目前，我们有 三大方向 岗位招聘，希望你是 （或者能成为） 这三个方向的内容专家： 岗位均为全职，工作地点：北京中关村。 岗位面向： 编辑部 发自 凹非寺 量子位 | 公众号 QbitAI AI热潮还在汹涌，但如果你还不知道如何参与……那为什么不来 量子位 呢？ 我们是一家以 追踪AI新进展 为核心的内容平台，经过8年积累，目前拥有顶流影响力，广泛且备受认可的产业资源，以及时代风口的最佳观 测和学习生态位。 主编 ：具备选题和带队能力及经验； 主笔 ：具备原创深度稿件能力； 编辑 ：热爱表达，喜欢挖掘信息，能够用大白话让所有人看懂AI新进展。 加入我们，你可以获得： 以下是岗位详情： 所有岗位不同能力层级职位均在开放，欢迎结合个人履历和经验申请。 AI产业方向 岗位职责： AI产业方向 ：关注基建层创新，包含芯片、AI Infra、云计算； AI财经方向 ：关注AI领域创投和财报，跟踪产业链资本动向； AI产品方向 ：关注AI在应用和硬件终端方向的进展。 社招：覆盖编辑、主笔、主编各个层级，按能力匹配岗位； 校招：应届毕业生，接受实习且可转正。 站在AI浪潮之巅 ：第一时间接触和了解AI领域最新技术和产品，构 ...

henry 发自 凹非寺 量子位 | 公众号 QbitAI 2026年的春晚，注定是属于机器人的。 不知道有多少人像我一样，在看到马丽的"铁哥们"时，没绷住，直接笑出了声。 这个叫"小盖"的家伙，是真的物理意义上的"铁哥们"。 节目里，腾哥上春晚得靠它，它甚至还给自己排了一出表演干活的"剧本"： 盘盘核桃、叠叠衣服、呛呛腾哥，直接把春晚的笑点拉满了。 但笑过之后，很难忽视的一点是——小盖这次在春晚做的事情不太一样。 它不是在秀动作，而是在执行任务： 货架取物、清理碎片、整理环境，甚至串烤肠、叠衣服。 据悉，这是春晚舞台上 第一个上台认真执行任务 、在自己工位上"干活"的机器人。 而把"小盖"带上春晚舞台干活的，正是被官方定位为具身大模型机器人的 银河通用机器人 ，以及它背后那套首次公开的"大脑-小脑-神经控 制"一体化系统 银河星脑AstraBrain 。 也正是这套系统，让舞台上那些看似随意的取物、清理、整理，不再依赖动作脚本，而是来自完整的感知-决策-执行闭环。 而在节目外，小盖也已经将这套能力延伸到了全国各地的100家银河太空舱，真正让机器人走进生活。 春晚大舞台，"干活"银河来 咱不演"马丽单飞，沈腾 ...

编辑部 发自 凹非寺 量子位 | 公众号 QbitAI AI热潮还在汹涌，但如果你还不知道如何参与……那为什么不来 量子位 呢？ 我们是一家以 追踪AI新进展 为核心的内容平台，经过8年积累，目前拥有顶流影响力，广泛且备受认可的产业资源，以及时代风口的最佳观 测和学习生态位。 目前，我们有 三大方向 岗位招聘，希望你是 （或者能成为） 这三个方向的内容专家： 站在AI浪潮之巅 ：第一时间接触和了解AI领域最新技术和产品，构建完整的AI认知体系。 玩转AI新工具 ：将各种AI新技术、新工具应用于工作，提升工作效率和创造力。 打造个人影响力 ：通过撰写独家原创内容，建立个人知名度，成为AI领域的意见领袖。 拓展行业人脉 ：与AI领域大咖零距离接触，参与重要科技活动和发布会，拓展行业视野。 获得专业指导 ：应届新人会由主编级编辑出任mentor，提供一对一指导，帮你更快进步获得成长。 加入活力团队 ：与一群志同道合的年轻人一起工作，享受扁平、简单、开放、多劳多得能者上位的团队氛围。 获得丰厚回报 ：行业TOP薪资待遇，五险一金、餐补、项目绩效、商务绩效、加班补助等福利一应俱全。 主编 ：具备选题和带队能力及经验； ...

Core Viewpoint - The article highlights the significant advancements in AI technology showcased during the Spring Festival Gala, particularly focusing on the capabilities of the Seedance 2.0 model and its integration with various AI applications in performance and interaction [2][42]. Group 1: AI Technology in Performance - The performance of "Yufeng Song" by Zhang Jie featured a background video created using the Seedance 2.0 model, which successfully interpreted and animated traditional Chinese ink painting styles, a task that many foreign models struggled with [4][5]. - Seedance 2.0 was utilized in multiple performances, including the creative dance show "He Huashen," where it demonstrated micro-control capabilities to create detailed visual effects [7][10]. - The model's ability to follow physical and biomechanical principles allowed for realistic animations of galloping horses, showcasing its advanced command-following and multi-modal material reference capabilities [8][10]. Group 2: Video Quality Enhancement - The collaboration with the Volcano Engine video cloud team enabled the enhancement of video quality to meet the Spring Festival Gala's high standards, utilizing super-resolution algorithms to upscale 720P to 8K and frame interpolation to increase frame rates from 24 to 50 FPS [15][17]. - The integration of 4D Gaussian splashing technology allowed for the creation of immersive visual experiences, where virtual dancers interacted seamlessly with real stage lighting [20][22]. Group 3: AI Interaction and User Engagement - The Spring Festival Gala introduced AI-driven interactive features through the Doubao app, allowing users to generate personalized avatars and greetings, marking a shift from traditional transactional interactions to more complex, computationally intensive engagements [28][30]. - The Ark platform played a crucial role in managing the high traffic during the event, utilizing a federated system to optimize resource allocation and ensure rapid response times for user requests [31][29]. Group 4: Broader Implications and Industry Impact - The article emphasizes the widespread adoption of Doubao's AI models across various industries, including automotive, mobile, and robotics, highlighting its robust partnerships with major companies [40][41]. - The successful implementation of AI technologies during the Spring Festival Gala serves as a demonstration of their practical value and potential for real-world applications, reinforcing the notion that effective AI solutions can deliver tangible benefits [43][44].

范式转移：从"模型森林"到"智能中枢化" 在现代广告推荐技术栈中，依赖大量离线任务支撑，如：query-ad相关性标注、用户画像生成、关键词扩写、创意优化……这些离线任务 通常用来为在线模型提供特征、数据和标签，工程师们为每个子任务都微调专属的BERT或小型LLM。这种"一任务一模型"的体系存在很多 痛点，如： AdNanny团队 投稿 量子位 | 公众号 QbitAI 微软用一个671B的"推理中枢"，把广告系统的脏活累活都管了，性能还全面碾压一众前辈。 在工业级广告推荐系统中，普遍正面临一个吊诡的现状：在通用大语言模型 （LLM） 的推理能力已经登峰造极的同时，为了追求毫秒级的 响应，通常无法直接把LLM用到线上而是在离线端堆积了成百上千个"小模型"——有的管相关性标注，有的管用户画像，等等。 这种 "模型森林" 范式正逐渐成为进化的阻碍。模型间知识割裂、运维成本高昂、决策过程黑盒化。 近日，微软Bing Ads与DKI团队发表论文《AdNanny: One Reasoning LLM for All Offline Ads Recommendation Tasks》，宣布基于 DeepSeek-R1 6 ...

编辑部 发自 凹非寺 量子位 | 公众号 QbitAI AI热潮还在汹涌，但如果你还不知道如何参与……那为什么不来 量子位 呢？ AI产业方向 ：关注基建层创新，包含芯片、AI Infra、云计算； AI财经方向 ：关注AI领域创投和财报，跟踪产业链资本动向； AI产品方向 ：关注AI在应用和硬件终端方向的进展。 社招：覆盖编辑、主笔、主编各个层级，按能力匹配岗位； 校招：应届毕业生，接受实习且可转正。 我们是一家以 追踪AI新进展 为核心的内容平台，经过8年积累，目前拥有顶流影响力，广泛且备受认可的产业资源，以及时代风口的最佳观 测和学习生态位。 目前，我们有 三大方向 岗位招聘，希望你是 （或者能成为） 这三个方向的内容专家： 岗位均为全职，工作地点：北京中关村。 岗位面向： 加入我们，你可以获得： 以下是岗位详情： 所有岗位不同能力层级职位均在开放，欢迎结合个人履历和经验申请。 AI产业方向 岗位职责： 站在AI浪潮之巅 ：第一时间接触和了解AI领域最新技术和产品，构建完整的AI认知体系。 玩转AI新工具 ：将各种AI新技术、新工具应用于工作，提升工作效率和创造力。 打造个人影响力 ：通过撰写独家原创内 ...

Core Viewpoint - Alibaba's new AI model Qwen3.5-Plus has been released, claiming the title of the strongest open-source model, outperforming many closed-source models in various benchmarks [1][3]. Performance and Features - Qwen3.5-Plus has 397 billion parameters, with only 17 billion activated during inference, yet it outperforms the trillion-parameter Qwen3-Max [4]. - The model reduces deployment memory usage by 60% and increases maximum inference throughput by up to 19 times, significantly optimizing deployment costs and efficiency [5][60]. - Qwen3.5-Plus achieves state-of-the-art performance across multiple dimensions, including reasoning and programming, with a score of 87.8 on the MMLU-Pro test, surpassing GPT-5.2 [17]. Accessibility and Pricing - The API pricing for Qwen3.5 is highly competitive, with input costs as low as 0.8 yuan per million tokens, which is 1/18 of the cost of similar models like Gemini-3-Pro [9]. - The model supports 201 languages, expanding its vocabulary from 150k to 250k, and improves encoding efficiency for less common languages by 60% [9]. Technological Innovations - Qwen3.5-Plus incorporates several key technological advancements, including a mixed attention mechanism that dynamically allocates computational resources based on the importance of information [53]. - The model employs a sparse MoE architecture, activating only 17 billion parameters during inference, which significantly reduces computational costs while retaining knowledge advantages [55]. - A native multi-token prediction mechanism allows for batch output, nearly doubling inference speed compared to traditional models [56]. Multi-Modal Capabilities - Qwen3.5-Plus is designed for native multi-modal understanding, processing text and visual data simultaneously without the need for separate alignment networks [64]. - The model can handle long video inputs of up to 2 hours, enabling precise analysis and summarization of lengthy content [26]. Market Position and Impact - Since its inception, Alibaba has open-sourced over 400 models, achieving over 1 billion downloads globally, and establishing itself as a leader in the AI model space [71][72]. - The competitive pricing and open-source nature of Qwen3.5-Plus aim to democratize access to advanced AI technologies, similar to the paths taken by Linux and Android in their respective domains [73].

GAPO团队 投稿 量子位 | 公众号 QbitAI 程序员们又能少掉头发了！ 新研究通过过滤掉训练中的噪声和异常值，显著提升代码大模型在实际编辑任务中的准确性和效率。 在AI辅助编程成为软件开发核心生产力的今天，大语言模型 （LLMs） 已深度融入代码编辑、调试与优化全流程。 然而，当企业试图用 真实复杂用户环境中采集的数据 开展强化学习 （RL） 训练时，一个棘手的实际问题浮出水面：复杂上下文 （context） 导致大模型的输出答案频繁出现异常内容，即rollout噪声更普遍，使得reward出现异常值 （outliers） ，直接造成优势值 （advantage） 估计不准确，严重拖累强化学习效果。 上海交通大学、腾讯CodeBuddy等团队联合提出的 Group Adaptive Policy Optimization（GAPO） 方法，精准直击这一产业落地关键 瓶颈，为代码LLM的工业化训练提供了兼具科研创新性与工程实用性的突破方案，引发AI科研界与产业界广泛关注。 真实场景的核心梗阻：复杂上下文→rollout噪声→优势估计失真 代码编辑的核心难点在于，真实用户场景的输入提示绝非简单的代码片段， ...

Core Viewpoint - OpenAI's internal model has demonstrated significant progress in solving real-world mathematical problems, indicating an evolution in its reasoning capabilities, especially in research-level contexts [1][2][52]. Group 1: Model Performance - OpenAI's internal model attempted to solve ten real mathematical problems, with five solutions deemed fundamentally correct [2][11]. - The problems were not standard test questions but derived from actual research scenarios faced by mathematicians, which reduces the likelihood of the model simply recalling answers from training data [5][6]. - The model's performance is noteworthy as it managed to provide reliable answers to specific problems, showcasing its ability to engage in autonomous reasoning rather than mere knowledge recall [52][54]. Group 2: Testing Methodology - The evaluation was conducted over a week, primarily querying the current training model without providing proof strategies or mathematical hints [14]. - Feedback from experts was utilized to refine the model's answers, indicating a collaborative approach to validating the model's outputs [16][18]. - The testing involved a unique set of ten research-level mathematical questions, which are part of the 1st Proof project aimed at assessing AI capabilities in a research-like environment [45][49]. Group 3: Community Engagement and Feedback - The community has actively participated in validating the model's answers, with discussions highlighting the model's impressive advancements in mathematical reasoning [46][52]. - Experts have noted that the framework captures progress in both competition-level mathematics and research-oriented mathematical reasoning [47][48]. - The shift in evaluation paradigms is evident, moving from traditional test scores to real-world problem-solving assessments, which could lead to transformative changes in STEM research [49][51][54].

Core Viewpoint - The article discusses the launch of Qwen AI Slides, an AI-powered PPT generation tool that aims to simplify the process of creating presentations by automating content structure and visual design. Group 1: Product Features - Qwen AI Slides offers a comprehensive solution for generating presentations, including content structure and visual elements, catering to students and professionals alike [1]. - The tool supports three input methods: simple prompts, complex prompts, and document uploads, enhancing user flexibility [13]. - The AI's ability to generate infographics and visual timelines exceeded expectations, showcasing its advanced content generation capabilities [17][18]. Group 2: Performance Evaluation - The AI demonstrated strong semantic understanding, effectively breaking down complex prompts into coherent presentation structures [25]. - Text rendering was generally stable, with no significant deformation of characters, although some complex Chinese characters posed challenges [33][38]. - The visual design capabilities were assessed through a business report theme, where the AI successfully matched chart types to content and maintained a cohesive color scheme [42][44]. Group 3: Limitations and Recommendations - Despite its strengths, the AI's output occasionally contained minor flaws in layout and alignment, indicating that human intervention may still be necessary for fine-tuning [46][50]. - The AI lacks the ability to make incremental edits based on new prompts, requiring users to regenerate slides entirely for modifications [54]. - For users with high-quality presentation demands, using complex prompts is recommended to ensure better results [26].