Core Viewpoint - The article discusses the rapid updates and improvements in the OpenClaw platform, highlighting the transition to version 3.23, which addresses previous issues and introduces new features, particularly focusing on the integration of DeepSeek and Qwen models with a pay-as-you-go pricing model. Group 1: Update Frequency and Issues - The update frequency of OpenClaw is notably high, with version 3.23 released just 12 hours after the significant 3.22 version, which was described as having the largest changes in history [2] - The 3.22 version caused widespread issues with IM plugins like WeChat due to the aggressive removal of old APIs, leading to a UI crash [5][6] - Version 3.23 rectified these issues by restoring missing runtime files and enhancing API compatibility checks [5][8] Group 2: New Features and Integrations - The 3.23 version officially integrates the DeepSeek plugin, allowing users to access DeepSeek models directly via API Key [2][14] - Qwen has been rebranded as Qwen (Alibaba Cloud Model Studio) and now supports a standard pay-as-you-go endpoint for both Chinese and global API Keys [12][13] - The update allows domestic developers to run domestic models in OpenClaw through a more affordable and legitimate pathway [15] Group 3: Security and Performance Enhancements - The new version implements SHA-256 hashing for all inline scripts, enhancing security by rejecting any malicious script injections not on the official whitelist [17][18] - A bug affecting Mac users that caused repeated pop-ups when connecting to Chrome has been fixed, improving response times by nearly double [19] - The update also optimizes the reasoning chain for Anthropic's Claude 3.7, ensuring uninterrupted AI logic during deep reasoning tasks [20]

Core Viewpoint - The article emphasizes the ongoing AI boom and invites individuals to join the company "Quantum Bit" to track AI advancements and become content experts in various AI-related fields [1]. Group 1: Job Opportunities - The company is hiring for three main directions: AI Industry, AI Finance, and AI Product, all full-time positions based in Beijing [3]. - Positions are open for both experienced professionals and fresh graduates, with opportunities for internships that can lead to full-time roles [4]. Group 2: AI Industry Direction - Responsibilities include tracking innovations in foundational layers such as chips, AI infrastructure, and cloud computing [6]. - Candidates should have a basic understanding of chips, GPUs, NPUs, servers, and model training architectures, as well as familiarity with the AI industry's supply chain and ecosystem [8]. Group 3: AI Finance Direction - Focus on venture capital, AI startups, public companies, and capital movements within the industry [7]. - Candidates should be data-sensitive and interested in financial reports, equity structures, and strategic planning [8]. Group 4: AI Product Direction - Responsibilities involve monitoring AI applications in software and hardware, writing in-depth product reviews, and interviewing entrepreneurs and product experts [9][11]. - Candidates should have a keen sense of trends in smart hardware and AI terminals, along with strong logical and structured expression abilities [11]. Group 5: Benefits and Growth - Employees can build personal influence by writing original content and expanding their industry network through interactions with AI experts [8]. - The company offers competitive salaries, comprehensive benefits, and a dynamic team environment that encourages growth and recognition based on merit [8][10]. Group 6: Company Overview - As of 2025, Quantum Bit aims to have over 2.4 million subscribers on WeChat and over 7 million users across platforms, with a daily reading volume exceeding 2 million [10]. - The company is recognized as a top media outlet in the AI and frontier technology sectors [10].

Core Insights - The article discusses the latest advancements in the LeCun world model, specifically the open-sourced LeWorldModel, which allows for extremely simplified training on a single GPU, achieving rapid planning in just one second [1][2]. Group 1: Model Architecture and Training - LeWorldModel (LeWM) is based on the JEPA architecture, enabling direct pixel input to predict future states with remarkable speed [2][3]. - The model simplifies the JEPA approach by using an encoder to convert images into latent features and a predictor to forecast the next features based on actions, employing Gaussian regularization to prevent collapse [6][11]. - The architecture consists of two core components: an encoder that compresses images into a small string of numbers (latent features) and a predictor that estimates the next features based on current features and intended actions [7][8]. Group 2: Performance Metrics - LeWM demonstrates superior performance in various tasks, achieving a 96% success rate in the Push-T task, which is 18% higher than the previous PLDM method and even surpasses the DINO-WM model with body input [17]. - In the Reacher task, LeWM outperforms PLDM and is comparable to DINO-WM, while in the OGBench-Cube task, it remains competitive despite slightly trailing DINO-WM [17]. - The model's planning speed is 48 times faster than DINO-WM, completing tasks in under one second compared to approximately 47 seconds for DINO-WM [19][20]. Group 3: Loss Functions and Training Simplification - The key innovation of LeWM lies in its use of only two loss functions: prediction loss, which encourages the predictor to accurately guess the next frame's features, and a regularization loss that enforces a standard Gaussian distribution on feature vectors to prevent model collapse [11][12]. - The total loss function is a combination of prediction loss and a weighted regularization loss, with the regularization weight being the only hyperparameter that requires tuning, significantly simplifying the training process [13]. Group 4: Experimental Results and Insights - Experimental results indicate that LeWM outperforms the previous end-to-end JEPA method (PLDM) and matches or exceeds the performance of DINO-WM, while being easier to train, faster, and requiring fewer parameters [14]. - The model effectively captures the core structure and dynamics of the environment, accurately predicting object movements and identifying "physically impossible" scenarios [24][25]. - In experiments with visual and physical disturbances, the model reacted differently, showing surprise at physical violations while remaining indifferent to mere color changes [26][28].

Core Viewpoint - The article emphasizes the transition of AI from a niche technology to a mainstream tool that is now widely recognized and utilized by the general public, marking a significant shift in its adoption and application [2][5][18]. Group 1: AI Adoption and Impact - AI has evolved from being a subject of interest in the tech community to becoming a household name, with applications in daily tasks such as cooking, cleaning, and healthcare [2][3]. - The upcoming 2026 China AIGC Industry Summit aims to facilitate the understanding and practical use of AI, inviting entrepreneurs, developers, and industry players to engage in discussions and share experiences [5][12]. - The summit will feature over 60 industry leaders sharing insights on AI's impact on content production, research efficiency, marketing strategies, team collaboration, and decision-making processes [9][18]. Group 2: Summit Details and Structure - The summit will take place in May 2026 in Beijing, focusing on the theme "Everyone, Let's Get AI Started" [6][24]. - The agenda includes two main sessions: the morning session will discuss the necessity of adopting AI and showcase successful case studies, while the afternoon session will explore the integration of AI across various sectors such as healthcare, gaming, and creative industries [13][14]. - The event is expected to attract significant attention, with over a thousand attendees in person and more than 3.5 million online viewers, highlighting the growing interest in AI [12]. Group 3: Recognition and Awards - The article mentions the evaluation of noteworthy AIGC companies and products for 2026, based on their performance and feedback over the past year, with results to be announced at the summit [19][20]. - The selection process will be grounded in objective data and expert opinions, ensuring credibility and professionalism in the recognition of outstanding contributions to the AI field [19].

Core Viewpoint - The article discusses the launch of Alibaba's enterprise-level agent, Accio Work, which enables small and medium-sized enterprises and individuals to easily set up and manage online stores, significantly lowering the barriers to entry for global e-commerce [4][5][10]. Group 1: Product Features - Accio Work allows users to create an online store in just 30 minutes, providing a comprehensive AI agent team that assists with product selection, procurement, and operations without the need for complex installations [3][11]. - The platform includes built-in business skills and offers a user-friendly interface that resembles a chat window, allowing users to interact with agents directly or in groups [6][11]. - Users can customize agents based on their specific tasks, and the platform supports the creation of scheduled tasks for ongoing operations [12][19]. Group 2: Market Analysis and Strategy - Accio Work provides detailed market trend analysis to help users position their stores effectively, focusing on high-end products with therapeutic benefits [26][30]. - The platform calculates profit margins for various products, ensuring users can make informed decisions about product offerings [29]. - The brand design process is streamlined, with agents creating visual identities that resonate with target audiences [30]. Group 3: Operational Efficiency - The platform automates various aspects of e-commerce, including store setup, product listing, and marketing strategies, which traditionally required extensive human resources [57][58]. - Accio Work's ability to handle complex e-commerce processes simplifies operations for users, allowing them to focus on growth rather than logistics [56][58]. - The transition from human-to-human interactions to agent-to-agent interactions is expected to enhance efficiency in trade processes [66]. Group 4: Industry Implications - The article highlights a significant talent gap in the global e-commerce sector, driven by information barriers, complex processes, and high costs, which Accio Work aims to address [48][51][52]. - The platform leverages Alibaba's extensive experience in global trade to offer localized and differentiated operational strategies, providing a competitive edge over similar products [56]. - The rise of one-person companies supported by tools like Accio Work is anticipated to reshape the e-commerce landscape, making it more accessible to a broader audience [65][74].

Core Viewpoint - The article discusses the challenges and advancements in 3D reconstruction for long sequences in real-time settings, highlighting the introduction of LongStream as a solution to these challenges [1][2]. Group 1: Challenges in Long Sequence 3D Reconstruction - Existing methods perform well in short sequences but struggle with real-time long video scenarios, leading to significant issues in accuracy and stability [2]. - Key problems include reliance on the first frame for pose anchoring, attention sink phenomena, and KV cache pollution, which degrade performance over time [5][6]. Group 2: Innovations of LongStream - LongStream introduces a Gauge-decoupled streaming visual geometry architecture that addresses the limitations of traditional methods by: 1. Eliminating first-frame anchoring, allowing for relative pose predictions that enhance robustness in long sequences [10]. 2. Implementing cache-consistent training to minimize the training-inference gap and reduce attention sink effects [11]. 3. Utilizing periodic cache refresh to mitigate memory saturation and geometric drift, maintaining reconstruction consistency [11]. Group 3: Experimental Results - LongStream demonstrates competitive performance across various benchmarks, including KITTI, Waymo, and TUM-RGBD, achieving stable reconstruction with low memory usage and maintaining 18 FPS streaming inference [12][16]. - In comparison to baseline methods, LongStream shows significantly lower Average Trajectory Error (ATE) across multiple datasets, indicating superior long-sequence stability and accuracy [17][18]. Group 4: Importance of LongStream - The significance of LongStream lies in its ability to support continuous online 3D world modeling, which is crucial for applications in robotics, autonomous driving, AR glasses, and embodied AI [19][21]. - This approach shifts the paradigm from offline reconstruction to real-time world maintenance, making it a vital development for future visual systems [22].

衡宇 发自 凹非寺 量子位 | 公众号 QbitAI 顺着周凯文人事变动这条线往下扒，我们发现了三条很值得和大家分享的信息。 第一 ，这家叫诺因智能的具身智能创业公司，成立时间很短，不满一年，但已经聚集了一批履历极强的技术人员。 又一位华为天才少年加入具身智能创业战场。 6G冲浪的量子位最新发现，去年从华为诺亚方舟实验室转入学界的周凯文，在入职港中文不到半年后，悄然更新了个人主页。 他已加入具身创企 诺因智能 ，担任合伙人兼算法主管。 第二 ，该公司选择的方向，是当前争议最大、也最难落地的ToC具身智能机器人。 第三 ，也是比较亮眼的一点，诺因智能刚发了个新具身模型，已经在一个具身智能权威榜单上拿下第一。 综上，这家半年内连融三轮，但相当低调的具身公司，已经越来越掩藏不住了。 从一条人事变动，扒一扒ToC具身低调玩家 在火热的具身智能赛道，顶级人才的流动本身就是一个重要信号。 周凯文 2013年以信息奥林匹克竞赛保送复旦大学，2019年在香港中文大学拿到计算机科学与工程的硕士学位，而后又在港中文攻读博士学 位。 2022年博士毕业后，周凯文以"华为天才少年"身份加入华为诺亚方舟实验室。业内人士告知，他 是诺亚 ...

一水 发自 凹非寺 量子位 | 公众号 QbitAI 故事省流版是这样的： 有一个人搞了一个"一人公司"，还请了一位AI当联创 （大名叫Kyle Law） 。 领英堕落？"互联网已死"理论成真？ 全球知名招聘网站LinkedIn （中文名领英） 最近的一场风波，再次把一个问题推至台前—— 当内容的生产、分发、互动甚至"用户"本身都开始被AI替代时，互联网还是我们熟悉的那个互联网吗？ 虽然问题暂无终局答案，但至少领英用 微妙的态度 表明： 这一次，事情真的变得有点不一样了。 LinkedIn对AI前后变脸，CEO在线控诉 这位AI联创在领英那叫一个活跃——发帖、互动的频率约等于"罗伯特"，由此还积攒了数百个联系人和粉丝。 然后它就被领英官方注意到了，并且几个月前，领英私下还请CEO及Kyle Law参加了内部座谈会，据称反响热烈。 领英组织者似乎很喜欢。聊天室热闹起来了……有人说，"当AI嘉宾的魅力比一些真正的CEO还要强……"。 本以为这是个大家都好的happy ending，结果没想到领英光速变脸—— 距离座谈会过去才两天，反手就把这位AI联创的号封了。 于是现在，创办这家公司的老哥开始在线发帖控诉。 就怎 ...

Core Viewpoint - The latest upgrade of Claude Code by Anthropic effectively addresses the shortcomings of the open-source project OpenClaw, enhancing its capabilities to operate like a human using a computer through real-time screen capture and simulation of mouse and keyboard actions [3][4]. Group 1: Claude Code Upgrade Features - The upgrade eliminates the need for API interfaces or CLI modifications, allowing traditional software, including legacy enterprise management systems and professional creative software like Photoshop, to be controlled by the agent [8]. - Two new high-demand features were introduced: remote control capability, enabling users to assign tasks to Claude from their mobile devices, and a scheduling function for automatic daily task execution [9]. Group 2: Security and Accessibility - The security design includes layered access, prioritizing previously authorized integrations (e.g., Slack, calendar, Google Workspace) and only requesting desktop operation permissions when no corresponding connector exists. Sensitive operations will prompt user confirmation before execution [11]. - The new functionality is currently available to all Claude Pro and Max users, requiring the latest desktop version and mobile account pairing to enable the Computer Use preview feature [13]. Group 3: Platform Availability - The feature is currently exclusive to macOS, with plans to quickly iterate based on user feedback and expand support to Windows and Linux systems in the near future [14].

Core Viewpoint - The article discusses the significant update of the software "龙虾" (Lobster), highlighting its self-update capability and extensive enhancements across various functionalities, including security, user interface, and plugin management. Update Highlights - The update is described as the largest ever, with numerous features and improvements [3] - Key features include self-updating capability of the software [4] - Enhancements in security, user interface, Android mobile support, and social media integration are noted [5] Plugin Updates - The plugin distribution mechanism has been optimized for security and development standards, with the old extension API removed [6] - New models MiniMax M2.7 and GPT-5.4-mini/nano have been introduced, along with a default mode for agents [6] - The installation of plugins now prioritizes ClawHub, with a streamlined import process [7][8] - The Matrix plugin is now officially supported by matrix-js-sdk, improving compatibility and encryption [9] Security Enhancements - The update strengthens identity verification and execution auditing, introducing native SSH sandbox support [10] - The core library now includes shared remote execution and file system tools, with a focus on sandbox lifecycle management [11] - The architecture has been updated to allow for pluggable backend designs, enhancing flexibility [11] Interaction and Performance - The system now employs a Compact Directory Fallback strategy to retain registered skill entries when prompts exceed limits [13] - Plugins can dynamically adjust context formats based on model IDs during the assembly phase [14] - Improvements in message handling during low-frequency interactions help save server resources [16] - A new feature for inbound room event deduplication has been introduced to manage message storms during gateway restarts [17][18] Model Updates - The update includes deep synchronization with mainstream model libraries, adding support for gpt-5.4-mini and gpt-5.4-nano [19] - The default model has switched to gpt-5.4, with all default values centralized for seamless future upgrades [19] - The MiniMax family has been upgraded to M2.7, with new high-speed versions introduced [19] UI and Mobile Optimizations - The user interface now features a "Roundness" adjustment slider for visual style customization [23] - The mobile version supports system-wide dark mode and has improved SMS and call log search capabilities [24] - Enhanced functionalities for social media platforms like 飞书 (Feishu) and Telegram have been added [25][26] Important Fixes - The update addresses several critical security vulnerabilities, including potential Windows password leaks and command forgery risks [29][30] - Performance improvements have reduced cold start times significantly, enhancing user experience [31] - Compatibility with OpenAI and third-party models has been strengthened, ensuring smoother operations [31] Community Feedback - Users have expressed satisfaction with the update, particularly regarding the plugin market and sandbox features [33] - Some users reported issues with specific plugins, indicating potential challenges in the update process [34][35]