近日，有网络安全研究人员标记出三个恶意的 npm（Node.js 包管理器）软件包，这些软件包的攻击 目标是一款颇受欢迎的由 AI 驱动的源代码编辑器 Cursor，且针对的是苹果 macOS 版本用户。 迄今 为止，这三个软件包的下载量总共已超过 3200 次。 软件供应链安全公司 Socket 的研究人员 Kirill Boychenko 表示："这些软件包伪装成提供'最便宜的 Cursor API'的开发者工具，窃取用户凭据，从由威胁行为者控制的基础设施中获取有效加密负载， 覆盖 Cursor 的 main.js 文件，并禁用自动更新以保持其持续性。" 整理 | 华卫 Cursor 用户被"劫持"全过程 有问题的软件包如下所列：sw-cur （2,771 次下载）、sw-cur1 （307 次下载） 和 aiide-cur （163 下载）。值得注意的是，目前这三个软件包仍可以继续从 npm 注册表下载。 其中，"aiide-cur "于今年 2 月 14 日首次发布，是由一个名为"aiide"的用户上传的，其 npm 库被描述 为"用于配置 macOS 版本的光标编辑器的命令行工具"。另外两个软件包则 ...

作者丨 Gergely Orosz & Addy Osmani 译者丨明知山 策划丨褚杏娟 可以肯定的是，生成式 AI 将继续改变我们开发软件的方式。 回顾 2022 年 11 月，ChatGPT 首次问世，这是大语言模型（LLM）开始被广泛运用的开端。尽管 LLM 的构建方式出人意料地简单，但它们在各个领域都取得了令人印象深刻的结果。编写代码无疑 是它们的强项之一。这并不令人感到惊讶，因为： 去年，我们的 AI 工具使用情况调查发现，大约 75% 的开发者使用某种 AI 工具进行与软件工程相关 的工作。然而，我们似乎仍处于工具创新周期的早期阶段，而更复杂的方法，如软件工程 AI 智能 体，很可能成为 2025 年创新的核心。 主流媒体对软件工程行业的描绘越来越戏剧化。3 月份，Business Insider 报道了"软件工程师越来越 接近 AI 是否会让他们失业的真相"；9 月份，福布斯抛出疑问："软件工程师是否正在变得过时？"尽 管这类文章广为传播，但它们多出自非软件工程师之手，这些作者既不使用 AI 工具，也不了解这些 新型 GenAI 编码工具的效率（以及局限性）。 那么我们能从 GenAI 工具重 ...

5 月 9 日晚间，阿里巴巴创始人马云现身杭州阿里总部，身着白色 T 恤并佩戴"风清扬"工牌，专门前 往了代表阿里创业精神的复刻版"湖畔小屋"，并鼓励员工坚持创业精神，持续创新。从 2024 年底至 今，马云已多次在杭州露面。据悉，去年 12 月 8 日，马云亮相蚂蚁园区，并在支付宝和蚂蚁集团二 十周年活动现场致辞，着重提及了对于 AI 的思考。 整理 | 傅宇琪、褚杏娟 消息人士称马云回归是绝不可能；雷军：过去 1 个月是创办小米以来最艰难时间；零一万物联创戴宗 宏离职创业；阿里通义应用视觉团队负责人薄列峰离职；国行版苹果 AI 合作细节：百度技术占比仅 35%；苹果 App Store 佣金去年破百亿，四年翻倍引争议；饿了么租宇树人形机器人在街头宣传"闪 购"；特朗普政府拟取消拜登时代的 AI 芯片限制；OpenAI 任命新 CEO，重组计划被迫改变；淘宝闪 购上线，节后第一天就崩了 行业热点 消息人士称马云回归是绝不可能的，并且马云是创始人也从来没离开过 据贝壳财经，5 月 10 日消息，从多位阿里巴巴内部人士处获悉，阿里巴巴已全面打通了内网（内部 论坛）权限，并调整员工跨业务流动机制、启动工牌焕新。 ...

Core Viewpoint - Meta aims to compete directly with OpenAI by launching a consumer-facing AI chatbot application and a developer API for its Llama model, promoting an open-source AI ecosystem that challenges closed AI providers like OpenAI [1][5][6]. Group 1: Meta AI Application - The Meta AI application is designed to provide personalized responses based on user preferences and interactions, integrating image generation and editing features [1][3]. - The application supports both voice and text interactions, including full-duplex voice communication, and is currently available in the U.S. and Canada [3]. - An "Explore Feed" feature allows users to share and discover how others are using AI, potentially amplifying trends in generative AI [3]. Group 2: Llama API - The Llama API is positioned as a challenge to OpenAI's API business, allowing developers to connect applications to the Llama model with minimal code [5]. - Meta offers a limited free trial of the Llama API, emphasizing that models built on it remain the property of the developers and are not locked to Meta's servers [5][6]. Group 3: Open Source Strategy - Meta's strategy appears to focus on strengthening the open-source model ecosystem while limiting the growth of proprietary AI models like those from OpenAI [6][7]. - The company has reported 1.2 billion downloads of its Llama models, with around 1 billion users utilizing the Meta AI assistant [8]. Group 4: Discussion on AI Development - A dialogue between Mark Zuckerberg and Satya Nadella highlighted the importance of open-source models and the potential for AI to significantly enhance productivity across various sectors [19][27]. - Nadella emphasized the need for a new production factor to address real-world challenges, drawing parallels to the economic growth during the Industrial Revolution [27][28]. Group 5: Distillation Factory Concept - The "distillation factory" concept was discussed as a means to create smaller, more efficient models from larger ones, facilitating easier access for developers [30][32]. - Both companies expressed optimism about the future of AI development and the role of developers in transforming potential into reality [36][37].

Core Viewpoint - The article discusses the significant impact of large models on recommendation systems, emphasizing that these models have already generated tangible benefits in the industry rather than focusing on future possibilities or academic discussions [1]. Group 1: Impact of Large Models on Recommendation Systems - Large models have transformed the way knowledge is learned, shifting from a closed system reliant on internal data to an open system that integrates vast external knowledge [4]. - The structure of large models, typically based on transformer architecture, differs fundamentally from traditional recommendation models, which raises questions about whether they can redefine the recommendation paradigm [5]. - Large models have the potential to create a "new world" by enabling personalized content generation, moving beyond mere recommendations to directly creating tailored content for users [6]. Group 2: Knowledge Input Comparison - A comparison highlights that large models draw knowledge from an open world, while traditional systems rely on internal user behavior data, creating a complementary relationship [7]. - Large models possess advantages in knowledge quantity and embedding quality over traditional knowledge graph methods, suggesting they are the optimal solution for knowledge input in recommendation systems [8]. Group 3: Implementation Strategies - Two primary methods for integrating large model knowledge into recommendation systems are identified: generating embeddings from large language models (LLMs) and producing text tokens for input [10][11]. - The integration of multi-modal features through large models allows for a more comprehensive representation of item content, enhancing recommendation capabilities [13][15]. Group 4: Evolution of Recommendation Models - The exploration of large models in recommendation systems has progressed through three stages, from initial toy models to more industrialized solutions that significantly improve business metrics [20][24]. - Meta's generative recommendation model (GR) exemplifies a successful application of large models, achieving a 12.4% increase in core business metrics by shifting the focus from click-through rate prediction to predicting user behavior [24][26]. Group 5: Content Generation and Future Directions - The article posits that the most profound impact of large models on recommendation systems lies in the personalized generation of content, integrating AI creators into the recommendation process [28][29]. - Current AI-generated content still requires human input, but the potential for fully autonomous content generation based on user feedback is highlighted as a future direction [41][43]. Group 6: Industry Insights and Recommendations - The search and recommendation industry is viewed as continuously evolving, with the integration of large models presenting new growth opportunities rather than a downturn [45]. - The article suggests that the key to success in the next phase of recommendation systems lies in the joint innovation and optimization of algorithms, engineering, and large models [46].

作者 | Tina、核子可乐 一款号称拥有 2200 万下载量的 AI 助手，评分却低至冰冷的 2.3 分！面对激烈竞争，JetBrains 选择了直接批量删除用户评论和反馈， 于是这款问题缠身的 AI 助手，又一次将这家知名开发工具公司推向舆论的风口浪尖。 2200 万下载 AI 助手为何口碑崩塌？ 尽管 JetBrains 的工具深受开发者喜爱，但该公司在 AI 辅助方面却落后了。 JetBarins 于 2023 年 12 月发布一款 AI 助手插件，旨在帮助程序员编写代码。这家总部位于捷克的软件开发工具厂商当时表 示，"JetBrains AI Assistant 与 GitHub Copilot 类似，但与 JetBrains 的开发环境（IDE）、代码编辑器和其他产品深度集成。" 如今这款 AI 助手插件下载量已经超过 2200 万次，但评分仅为 2.3 分。满分 5 分，其中还有大量的 1 分评价。 近期，用户注意到部分负面评论遭到删除。 一位用户抱怨称，"我之前的评论被无缘无故删除了。JetBrains 似乎在清理负面反馈，这摧毁了我对这家公司的信任和信心。他们不再 重视客户的声音。我仍然会打 ...

Core Viewpoint - Alibaba's new technology "ZeroSearch" significantly reduces the cost and complexity of training AI systems for information retrieval, eliminating the need for expensive commercial search engine APIs [1][2][14]. Summary by Sections Technology Overview - ZeroSearch is a reinforcement learning framework that allows large language models (LLMs) to develop advanced search capabilities through simulation, outperforming models based on real search engines while incurring zero API costs [2][3]. - The technology is compatible with various model series, including Qwen-2.5 and LLaMA-3.2, and does not require a separate supervised preheating phase [2][3]. Performance Metrics - In comprehensive experiments across seven question-answer datasets, ZeroSearch's performance matched or exceeded that of models trained with real search engines [3][5]. - A 3 billion parameter LLM can achieve search capabilities comparable to Google, while a 14 billion parameter module can surpass Google's performance [3][5]. Cost Efficiency - Training using Google search via SerpAPI for approximately 64,000 queries costs around $586.70, while using a 14 billion parameter simulated LLM on four A100 GPUs costs only $70.80, representing an 88% reduction in costs [7][8]. Methodology - ZeroSearch begins with a lightweight supervised fine-tuning process that transforms LLMs into retrieval modules capable of generating relevant and irrelevant documents in response to queries [9][11]. - The system employs a course-based learning deployment mechanism, gradually increasing the difficulty of generated documents to simulate challenging retrieval scenarios [11][12]. Implications for AI Development - ZeroSearch represents a significant shift in AI training methods, enabling AI systems to improve without relying on external tools like search engines [14][15]. - This technology creates a more equitable competitive environment for small AI companies and startups by drastically lowering the entry barrier associated with high API costs [14][15].

Core Viewpoint - Databricks is in negotiations to acquire Neon, an open-source database startup, for approximately $1 billion, which may exceed this amount when including employee retention incentives. The deal is seen as a strategic move to enhance Databricks' AI capabilities and infrastructure [1][16]. Group 1: Company Overview - Neon is a four-year-old open-source database company founded by Nikita Shamgunov, Heikki Linnakangas, and Stas Kelvich, focusing on PostgreSQL [2][3]. - The current CEO, Shamgunov, has a strong background in computer science and has previously contributed to SQL Server at Microsoft and co-founded MemSQL (now SingleStore) [5][6]. - The company aims to create a PostgreSQL variant suitable for AI applications, allowing customers to pay for database usage on demand, with a focus on efficiency for AI agents [11][12]. Group 2: Technology and Features - Neon employs a serverless architecture that separates storage and compute, allowing for automatic scaling based on workload demands [7][8]. - The technology includes features like copy-on-write for checkpointing and time-point recovery, as well as connection pooling to enhance performance [8][9]. - Neon supports vector data storage and utilizes HNSW indexing for efficient high-dimensional vector searches, making it valuable for natural language processing tasks [11][12]. Group 3: Investment and Financials - Neon has raised over $130 million in funding, including a recent $46 million round led by Menlo VC, bringing its total funding to approximately $104 million [14]. - The company previously received a $25 million strategic investment from Microsoft's M12, enhancing its collaboration with Azure [13][14]. Group 4: Databricks' Strategic Moves - Databricks, founded in 2013, has shifted its focus towards AI, acquiring companies like MosaicML for $1.3 billion to bolster its AI capabilities [16][17]. - The company has been actively enhancing its platform through various product developments and acquisitions, including the launch of Databricks Apps for building customized AI applications [17][18]. - Databricks is reportedly facing challenges in its transition to AI, with some industry insiders expressing concerns about its current direction and operational efficiency [20].

大模型如何真正驱动企业核心业务提效？客服、财务、营销三大场景的 AI 革命已拉开帷幕！华为云 AI 应用首席架构师郑岩，携手蚂蚁集团高级技术专家杨浩、明略科技高级技术总监吴昊宇，聚焦"场 景探索 - 技术落地 - 未来展望"，与你探讨提效策略。 直播介绍 直播时间 5 月 9 日 20:00-21:30 直播主题 财务·客服·营销，大模型如何驱动业务提效 直播嘉宾 主持人 ：郑岩 华为云 AI 应用首席架构师 嘉宾 ： 直播亮点 杨浩 蚂蚁集团 / 高级技术专家 吴昊宇 明略科技 / 高级技术总监 实战场景剖析：精准评估落地价值，量化"价值锚点"。 技术落地秘籍：模型选型、评测设计与 RAG 应用深度优化。 未来展望：AI Native 智能体特质及组织"超能力"布局。 如何看直播？ 扫描下图海报 【二维码】 ，或戳直播预约按钮，预约 InfoQ 视频号直播。 如何向讲师提问？ 文末留言写下问题，讲师会在直播中为你解答。 ...

作者 | 罗燕珊 策划 | AICon 全球人工智能开发与应用大会 在 AI 开发者社区颇受欢迎的 "MCP 应用市场" MCP.so，你用过吗？ 随着智能体生态持续升温，AI 应用的标准化与可扩展性越来越受到开发者关注。其中，MCP 协议 （Model Context Protocol）自 2024 年 11 月由 Anthropic 推出以来，在 AI 开发者和工具社区中引 起了广泛讨论。它被视为一种开放标准，旨在简化 AI 模型与外部工具和数据源的集成过程。 而 MCP.so 这个目前收录超 10000 个 MCP Server 、支持网页直接调用 AI 工具、集成 Chat 能力的 全球最大 MCP 应用市场 ，其实来自一位中国独立开发者 —— 艾逗笔（idoubi）。 艾逗笔，前腾讯高级工程师、微信后台开发，现独立开发者。他作为全栈开发，技术涉猎面广，当前 专注 AI 应用出海，独立打造了多款产品，其中就包括最近热度较高的 MCP.so： 近期，随着 MCP 协议的热度不断提升， MCP.so 也迎来访问量激增。对此， InfoQ 对艾逗笔进行了 一次简短的采访： InfoQ：MCP.so 近期流量 ...