Workflow
人工
icon
Search documents
ICML 2025 Oral!北大和腾讯优图破解AI生成图像检测泛化难题:正交子空间分解
机器之心· 2025-07-12 04:57
Core Viewpoint - The article discusses the advancements in AI-generated image detection, particularly focusing on the challenges of distinguishing between real and generated images, emphasizing the complexity beyond simple binary classification [1][5][31]. Group 1: Research Findings - A study conducted by researchers from Peking University and Tencent Youtu Lab reveals that AI-generated image detection is more complex than a straightforward "real-fake" binary classification [1][5]. - The research introduces a new solution based on orthogonal subspace decomposition, which enhances the generalization ability of detection models from "memorization" to "understanding" [1][3][31]. - The study highlights the asymmetry in the binary classification of AI-generated images, where models tend to overfit to fixed fake patterns in the training set, limiting their generalization capabilities [5][7][9]. Group 2: Methodology - The proposed method utilizes Singular Value Decomposition (SVD) to create two orthogonal subspaces: one for retaining pre-trained knowledge and another for learning new AIGI-related knowledge [16][18]. - The approach involves freezing the principal components while fine-tuning the residual components, allowing the model to learn fake detection information while preserving original knowledge [17][18][25]. - The method's effectiveness is validated through attention map visualizations, demonstrating the orthogonality between retained semantic information and learned fake features [25][27]. Group 3: Experimental Results - The proposed method shows improved generalization performance in tasks such as DeepFake face detection and AIGC full-image generation detection, outperforming traditional methods [21][23]. - Quantitative analysis indicates that traditional methods lead to a significant reduction in the effective dimensionality of the feature space, while the new method maintains a high-rank feature space [10][14][22]. Group 4: Insights and Future Directions - The article emphasizes that the relationship between real and fake images is hierarchical rather than independent, suggesting that understanding this relationship is crucial for effective detection [29][30]. - The research proposes that the orthogonal decomposition framework can be applied to other AI tasks, providing a new paradigm for balancing existing knowledge with adaptability in new domains [31].
密室逃脱成AI新考场,通关率不足50%,暴露空间推理短板丨清华ICCV25
量子位· 2025-07-12 04:57
清华大学团队 投稿 量子位 | 公众号 QbitAI 近年来,多模态大模型(MLLMs)发展迅猛,从看图说话到视频理解,似乎无所不能。 但你是否想过:它们真的"看懂"并"想通"了吗? 模型在面对复杂的、多步骤的视觉推理任务时,能否像人类一样推理和决策? 为评估多模态大模型在视觉环境中,完成复杂任务推理的能力。清华大学团队受密室逃脱游戏启发,提出 EscapeCraft:一个3D密室逃脱环境 ,让大模型在3D密室中通过自由探索寻找道具,解锁出口。 该论文目前已入选ICCV 2025。 EscapeCraft 环境 沉浸式互动环境,灵感源自密室逃脱 研究团队打造了可自动生成、灵活配置的 3D 场景 EscapeCraft,模型在里面自由行动:找钥匙、开箱 子、解密码、逃出房间……其中每一步都需整合视觉、空间、逻辑等多模态信息。 任务可扩展,应用无限可能 EscapeCraft以逃出房间为最终目的,重点评测逃脱过程中的探索和决策行为、推理路径等。支持不同房 间风格、道具链长度与难度组合,还可扩展到问答、逻辑推理、叙述重建等任务。它是一个 高度灵活、 可持续迭代的通用评测平台 ,也可以为未来的智能体、多模态推理、强化 ...
广西南宁出台多项措施 打造开放活力的AI人才生态环境
Huan Qiu Wang Zi Xun· 2025-07-12 04:30
Core Viewpoint - The Guangxi Nanning Human Resources and Social Security Bureau has introduced ten measures to support the development of the artificial intelligence (AI) industry, aiming to establish Nanning as a hub for AI innovation and talent cultivation in the ASEAN region [1][3]. Group 1: Talent Development Initiatives - The ten measures include large-scale vocational skills training to enhance the cultivation of AI-skilled talents, particularly in areas such as AI trainers and generative AI [1]. - Nanning will implement the "Skills Illuminate the Future" training initiative, focusing on AI application skills and establishing an AI industry talent cultivation alliance to promote industry-education integration [1][3]. - The city plans to host AI vocational skills competitions to encourage learning and training, with winners receiving professional skill certificates and rewards [3]. Group 2: Employment and Recruitment Efforts - Nanning has launched a public "Skills Night School" to provide practical courses on intelligent office software, facilitating AI empowerment across various industries and promoting employment and entrepreneurship [3][4]. - An online recruitment event titled "Smart Nanning · AI Empowering Employment" was held, featuring job opportunities in AI fields, with some positions offering monthly salaries of up to 30,000 RMB [4]. Group 3: Ecosystem and Policy Support - The city is creating a supportive ecosystem for AI talent by selecting leading AI companies, universities, and industry associations to establish a "Nanning Talent Home" for diverse and professional talent services [4]. - Nanning is leveraging its policies for returning overseas students to attract outstanding AI teams and talents, providing policy support and streamlined professional services [4].
阿里通义千问,重大更新!三大亮点
Zheng Quan Shi Bao· 2025-07-12 04:09
Core Insights - Alibaba's Tongyi Qianwen team announced significant updates to its AI chat product Qwen Chat, enhancing user interaction and adding practical features [1][2] Group 1: Product Updates - Users can now start conversations directly from the Tongyi Qianwen homepage, improving accessibility and immediacy [2] - Qwen Chat integrates multiple functionalities, including "in-depth research," "image generation," "web development," "deep thinking," and "search," allowing users to generate high-quality images from text descriptions and assist front-end engineers in coding [2][3] - A new desktop client has been introduced, enabling one-click access to the Model Context Protocol (MCP) and facilitating cross-application calls and automated task execution [2][3] Group 2: Competitive Positioning - Alibaba's Tongyi Qianwen is recognized as the largest open-source model globally, with a 23% market share in China's AI infrastructure (AIIaaS), surpassing the combined share of the second and third players [4] - The company reported that its "cloud + AI" strategy has become a new growth engine, with Alibaba Cloud achieving a revenue of 1180.28 billion yuan in the 2025 fiscal year, marking an 11% year-on-year increase [4] - AI-related product revenues have seen triple-digit growth for seven consecutive quarters, indicating strong demand in the market [4] Group 3: Investment and Future Plans - Alibaba plans to invest 380 billion yuan in AI infrastructure over the next three years, exceeding its total tech investments from the past decade [5] - The company announced a plan to issue zero-coupon exchangeable bonds worth approximately 12 billion Hong Kong dollars to fund cloud computing infrastructure and support international e-commerce development [5]
阿里通义千问,重大更新!三大亮点
证券时报· 2025-07-12 03:56
Core Viewpoint - Alibaba's Tongyi Qianwen team has made significant updates to its AI chat product Qwen Chat, enhancing user interaction and adding multiple practical features, aiming to improve product usability and integration within the AI platform [1][4]. Group 1: Product Updates - Users can now start conversations directly from the Tongyi Qianwen homepage, reducing usage costs and enhancing accessibility [3]. - Qwen Chat integrates multiple functionalities, including "in-depth research," "image generation," "web development," "deep thinking," and "search," allowing users to generate high-quality images from text descriptions and assist front-end engineers in coding [3]. - A new desktop client has been introduced, enabling cross-application calls and automated task execution, which enhances work efficiency by bridging different AI models and external tools [3]. Group 2: Competitive Positioning - The Qwen model family has been continuously updated, with the latest Qwen3 model outperforming top global models in various benchmark tests, showcasing its competitive edge [4]. - Alibaba is addressing the "product strength" gap of its large models by enhancing usability and user perception, indicating a strategic intent to build a unified AI platform that is user-friendly and feature-rich [4]. Group 3: Market Presence and Financial Performance - Tongyi Qianwen is recognized as the largest open-source model globally, with Alibaba Cloud holding a 23% market share in China's AI infrastructure (AIIaaS) market, surpassing the combined share of its closest competitors [5]. - The "cloud + AI" strategy has become a new growth engine for Alibaba, with the cloud intelligence group achieving a revenue of 1180.28 billion yuan in the 2025 fiscal year, marking an 11% year-on-year increase [5]. - AI-related product revenues have seen triple-digit growth for seven consecutive quarters, indicating strong demand and market penetration in traditional vertical industries [5]. Group 4: Investment and Future Plans - Alibaba plans to invest 380 billion yuan in AI infrastructure over the next three years, exceeding its total technology investment over the past decade [6]. - The company has announced the issuance of zero-coupon exchangeable bonds to fund cloud computing infrastructure and support international expansion [6].
锲而不舍落实中央八项规定精神丨上海浦东、陕西西安等地 坚持查在细处、改在实处,推动学习教育见行见效
Yang Guang Wang· 2025-07-12 03:44
Group 1 - Shanghai Pudong and Xi'an, Shaanxi are implementing the central government's eight regulations, focusing on problem-oriented approaches and categorized guidance to ensure effective learning and education [1][2] - In Shanghai Pudong, nearly 700 enterprise service specialists have been assigned to help businesses understand and apply policies, resulting in significant support for companies, including an AI firm that successfully entered overseas markets with their assistance [1] - A centralized office mechanism has been established in Shanghai Pudong to address issues related to investment promotion, attracting investment, and enterprise services, having visited 4,245 enterprises and resolved 612 issues to date [1] Group 2 - Xi'an is addressing prominent issues in the business environment through extensive outreach, selecting 1,000 enterprises to monitor key indicators related to policy implementation and regulatory enforcement, with monthly reviews to rectify issues [2] - A complaint and supervision platform for optimizing the business environment has been launched in Xi'an, creating a dynamic problem list to ensure accountability and resolution of issues [2] - Xi'an has processed 873 enterprise complaints and resolved 174 issues since the start of the learning initiative, with a focus on addressing urgent problems raised by the public [3]
多项全国首个!2025具身智能生态大会发布最新科技成果
Guo Ji Jin Rong Bao· 2025-07-12 03:24
Group 1: Core Insights - Embodied intelligence is recognized as a strategic frontier in the AI field, driving industrial transformation and reshaping economic development [1] - The 2025 Embodied Intelligence Ecological Conference in Chongqing gathered top experts and representatives to discuss the integration of embodied intelligence with financial services, particularly in enhancing elderly financial services [1][4] - The conference showcased various intelligent technologies, including wheeled robots and humanoid robots, emphasizing the dynamic role of AI in financial technology [1] Group 2: Industry Developments - The Chongqing Financial Regulatory Bureau has implemented a series of policies to support technological innovation in finance, enhancing the efficiency of financial resource allocation and credit accessibility [2] - By the end of 2024, the robot industry chain in Chongqing's Liangjiang New Area is expected to exceed 3 billion yuan, with 23 robot companies contributing to a developing industrial ecosystem [1][2] Group 3: Elderly Financial Services - With over 300 million people aged 60 and above in China, the country is entering a moderately aging society, necessitating advancements in elderly care technology and financial services [4] - The integration of blockchain and embodied intelligence in elderly finance is highlighted as a significant challenge for the financial industry, with proposals for seven pillars to enhance elderly financial services [4] Group 4: Technological Innovations - The conference announced the establishment of the IEEE P3707 international standard for the application of embodied AI in the elderly care sector, covering critical evaluation dimensions such as functionality and risk governance [8] - A national initiative for AI patent open licensing was launched, aimed at facilitating the transformation of AI innovations into practical applications [10]
刚刚,OpenAI想收购的Windsurf,被谷歌DeepMind抢走了核心团队
机器之心· 2025-07-12 02:11
加入 DeepMind 的 Windsurf CEO Varun Mohan 与联合创始人 Douglas Chen 也发布了一份声明,其中表示:「我们很高兴能与 Windsurf 团队的部分成员一起加入谷 歌 DeepMind。我们为 Windsurf 过去四年取得的成就感到自豪,并期待看到它与他们世界一流的团队一起前进,开启新的发展阶段。」 至于达成这笔交易的具体金额,目前还没有人透露。作为参考,此前有报道称,OpenAI 将斥资 30 亿美元收购 Windsurf。 那么,OpenAI 与 Windsurf 的交易究竟出了什么问题? 机器之心报道 编辑:Panda 就在所有人还在惊叹于月之暗面开源 Kimi K2 模型的同时,谷歌 DeepMind 却宣布截胡了 OpenAI 原本打算收购的 Windsurf。 诺贝尔奖得主、DeepMind 联创兼 CEO Demis Hassabis 与 DeepMind CTO Koray Kavukcuoglu 在 上向 Windsurf CEO Varun Mohan、联合创始人 Douglas Chen 以及 部分加入 DeepMind 团队的研发人员表示了 ...
谷歌据悉斥资24亿美元收购AI初创公司Windsurf资产和人员,此前消息称OpenAI以30亿美元收购Windsurf的计划告吹。(彭博)
news flash· 2025-07-12 01:57
谷歌据悉斥资24亿美元收购AI初创公司Windsurf资产和人员,此前消息称OpenAI以30亿美元收购 Windsurf的计划告吹。(彭博) ...
奥特曼30亿刀收购案黄了!谷歌迅速出手:Windsurf核心团队打包带走
量子位· 2025-07-12 01:49
鱼羊 发自 凹非寺 量子位 | 公众号 QbitAI 事情变得有意思了起来。 OpenAI豪掷30亿美元收购AI编程初创公司Windsurf的事,黄了。 并且谷歌已经迅速出手—— 把包括Windsurf CEO、联合创始人在内的核心团队一整个打包带走 ,上演硅谷经典" 雇佣式收购 "。 港真,网友都有点猝不及防了:这是什么鬼热闹。 毕竟OpenAI这笔收购案闹得轰轰烈烈,Windsurf作为一个基础模型靠大厂的AI编程工具,还因此惨遭Claude断供,API直接不给用。 这下可好,整个团队彻底投入谷歌Gemini怀抱。 Google DeepMind一把手哈萨比斯亲自出面表示欢迎: 推文中透露,此番Windsurf团队加入谷歌,将参与到Gemini项目之中,负责Google DeepMind的编程Agent开发。 OpenAI为何出局? 原因之一,在于与OpenAI关系日渐微妙的微软。 根据OpenAI和微软的协议,微软可以访问OpenAI的所有知识产权。 但桥豆麻袋,微软自己手握GitHub Copilot,正正经经是Windsurf的竞品。因此对这事儿OpenAI的态度是: 尴尬的情况使得收购案陷入了僵局 ...