Weekly Digest — 2026-W09

FULL Augment Code, Claude Code, Cluely, CodeBuddy, Comet, Cursor, Devin AI, Junie, Kiro, Leap.new, Lovable, Manus, NotionAI, Orchids.app, Perplexity, Poke, Qoder, Replit, Same.dev, Trae, Traycer AI, VSCode Agent, Warp.dev, Windsurf, Xcode, Z.ai Code, Dia & v0. (And other Open Sourced) System Prompts, Internal Tools & AI Models

Financial data platform for analysts, quants and AI agents.

A comprehensive collection of Agent Skills for context engineering, multi-agent architectures, and production agent systems. Use when building, optimizing, or debugging agent systems that require effective context management.

a.k.a. Awesome ChatGPT Prompts. Share, discover, and collect prompts from the community. Free and open source — self-host for your organization with complete privacy.

A latent text-to-image diffusion model

The agentic meeting stack with free prep, notes, and chat

Your voice in, polished text out — in seconds

Everything your agent needs to run reliably

Test your APIs in plain English, straight from the terminal

Openclaw in the cloud

Switch from ChatGPT to Claude with import memory feature

Training stability remains a central challenge in reinforcement learning (RL) for large language models (LLMs). Policy staleness, asynchronous training, and mismatches between training and inference engines all cause the behavior policy to diverge from the current policy, risking training collapse. Importance sampling provides a principled correction for this distribution shift but suffers from high variance; existing remedies such as token-level clipping and sequence-level normalization lack a unified theoretical foundation. We propose Variational sEquence-level Soft Policy Optimization (VESPO). By incorporating variance reduction into a variational formulation over proposal distributions, VESPO derives a closed-form reshaping kernel that operates directly on sequence-level importance weights without length normalization. Experiments on mathematical reasoning benchmarks show that VESPO maintains stable training under staleness ratios up to 64x and fully asynchronous execution, and delivers consistent gains across both dense and Mixture-of-Experts models. Code is available at https://github.com/FloyedShen/VESPO

Recent advancements in large reasoning models (LRMs) have greatly improved their capabilities on complex reasoning tasks through Long Chains of Thought (CoTs). However, this approach often results in substantial redundancy, impairing computational efficiency and causing significant delays in real-time applications. Recent studies show that longer reasoning chains are frequently uncorrelated with correctness and can even be detrimental to accuracy. In a further in-depth analysis of this phenomenon, we surprisingly uncover and empirically verify that LRMs implicitly know the appropriate time to stop thinking, while this capability is obscured by current sampling paradigms. Motivated by this, we introduce SAGE (Self-Aware Guided Efficient Reasoning), a novel sampling paradigm that unleashes this efficient reasoning potential. Furthermore, integrating SAGE as mixed sampling into group-based reinforcement learning (SAGE-RL) enables SAGE-RL to effectively incorporate SAGE-discovered efficient reasoning patterns into standard pass@1 inference, markedly enhancing both the reasoning accuracy and efficiency of LRMs across multiple challenging mathematical benchmarks.

Extended reality (XR) demands generative models that respond to users' tracked real-world motion, yet current video world models accept only coarse control signals such as text or keyboard input, limiting their utility for embodied interaction. We introduce a human-centric video world model that is conditioned on both tracked head pose and joint-level hand poses. For this purpose, we evaluate existing diffusion transformer conditioning strategies and propose an effective mechanism for 3D head and hand control, enabling dexterous hand--object interactions. We train a bidirectional video diffusion model teacher using this strategy and distill it into a causal, interactive system that generates egocentric virtual environments. We evaluate this generated reality system with human subjects and demonstrate improved task performance as well as a significantly higher level of perceived amount of control over the performed actions compared with relevant baselines.

Visual analogy learning enables image manipulation through demonstration rather than textual description, allowing users to specify complex transformations difficult to articulate in words. Given a triplet {a, a', b}, the goal is to generate b' such that a : a' :: b : b'. Recent methods adapt text-to-image models to this task using a single Low-Rank Adaptation (LoRA) module, but they face a fundamental limitation: attempting to capture the diverse space of visual transformations within a fixed adaptation module constrains generalization capabilities. Inspired by recent work showing that LoRAs in constrained domains span meaningful, interpolatable semantic spaces, we propose LoRWeB, a novel approach that specializes the model for each analogy task at inference time through dynamic composition of learned transformation primitives, informally, choosing a point in a "space of LoRAs". We introduce two key components: (1) a learnable basis of LoRA modules, to span the space of different visual transformations, and (2) a lightweight encoder that dynamically selects and weighs these basis LoRAs based on the input analogy pair. Comprehensive evaluations demonstrate our approach achieves state-of-the-art performance and significantly improves generalization to unseen visual transformations. Our findings suggest that LoRA basis decompositions are a promising direction for flexible visual manipulation. Code and data are in https://research.nvidia.com/labs/par/lorweb

Decoding sits between a language model and everything we do with it, yet it is still treated as a heuristic knob-tuning exercise. We argue decoding should be understood as a principled optimisation layer: at each token, we solve a regularised problem over the probability simplex that trades off model score against structural preferences and constraints. This single template recovers greedy decoding, Softmax sampling, Top-K, Top-P, and Sparsemax-style sparsity as special cases, and explains their common structure through optimality conditions. More importantly, the framework makes it easy to invent new decoders without folklore. We demonstrate this by designing Best-of-K (BoK), a KL-anchored coverage objective aimed at multi-sample pipelines (self-consistency, reranking, verifier selection). BoK targets the probability of covering good alternatives within a fixed K-sample budget and improves empirical performance. We show that such samples can improve accuracy by, for example, +18.6% for Qwen2.5-Math-7B on MATH500 at high sampling temperatures.

Humans can rearrange objects in cluttered environments using egocentric perception, navigating occlusions without global coordinates. Inspired by this capability, we study long-horizon multi-object non-prehensile rearrangement for mobile robots using a single egocentric camera. We introduce EgoPush, a policy learning framework that enables egocentric, perception-driven rearrangement without relying on explicit global state estimation that often fails in dynamic scenes. EgoPush designs an object-centric latent space to encode relative spatial relations among objects, rather than absolute poses. This design enables a privileged reinforcement-learning (RL) teacher to jointly learn latent states and mobile actions from sparse keypoints, which is then distilled into a purely visual student policy. To reduce the supervision gap between the omniscient teacher and the partially observed student, we restrict the teacher's observations to visually accessible cues. This induces active perception behaviors that are recoverable from the student's viewpoint. To address long-horizon credit assignment, we decompose rearrangement into stage-level subproblems using temporally decayed, stage-local completion rewards. Extensive simulation experiments demonstrate that EgoPush significantly outperforms end-to-end RL baselines in success rate, with ablation studies validating each design choice. We further demonstrate zero-shot sim-to-real transfer on a mobile platform in the real world. Code and videos are available at https://ai4ce.github.io/EgoPush/.

Rapid progress in video models has largely focused on visual quality, leaving their reasoning capabilities underexplored. Video reasoning grounds intelligence in spatiotemporally consistent visual environments that go beyond what text can naturally capture, enabling intuitive reasoning over spatiotemporal structure such as continuity, interaction, and causality. However, systematically studying video reasoning and its scaling behavior is hindered by the lack of large-scale training data. To address this gap, we introduce the Very Big Video Reasoning (VBVR) Dataset, an unprecedentedly large-scale resource spanning 200 curated reasoning tasks following a principled taxonomy and over one million video clips, approximately three orders of magnitude larger than existing datasets. We further present VBVR-Bench, a verifiable evaluation framework that moves beyond model-based judging by incorporating rule-based, human-aligned scorers, enabling reproducible and interpretable diagnosis of video reasoning capabilities. Leveraging the VBVR suite, we conduct one of the first large-scale scaling studies of video reasoning and observe early signs of emergent generalization to unseen reasoning tasks. Together, VBVR lays a foundation for the next stage of research in generalizable video reasoning. The data, benchmark toolkit, and models are publicly available at https://video-reason.com/ .

Following the rise of large foundation models, Vision-Language-Action models (VLAs) emerged, leveraging strong visual and language understanding for general-purpose policy learning. Yet, the current VLA landscape remains fragmented and exploratory. Although many groups have proposed their own VLA models, inconsistencies in training protocols and evaluation settings make it difficult to identify which design choices truly matter. To bring structure to this evolving space, we reexamine the VLA design space under a unified framework and evaluation setup. Starting from a simple VLA baseline similar to RT-2 and OpenVLA, we systematically dissect design choices along three dimensions: foundational components, perception essentials, and action modelling perspectives. From this study, we distill 12 key findings that together form a practical recipe for building strong VLA models. The outcome of this exploration is a simple yet effective model, VLANeXt. VLANeXt outperforms prior state-of-the-art methods on the LIBERO and LIBERO-plus benchmarks and demonstrates strong generalization in real-world experiments. We will release a unified, easy-to-use codebase that serves as a common platform for the community to reproduce our findings, explore the design space, and build new VLA variants on top of a shared foundation.

Compound AI systems promise capabilities beyond those of individual models, yet their success depends critically on effective orchestration. Existing routing approaches face two limitations: (1) input-level routers make coarse query-level decisions that ignore evolving task requirements; (2) RL-trained orchestrators are expensive to adapt and often suffer from routing collapse, repeatedly invoking one strong but costly option in multi-turn scenarios. We introduce SkillOrchestra, a framework for skill-aware orchestration. Instead of directly learning a routing policy end-to-end, SkillOrchestra learns fine-grained skills from execution experience and models agent-specific competence and cost under those skills. At deployment, the orchestrator infers the skill demands of the current interaction and selects agents that best satisfy them under an explicit performance-cost trade-off. Extensive experiments across ten benchmarks demonstrate that SkillOrchestra outperforms SoTA RL-based orchestrators by up to 22.5% with 700x and 300x learning cost reduction compared to Router-R1 and ToolOrchestra, respectively. These results show that explicit skill modeling enables scalable, interpretable, and sample-efficient orchestration, offering a principled alternative to data-intensive RL-based approaches. The code is available at: https://github.com/jiayuww/SkillOrchestra.

While Vision-Language-Action (VLA) models have seen rapid progress in pretraining, their advancement in Reinforcement Learning (RL) remains hampered by low sample efficiency and sparse rewards in real-world settings. Developing generalizable process reward models is essential for providing the fine-grained feedback necessary to bridge this gap, yet existing temporal value functions often fail to generalize beyond their training domains. We introduce TOPReward, a novel, probabilistically grounded temporal value function that leverages the latent world knowledge of pretrained video Vision-Language Models (VLMs) to estimate robotic task progress. Unlike prior methods that prompt VLMs to directly output progress values, which are prone to numerical misrepresentation, TOPReward extracts task progress directly from the VLM's internal token logits. In zero-shot evaluations across 130+ distinct real-world tasks and multiple robot platforms (e.g., Franka, YAM, SO-100/101), TOPReward achieves 0.947 mean Value-Order Correlation (VOC) on Qwen3-VL, dramatically outperforming the state-of-the-art GVL baseline which achieves near-zero correlation on the same open-source model. We further demonstrate that TOPReward serves as a versatile tool for downstream applications, including success detection and reward-aligned behavior cloning.

Sequential recommendation increasingly employs latent multi-step reasoning to enhance test-time computation. Despite empirical gains, existing approaches largely drive intermediate reasoning states via target-dominant objectives without imposing explicit feasibility constraints. This results in latent drift, where reasoning trajectories deviate into implausible regions. We argue that effective recommendation reasoning should instead be viewed as navigation on a collaborative manifold rather than free-form latent refinement. To this end, we propose ManCAR (Manifold-Constrained Adaptive Reasoning), a principled framework that grounds reasoning within the topology of a global interaction graph. ManCAR constructs a local intent prior from the collaborative neighborhood of a user's recent actions, represented as a distribution over the item simplex. During training, the model progressively aligns its latent predictive distribution with this prior, forcing the reasoning trajectory to remain within the valid manifold. At test time, reasoning proceeds adaptively until the predictive distribution stabilizes, avoiding over-refinement. We provide a variational interpretation of ManCAR to theoretically validate its drift-prevention and adaptive test-time stopping mechanisms. Experiments on seven benchmarks demonstrate that ManCAR consistently outperforms state-of-the-art baselines, achieving up to a 46.88% relative improvement w.r.t. NDCG@10. Our code is available at https://github.com/FuCongResearchSquad/ManCAR.

Unified multimodal models can both understand and generate visual content within a single architecture. Existing models, however, remain data-hungry and too heavy for deployment on edge devices. We present Mobile-O, a compact vision-language-diffusion model that brings unified multimodal intelligence to a mobile device. Its core module, the Mobile Conditioning Projector (MCP), fuses vision-language features with a diffusion generator using depthwise-separable convolutions and layerwise alignment. This design enables efficient cross-modal conditioning with minimal computational cost. Trained on only a few million samples and post-trained in a novel quadruplet format (generation prompt, image, question, answer), Mobile-O jointly enhances both visual understanding and generation capabilities. Despite its efficiency, Mobile-O attains competitive or superior performance compared to other unified models, achieving 74% on GenEval and outperforming Show-O and JanusFlow by 5% and 11%, while running 6x and 11x faster, respectively. For visual understanding, Mobile-O surpasses them by 15.3% and 5.1% averaged across seven benchmarks. Running in only ~3s per 512x512 image on an iPhone, Mobile-O establishes the first practical framework for real-time unified multimodal understanding and generation on edge devices. We hope Mobile-O will ease future research in real-time unified multimodal intelligence running entirely on-device with no cloud dependency. Our code, models, datasets, and mobile application are publicly available at https://amshaker.github.io/Mobile-O/

Byron Kaye / Reuters : Australia's eSafety Commissioner threatens action against app stores and search engines if AI services operating in Australia don't verify user ages by March 9 —  Australia's internet regulator said it may push search engines and app stores to block artificial intelligence services that fail …

Meir Orbach / CTech : Israel-based Guidde, which is developing a platform to accelerate the adoption of AI in organizations, raised a $50M Series B round led by PSG Equity —  The Israeli startup's platform turns employee workflows into structured knowledge for automation.  —  Guidde, a startup developing …

New York Times : Sources describe in detail the failed talks between Anthropic and DOD, and how officials at agencies, including the CIA, still hope for a peace agreement —  The Pentagon and Anthropic were close to agreeing on the use of artificial intelligence.  But strong personalities, mutual dislike and a rival company unraveled a deal.

Kohei Fujimura / Nikkei Asia : Chinese matchmaking apps like Wanmei Qinjia, which has 50M users and lets parents look for spouses for their children, surge as marriage rates continue to fall —  DALIAN, China — Apps that enable parents to search for spouses for their unmarried children have become increasingly popular in China …

Ross Andersen / The Atlantic : Source describes the failed Pentagon-Anthropic talks: through the end, the Pentagon wanted to use Anthropic's AI to analyze bulk data collected about Americans —  Right up until the moment that Pete Hegseth moved to terminate the government's relationship with the AI company Anthropic …

Kyt Dotson / SiliconANGLE : Nvidia partners with Cisco, Nokia, and others to build 6G networks based on open, software-defined AI radio access networking (AI-RAN) architecture —  Nvidia Corp. early Sunday announced ahead of the MWC Barcelona conference that its joining global telecom leaders in a commitment to build 6G …

浏览太多刺激大脑多巴胺的社媒内容是否会导致脑腐（Brain Rot）？根据多项研究，这可能是真的。研究显示，滚动浏览 TikTok、Instagram or YouTube Shorts 等平台上的短视频会影响注意力、记忆力和心理健康。有研究发现短视频使用量增加与认知能力下降和焦虑加剧相关。根据发表在《Translational Psychiatry》期刊上的一项研究，对逾 7000 名儿童的分析发现，屏幕使用时间越长，大脑部分区域的皮层厚度越小。皮层是负责高级思维、记忆和决策的大脑区域，它对控制成瘾行为也非常重要。另一项研究发现，如果儿童手机移除了社媒应用，但不限制他们使用手机，那么负面影响会显著减少。

I2P 匿名网络在 2 月 3 日遭遇了来自 Kimwolf 物联网僵尸网络的女巫攻击（Sybil attack）。所谓女巫攻击是指攻击者通过创建女巫（Sybil）节点操控整个网络系统，破坏了系统的正常运行。I2P 去中心化匿名网络通常只有 1.5-2 万个活跃设备，但当天涌入的恶意节点多达 70 万个，恶意节点的数量是合法节点的 39 倍。Kimwolf 的主要 CC 指令控制服务器此前遭到了 Google 等公司的破坏，该僵尸网络的运营者在 Discord 上表示它尝试将 I2P 网络作为备用的 CC 基础设施，结果意外破坏了 I2P 网络。I2P 团队在 6 天后释出了 v2.11.0，加入了针对女巫攻击的缓解措施，默认启用了后量子加密算法 ML-KEM 和 X25519。

Nala Ginrut 写道： 当 AI 成为生产力基础设施时，我们是否仍然保有迁移能力与选择权？如果今天是窗口期，那么在窗口期内做出怎样的准备，才能避免在锁定期和收缩期中被动应对？ 这里涉及一个概念，我称之为“技术格局”。它并不意味着对抗或拒绝平台，也不是强调自给自足，而是指在关键生产工具上，个体能够保留基本的迁移能力与选择空间。

DNA 技术和家谱基因数据库再次帮助警方破解了一起陈年悬案。加州 Cloverdale 的 13 岁女孩 Sarah Geer 于 1982 年 5 月 23 日晚上离开朋友家后失踪，第二天早上一名消防员发现了她的尸体。她的死被定为谋杀，但因为技术限制，未能确定谋杀嫌疑人。这起案件被搁置了逾 40 年。FBI 根据 Sarah 身上收集的 DNA 以及家谱基因数据库判断凶手是四兄弟之一，调查人员对他们进行了监视，收集了丢弃的香烟，确定现年 64 岁的 James Unick 是凶手。在 Sarah 遇害近 44 年后，陪审团于 2 月 13 日裁定其谋杀罪名成立。当地检方在一份声明中表示，虽然 44 年的等待实在太久，但正义终得伸张。

微软 Xbox 和游戏业务 CEO Phil Spencer 在公司工作 38 年之后离职，被广泛视为其接任者的 Xbox 总裁 Sarah Bond 也已辞职，游戏业务的新 CEO 将是负责 CoreAI 产品的 Asha Sharma。Spencer 是在 2014 年 3 月被任命为 Xbox 负责人，他任内推出了游戏订阅服务 Xbox Game Pass，最为人所知的事情可能是以 690 亿美元收购动视暴雪。他还收购了一系列游戏工作室，其中包括 2020 年以 75 亿美元收购 Bethesda 母公司 ZeniMax，完全控制了著名的游戏 IP 如 《辐射》和《上古卷轴》（Bethesda）、《毁灭战士》和《雷神之锤》（id Software）。

NASA 计划于 3 月 6 日执行 Artemis II 载人绕月任务。执行该任务的登月火箭 Space Launch System (SLS）已经竖立在佛罗里达肯尼迪太空中心的发射台上。NASA 官员将在下周对其进行为期数天的飞行准备评估，确保火箭各个方面都准备就绪。本月早些时候 SLS 在首次测试火箭燃料加注时遭遇了液氢泄漏问题，官员称在更换部分密封件之后该问题看起来已经解决了。

根据发表在《JGR Oceans》期刊上的一项研究，过去 20 年从太平洋流入北冰洋“加拿大海盆”的海水的热输送量增至 1.5 倍。分析认为，除流入水温升高外，北冰洋海冰减少也进一步推高了水温。受全球变暖影响，北冰洋海冰正在减少，尤其是太平洋一侧的减幅较大。研究团队自 2000 年起，在阿拉斯加州巴罗角近海观测海水温度与流速，从太平洋经白令海峡的海水主要汇入该处。结果显示，流速未见变动趋势，但水温呈长期上升。海洋热输送量也呈增加趋势，在 2000年-2022 年间增至原来的 1.5 倍。基于卫星海表温度等数据，研究还发现热输送自 2010 年代后半期起急剧增加。海冰较少的年份热输送较多，海冰较多的年份热输送较少。海冰较少时，海水更易吸收日照导致水温上升，进而加速海冰融化，形成反馈效应。

在索尼之后，曾以等离子电视闻名的松下宣布其电视业务将由创维接手。从 4 月开始，欧洲和北美的松下电视销售业务移交给创维，双方还将在产品研发和生产方面进行合作，松下将专注于在日本的销售和高端机型的生产，借助其他地区的销售和低价产品的生产委托给外部，有助于提高正在下滑的电视业务的收益。在销售方面，日本市场仍由松下自己负责，而欧美则由创维负责。剩下的亚洲市场今后将讨论包括与创维合作在内的各个国家和地区的最佳措施。在等离子电视时代，松下一度占据近半市场份额，2010 年松下控制了 40.7% 的等离子面板市场份额，超过三星（33.7%）和LG（23.2%），但随着消费者日益对 LCD 电视感兴趣，松下于 2014 年 3 月停产等离子电视。日本公司如夏普、东芝、日立以及索尼都基本退出了电视市场。

Mozilla 释出了 Firefox 148，引入了 AI 关闭开关，允许用户关闭所有 AI 功能，Mozilla 承诺未来的更新不会覆盖该设置。该开关位于 设置 > AI Controls 下。Mozilla 还允许用户最大限度退出数据收集，相关选项位于 设置 > 隐私设置 > Firefox 数据收集下。其它变化包括：集成 Trusted Types API 和 Sanitizer API 以遏制跨站脚本攻击(XSS)，改进了 PDF 中屏幕阅读器对数学公式的兼容性；Firefox Backup on Windows 10；支持 WebGPU 的 Service Worker 等等。

Ladybird 浏览器项目宣布将在 AI 帮助下使用 Rust 语言。Ladybird 是非盈利组织 Ladybird Browser Initiative 开发的开源浏览器，计划在今年内发布一个 alpha 版本，2028 年发布正式版本，它最初使用的语言是 C++，开发者表示他们一直在寻找一种内存安全语言替代 C++，他们在 2024 年评估过 Rust，但因为它在 C++ 风格的面向对象编程(OOP)上表现不佳而放弃，但一年之后它还是决定采用 Rust，而 Firefox 和 Chromium 都已开始在其代码库中引入 Rust。Ladybird 将首先用 Rust 重写部分代码，第一个目标是 JavaScript 引擎 LibJS，开发者在 AI 辅助编程工具 Claude Code 和 Codex 帮助下完成了 2.5 万行的代码。Rust 将主要用于开发子系统，浏览器引擎仍然继续使用 C++ 开发。

荷兰 ASML 的研究人员改进了极紫外光刻（EUV）设备所使用的光源功率，有望在这个十年结束前将芯片产量提高 50%。研究人员找到了方法将极紫外光源功率从目前的 600 瓦提高至 1000 瓦。更大的功率意味着每小时可以生产更多芯片，有助于降低单个芯片的成本。芯片的制造方法类似照片打印，利用极紫外光照射涂有光刻胶的硅晶圆，使用更大的极紫外光源，芯片工厂所需的曝光时间更短。ASML 极紫外光刻机执行副总裁 Teun van Gogh 表示，到 2030 年极紫外光刻机每台机器每小时能处理约 330 片硅晶圆，而目前是 220 片。

荷兰国防部副部长 Gijs Tuinman 透露，F-35 能被越狱安装第三方软件，就像以前的 iPhone。他没有透露多少越狱细节。F-35 战斗机包含了云端组件 ALIS/ODIN network，它除了用于处理软件更新和后勤数据外，还被用于在执行任务前上传高度敏感的任务数据，在任务结束后下载情报等数据。采购 F-35 战斗机的美国盟友中，只有以色列允许安装自己开发的软件，允许在 ALIS/ODIN network 之外操作战斗机。其它国家的 F-35 高度依赖于美国的维护和后勤保障体系，因此越狱可能会导致美国停止维护，最终导致战斗机无法正常工作。