region, an election -- and asks for its full membership plus a per-item attribute table, graded by Item-, Column-, and Row-F1. It spans 228 tables over 190 entities and sixteen categories across three difficulty tiers, set by two structural knobs I dial independently -- table width and a 2-D composite key -- so cross-product membership climbs from 0\% to 100\% across the tiers. A single normalization-aware comparator is shared between gold construction and grading, so stable date and count columns are not over-dropped on formatting alone. Across twenty web agents, the failure is consistent: agents recover the set but not the rows (e.g.\ Item-F1 92.8 against Row-F1 53.7), accuracy falls steadily as the knobs harden, and neither more search nor more spend closes the gap. Broken down by cell, the hard part is finding the right value, not formatting it: open-ended free-text cells fail most, while cells with a standard answer such as a date or a name usually come out right.

Vision-Language-Action (VLA) models can generalize across diverse manipulation tasks, but their imitation-learning-based policies remain brittle in precise physical interactions due to compounding execution errors; Can a reinforcement learning policy trained purely in simulation improve the robustness of real-world VLAs zero-shot? Residual RL, which learns a corrective policy on top of a frozen VLA, offers a natural framework, but existing approaches face a fundamental sim-to-real dilemma: privileged-state methods require lossy distillation for deployment; image-based methods suffer from the visual domain gap; and real-world RL is costly and unsafe. We propose an object-centric residual RL framework that refines VLA actions using object poses, enabling a compact observation space that transfers consistently between simulation and reality. To align the two domains, we additionally replay the same teleoperation demonstrations in simulation to train a sim counterpart of the real-world VLA. The residual RL policy is trained only in simulation with pose noise injection and dropout, and transfers zero-shot to the real robot. Across five manipulation tasks on a real Franka Research 3 (FR3) robot, our method improves the success rate from 42% to 76% zero-shot, and the improved rollouts can be further reused to retrain the base VLA for self-improvement without additional teleoperation. Project page: https://www.microsoft.com/en-us/research/articles/object-centric-residual-rl/

Murray Stassen / Music Business Worldwide : Tidal adopts an AI policy that blocks wholly AI-generated music from earning royalties and removes AI-generated music that impersonates artists —  TIDAL is introducing an AI policy that will automatically tag wholly AI-generated music in its app and block it from earning royalties.

Jaspreet Singh / Reuters : Digital Realty says it plans to acquire a majority stake in three fully leased Northern Virginia data centers from Blackstone-managed funds in a $7.8B deal —  Digital Realty (DLR.N) said on Monday it would acquire a majority stake in three fully leased Northern Virginia data centers …

Debby Wu / Bloomberg : Source: Taiwanese authorities raided Super Micro's Taiwan office as part of a probe into the alleged smuggling of Nvidia chips to China; SMCI closed down 8.1% —  Super Micro Computer Inc.'s offices in Taiwan were raided by government authorities on Monday, widening an investigation …

Bloomberg : As a new law bars DOD from working with companies whose lobbyists also represent blacklisted entities, DC lobbying firms drop companies like Alibaba and Tencent —  Washington's most powerful lobbying firms are rushing to cut ties with Chinese tech giants including Alibaba Group Holding Ltd …

Sean O'Kane / TechCrunch : Waymo and Uber quietly ended their partnership in Phoenix in May; Uber says it is readying the launch of a separate autonomous vehicle partnership in the city —  Waymo robotaxis are no longer available on Uber's ride-hail app in Phoenix, Arizona, ending a nearly three-year partnership in the city …

Reuters : Sources: component and supplier lists, and photos of iPhone 18 Pro models, are among files a ransomware group stole from Apple's Indian supplier Tata —  Sensitive lists of components and suppliers, and photos of Apple's upcoming iPhone 18 Pro models are part of files posted on the dark web …

Mike Wheatley / SiliconANGLE : Baz releases Baz Planner, which uses four specialized AI agents to analyze code at the planning stage, and extends its seed funding by $9M to $17M —  Agentic coding startup Baz Technologies Inc. said today it's launching a new platform that sits between developers and the code bases they're working …

Christine Mui / Politico : California strikes a deal with Anthropic to expand the use of Claude products across state agencies and local governments at a 50% discount —  “A lot of departments are going to switch their usage to this contract, and that's very much our intent,” Chris Given, California's chief information officer …

Joanna Glasner / Crunchbase News : In Q2, there were thirteen $1B+ US venture-backed startup exits, either through acquisition or IPO, the most exits since the 2021 market peak —  Startup exits valued at $1 billion or more are now more numerous than at any point since the 2021 market peak, Crunchbase data shows.

Laura Kreutzer / Wall Street Journal : Quantifind, whose AI products help banks combat financial crimes such as money laundering, raised $200M led by Summit Partners —  Its Graphyte platform aims to help banks and other organizations ferret out money-laundering and other financial crime, the company says

Catherine Perloff / The Information : Sources: Amazon is weighing using OpenAI's and its own Nova models to cut costs after Anthropic raised prices for using its models in Amazon products —  As Anthropic grows to become a leading AI model used by enterprises, it's been flexing its power with customers—including one of its most important early backers, Amazon.

TechCrunch : SCOTUS limits the law enforcement use of “geofence” warrants, saying people have “a reasonable expectation of privacy” in their cell-phone location data —  Zack Whittaker Lorenzo Franceschi-Bicchierai  —  The U.S. Supreme Court on Monday limited the law enforcement use of …

Marcus Mendes / 9to5Mac : WhatsApp rolls out username reservations globally, allowing users to claim a unique name before the privacy-focused feature launches later this year —  Following years of anticipation and months of preparation, WhatsApp today began rolling out username reservations worldwide …

Naga Avan-Nomayo / The Block : Changpeng Zhao says Binance's MiCA application in Greece was fully compliant and close to approval by at least one EU regulator, but “other forces” opposed it —  Quick Take  — Binance founder Changpeng Zhao said the exchange's MiCA application in Greece was fully compliant …

Chris Metinko / Axios : Straiker, which develops tech for securing enterprise AI agents, raised a $64M Series A, bringing its total funding to $85M —  Straiker, which secures AI agents, raised a $64 million Series A, CEO Ankur Shah tells Axios Pro exclusively. … - The platform can discover agents across …

世界各地的学生正使用 AI 驱动的智能眼镜在考试中作弊，而各国也开始加强对考生眼镜的检查。本月初的中国高考有逾千万学生参加，政府要求检查所有学生的眼镜。英格兰考试监管机构负责人警告，AI 眼镜和智能耳机等设备可能会加剧考试作弊现象。韩国报告了首起利用 AI 眼镜作弊的案例。专家担心相关个案可能预示着智能眼镜作弊正成为更为普遍的问题。研究过智能眼镜应用的 Thomas Corbin 说，“如果我们看到有些案例被报道，那么肯定还有更多案例没有被报道。”

福特公司是众多拥抱 AI 的公司之一，该公司在质检等运营环节部署了 AI。福特 CEO Jim Farley 去年 6 月表示，AI 将让大量白领落在后面。首席运营官 Kumar Galhotra 去年 10 月表示公司在整个工业系统部署 AI。福特汽车硬件工程副总裁 Charles Poon 如今表示 AI 驱动的质量检查未能达到预期。自动化工具缺少资深技术人员的训练和知识，他表示资深技术人员在公司能利用其知识改进技术前已离开了公司。这些员工已重新聘用，负责训练系统并指导年轻员工。

超新星爆炸将碳、氮、氧等组成生命的元素散布到宇宙中。这些粒子成为宇宙射线在宇宙中跋涉数百万年后，如今正不断撞击地球大气。搭载于国际太空站上的阿尔法磁谱仪（AMS-02）所收集数据的新发现为我们揭开了这些粒子的奥秘。AMS-02 自 2011 年启用以来，已累积收集超过 230 兆个宇宙射线事件。研究团队深入分析了周期表上介于氦与铁之间的 20 种元素，惊人地发现这些宇宙射线并非随机分布，而是可以归纳为四个独特的类别：两种「主要宇宙射线」（源自深空未受破坏的原始射线）及两种「次要宇宙射线」（在旅途中与星际气体碰撞产生的混合粒子）。粒子中质子数的奇偶性（奇数或偶数）与其演化路径密切相关，显示恒星内部的元素合成机制深刻影响了它们在太空中的行为。新发现直接挑战了现行的宇宙射线模型。目前科学家尚无法完全解释这些观测结果，这意味着现有理论可能遗漏了某些关键物理机制。

波音 747 曾是美国实力、发明、进步和平民主义的象征。如今它却成为所有这些价值观衰落的缩影。从 1970 年第一架 747 投入使用，到 2023 年停产该机型，波音公司共生产了 1574 架飞机，包括两架至今仍在服役的空军一号。类似 20 世纪的大多数科技创新，747 项目也是由军方推动的。1960 年代初，波音应政府要求设计一款大型军用运输机。洛克希德赢得了竞标生产了 C-5 银河运输机。波音失利之后放手让工程师在此基础上研发出最大的商用飞机。747 的用途远不止客运和货运。NASA 曾利用一架改装过的 747 将航天飞机运送到肯尼迪航天中心。美国的这两大象征似乎预示着 20 世纪的进步永无止境。但它们终究会落幕。航天飞机项目于 2011 年终止，而 747 飞机也逐渐从天空消失。如今亲眼目睹 747 变得越来越难，尤其是在美国。Atlas Air 和 Kalitta 仍在运营部分波音 747。汉莎航空运营着波音 747 客运航班最多的航线，大韩航空仍在运营 747 的国际航线，中国、伊朗和俄罗斯用它执飞类似巴士的国内航线。

Top500 上周公布了最新的超算榜单，深圳国家超算中心的灵晟首次亮相即登顶榜单。灵晟超算在 Linpack 测试中比排名第二的美国劳伦斯利弗莫尔国家实验室 El Capitan 超算快 22%，在 HPCG 测试中快 26%。它是首个仅靠 CPU 实现持续双精度浮点性能逾 2 Exaflops 的超算系统，美国的超算使用了 GPU 加速器。据 Chips and Cheese 根据相关幻灯片和相关 arXiv 论文报道，灵晟使用的 LX2 CPU 是基于 ARMv9.2 架构，支持 Scalable Matrix Extension(SME)指令集。相比下日本 ARM 超算富岳(Fugaku)是基于 ARMv8 架构，在今天已经相当老了。LX2 的每个核心都有 32 KB 的 L1 指令缓存和 32 KB 的 L1 数据缓存。芯片由两个计算模块（die）组成，每个模块包含四个 40 核心簇。每个簇有 2 个核心被禁用，因此每个簇有 38 个活跃核心，每个模块有 152 个活跃核心。每个簇配备 28.5 MB 的 L2 缓存，每个模块有 114 MB 的 L2 缓存，整个 LX2 封装有 304 个活跃核心和 228 MB 的总 L2 缓存。304 个核心以 1.55 GHz 运行，每个 LX2 CPU 提供 60.3 TFLOP/s 的 FP64 计算性能，功耗为 690 瓦。LX2 配备了八个“高带宽内存”，带宽为 4 TB/s（另一篇报道称 4 TB/s per chiplet，8 TB/s per socket）。所谓的高带宽内存可能不是 HBM。灵晟超算系统包含了逾 22,000 个节点和 1379 万个 CPU 核心。

Gartner 预测两年内开发者的 AI token 使用费用将超过其薪水。预测将开发者的薪水设定为每月 2000 美元，因此这并不意味着所有地区的 AI token 使用费用将会达到或超过开发者薪水，美国的开发者年薪通常高达六位数。Gartner 高级首席分析师 Nitish Tyagi 指出，在极端情况下美国开发者的 AI token 使用费用也可能会超过其薪水，有些开发者每月的 AI 支出会超过数万美元。美国部分科技公司也已经开始要求其员工控制 AI token 的使用。Tyagi 称，企业必须监管和控制 AI token 的使用，否者 AI 工具费用的增长速度可能会超过带来的生产力提升。

中国近四年来人口持续下降，超重肥胖率却加速上升，而且年轻化趋势明显，这意味着未来健康劳动力可能减少。国家卫健委、中国营养学会的《中国肥胖预防和控制蓝皮书》等公开数据显示，1992 年全国成人超重肥胖率为 27.2%，2002 年升至 29.9%，2005 年则加速增长到 42.3%，2020 年为 50.7%，2023 年则达到 57%。研究预测，若超重肥胖趋势得不到遏制，2030 年中国成人、儿童超重肥胖率将分别达 70.5% 和 31.8%。《柳叶刀》报告显示，中国 25 岁及以上的成年超重和肥胖患者人数早在 2021 年已达 4.02 亿。

在 Cox Communications, Inc. v. Sony Music Entertainment 一案中，索尼等唱片公司指控 Cox 在其用户的侵权活动中是共谋犯，需要承担侵权责任。今年 3 月美国最高法院站在了 Cox 这边，以 9 比 0 裁定 Cox 不用对其用户的行为承担共同责任。本案就间接侵权设立了新标准，从此之后原告必须证明被告故意诱导他人实施非法行为。《纽约时报》根据最高法的裁决修改了诉讼，指控微软积极鼓励 OpenAI 窃取其受版权保护的作品。诉讼称，微软的新超算是专门帮助 OpenAI 侵权而定制的，其目的就是在未经许可的情况下训练 AI 处理受版权保护的作品，该系统特别授予纽约时报文章更高的权重。《纽约时报》指控，通过建造新超算，微软不仅帮助 OpenAI 选择侵权作品，还提供了一种未经许可获取受版权保护作品的手段。

Linus Torvalds 在内核邮件列表上宣布释出 Linux 7.2-rc1。主要变化包括：Cache Aware Scheduling，性能优化、修复 Linux 7.1 的新 NTFS 驱动，完全移除 strncpy API，新 ARCTIC Fan Controller 驱动，AMD ISP4 驱动，初步支持 AMDGPU HDMI 2.1 FRL，等等。

AI 短剧近月来成了中国影视业炙手可热的风口。在这场变革中，创作者借助低门槛的 AI 生成技术，快速制作具画面、对白和音效的短剧，大幅压缩了传统制作所需的人力和时间。古装题材 AI 短剧《霍去病》等爆款作品的出现，让不少业者看到一片商业蓝海。据媒体报道，中国今年注册的 AI 短剧企业超过 2100 家。然而爆款剧终究是少数——截至今年 2 月上线的超过 12 万部 AI 短剧中，播放量破亿的低于 150 部，爆款率约千分之一。西红柿影业董事长陈健受访时坦言：“对于承制 AI 短剧的公司，扣除算力和员工成本，利润没剩多少。钱基本都给头部大厂赚走，我们更像是血汗工厂。”陈健说：“有时发出 100 次指令，只有一次能换来想要的效果……原以为能赚钱，后来发现是亏的，因为修改太多了，推高了人力和算力成本。”AI 虽提升了生产力，却未必能改变短剧业内卷和利益分配的既有逻辑。许多业者发现，真正赚钱的是平台公司和少数头部短剧企业，多数中小型承制公司则只能艰难求生。

币安已告知欧盟客户，由于该公司无法取得在欧盟运营所需的牌照，将从下周起停止向他们提供服务。这对这家全球最大的加密货币交易所来说，是一次重大挫折。自 7 月 1 日起，所有在欧盟经营的加密货币公司，都必须依据欧盟《加密资产市场监管条例》(Markets in Crypto-Assets Regulation, MiCA)持牌运营，否则将面临处罚。币安此前在希腊申请一张可覆盖整个欧盟的牌照，但上周遭到拒绝，距离最后期限生效已不足两周。该交易所目前计划转向法国申请牌照，此前曾在法国就牌照问题进行过磋商。

继哥伦比亚大学 Dieter Egli 团队之后，第二个科研团队报告称，利用高精度基因编辑技术改造了人类胚胎 DNA。多名科研人员表示，该实验的成功，让关于胚胎基因编辑的伦理讨论变得更加紧迫。这一次英国剑桥大学的 Kathy Niakan 带领研究人员使用单碱基编辑技术敲除了编码 NANOG 蛋白的基因。实验所用的精子、卵子与胚胎均来自科研捐献，胚胎仅培养约 1 周时间。经过编辑的胚胎无法形成完整外胚层，也就是未来发育成人体各类组织的细胞团；但胚胎仍能分化出胎盘、卵黄囊等胚胎附属支撑结构的前体细胞。

1918 年诺贝尔物理学家得主马克斯普朗克（Max Planck）的两篇论文在 2011 年被撤下。两篇论文于 1940 年代发表在德国期刊《Naturwissenschaften》上，如今隶属于 Springer Nature 集团，《Naturwissenschaften》撤稿的理由是侵犯版权。马克斯普朗克的做法是一稿多发，在以前很常见，但如今被认为“自我剽窃”会引发版权纠纷。魁北克大学物理史学家 Yves Gingras 和科学史学家 Mahdi Khelfaoui 对此做法提出批评。尤其滑稽的是 Springer Nature 以空白 PDF 的形式撤稿，而该空白 PDF 仍然以 39.95 美元的价格出售访问权限。他们怀疑自然集团是使用算法自动撤稿的。

2014 年世界气象组织 (WMO) 协调组织了一场活动，其中包括预测 2050 年的天气。法国主持人 ïvelyne Dhéliat 的预测是到 2050 年该国最高气温将达到 43 摄氏度。本周法国的热浪将这一预测提前 24 年实现，且超过了预测。法国周三 43 个地点有 19 个地点超过了对 2050 年的预测，最高气温高达 44.3 摄氏度。法国 1947 年以来的热浪半数发生在 2010 年之后。法国气象局表示，到 2100 年热浪可能会持续两个月。随着地球暖化，极端高温将变得越来越常见。

韩国计划训练所有军人像使用枪支一样熟练操作无人机。国防部长官安圭伯表示，目标是通过训练士兵像使用“第二件个人武器”一样使用无人机，让无人机成为所有部队的“通用作战工具”。韩国还计划为军事单位配备更多廉价且可消耗的无人机，用于侦察和打击任务，部署更多反无人机激光武器和微波武器。很多国家都在效仿乌克兰训练和装备无人机，韩国并非个例。乌克兰利用无人机和军用机器人弥补其在战场上与俄罗斯庞大军力对抗的劣势。韩国国防部今年将首先向军人提供 1.1 万架训练无人机，目标是到 2029 年部署 6 万架无人机。出于安全考虑，韩国军方希望采购 100% 国产零部件的无人机。