社交媒体已经变成了「媒体」，而不是「社交」。 我们不再是一个传统意义上的社交网络了，所以我们也不是在垄断社交网络。 扎克伯格在出席 Meta 反垄断案庭审作证时， 试图通过这种说法来削弱 Meta 在社交网络领域的垄断指控。这是一种策略性地重新定义， 目的是让 Meta 看起来不像是控制了一个特定市场，而是参与了一个更广泛、竞争更激烈的 媒体平台市场。 上周一，美国联邦贸易委员会（FTC）指控 Meta 收购 Instagram 和 WhatsApp 以此非法垄断社交媒体市场的案件在华盛顿开庭。该反垄断案在 2021 年 曾因对于 「个人社交网络服务」 的市场定义过于宽泛而被驳回。 据纽约客报道，Meta 创始人马克· 扎克伯格 在反垄断诉讼庭审期间表示，如今的社交媒体平台已今非昔比，Meta 近年来关注的重点是 「娱乐、了解世 界及发现新鲜事物的整体概念」。社交媒体已经逐渐从「连接人与人」演变成更类似传统媒体的形态，充斥着名人制作的推广视频、评论员对新闻事件的 评论内容、流行文化的聚合片段等。换句话说， 社交媒体已经变成了「媒体」，而不是「社交」。 以下为纽约客的评论文章《Mark Zuckerberg S ...

其中，该行指出，公司预计截至2月国内生活类场景下付费渗透率达到5.2%，突破此前5%的目标，主要 系生成式AI对传统功能的重构，比如用AI去双下巴功能取代传统的手动推图，实现对体验和效果的更 好满足，从而促进订阅转化率的提升。该行认为，公司基于长期沉淀的人像修图场景数据，通过低频功 能迭代+高频体验升级的模式，有利于用户留存与回流维持在健康水平。公司预计25年生活类付费会员 增长有望提供订阅收入弹性，维持生活类产品10%的长期订阅渗透率目标。 该行续指，根据SensorTower，1Q以来，定位欧美日韩市场的Airbrush、BeautyPlus MAU较为稳定；AI 换装等功能在东南亚出圈，带动海外版Wink、美颜相机MAU环比大幅增长。受益于较高付费意愿和高 级产品定位，公司预计截至24年Airbrush付费率远高于国内，而其他出海产品还在积极扩张新市场阶 段，短期尚未激进变现，对比成熟市场，远期付费率还有较大提升空间。 智通财经APP获悉，美图公司(01357)涨近3%，截至发稿，涨2.74%，报4.88港元，成交额9258.87万港 元。 中金发布研报称，近期，该行邀请美图管理层在2H24业绩后与投资 ...

曾几何时，商汤科技、旷视科技、云从科技、依图科技并称为"AI 四小龙"，承载着AI行业的无限期 待。时光流转，"AI 六小龙""杭州七小龙"等名号层出不穷，"AI 四小龙"在时代的洪流中逐渐失色。 从财报数据来看，"AI 四小龙"曾经的辉煌已如过眼云烟。 商汤科技2024 年度财报显示，全年总营收 37.72 亿元，却伴随着 43.06 亿元的净亏损，自 2018 年至 2024 年累计亏损超 546 亿元；云从科技2024 年总营收约 3.98 亿元，归母净亏损达6.63亿元。 持续亏损"老大难" 被誉为"AI 四小龙"之一的云从科技，往昔头顶着明星企业的光环，承载着行业与资本的诸多期待。然 而，在高研发投入下，亏损却不断增加，面临着商业化的难题。 数据显示，2024年，云从科技实现营业总收入3.98亿元，同比减少36.60%；营业利润-6.49亿元，上年同 期为-6.55亿元；归属于母公司所有者的净利润-6.37亿元，上年同期为-6.43亿元；归属于母公司所有者 的扣除非经常性损益的净利润-6.63亿元，上年同期为-6.89亿元。 从更长时间跨度来看，自2017 年起至 2024 年，云从科技仿佛陷入了一 ...

Core Viewpoint - The report from CICC indicates an upward revision of Meitu's revenue forecast for 2025 by 2% to 4.3 billion yuan, driven by better-than-expected growth in paid users and subscription business, with adjusted net profit forecast maintained at 850 million yuan [1] Group 1: Subscription Revenue Growth - The company expects a more than 40% year-on-year increase in subscription revenue for 2025, driven by both payment rate and user growth [2] - The domestic life scenario payment penetration rate is projected to reach 5.2% by February, surpassing the previous target of 5%, largely due to generative AI enhancing traditional functions [2] - The company aims to maintain a long-term subscription penetration rate target of 10% for life products, supported by user retention strategies [2] Group 2: Overseas Market Expansion - The user base for overseas life scenarios is accelerating, with stable MAU for Airbrush and BeautyPlus in the US and Europe, and significant growth in Southeast Asia for AI dressing features [3] - The company anticipates that the payment rate for Airbrush will be significantly higher than in the domestic market by 2024, with other overseas products still in the expansion phase [3] Group 3: Productivity Scene Revenue Contribution - The number of paid subscribers for Meitu Design Studio is expected to reach 1.13 million in 2024, contributing to a revenue increase to 200 million yuan, doubling year-on-year [4] - The company plans to focus on enhancing product capabilities in core e-commerce sectors, with an expected 2 percentage point increase in domestic product penetration over the next two years [4] - The company is optimistic about the overseas productivity market and has launched two new productivity tools, Vmake and X-design, while monitoring user growth [4]

截至收盘，道琼斯工业平均指数上涨486.83点，报40093.40点，涨幅1.23%；标普500指数上涨 108.91点，报5484.77点，涨幅2.03%；纳斯达克综合指数涨幅达2.74%，报17166.04点。 标普500指数的11个主要板块中，除主要消费品板块小幅收跌外，其余全部上涨。其中，科技板块表 现最为强劲，收涨3.5%，领涨大盘。 2025.04. 25 本文字数：809，阅读时长大约1.5分钟 作者 | 第一财经 胡弋杰 周四，主要美股指数集体收高，科技板块涨幅居前。投资者在评估一系列最新贸易表态与就业市场数 据的同时，继续关注美联储的政策动向。 商品市场方面，COMEX黄金期货结算价上涨54.5美元，报每盎司3348.60美元，涨幅1.65%。 国际原油价格温和走高，纽约WTI 6月合约收涨0.84%，报每桶62.79美元；伦敦布伦特6月合约上涨 0.65%，报每桶66.55美元。 微信编辑 | 七三 推荐阅读 白宫慌了！将设工作组紧急处理对中国加征关税危机 热门中概股方面多数上涨，纳斯达克中国金龙指数收涨0.68%。个股方面，蔚来、富途控股涨幅超过 6%，拼多多、百度、网易、哔哩哔哩等涨超 ...

生成式AI 一、 OpenAI图像生成模型gpt-image-1，API向所有开发者开放 1. OpenAI发布新图像生成模型gpt-image-1及其API，支持图像生成、编辑和变体功能，每 张图成本低至0.02美元； 2. 模型支持自定义尺寸、质量、格式、压缩度和背景透明度，能结合世界知识生成更符合上 下文的高质量图像； 3. Adobe、Figma、Canva等多家企业已将该API集成到产品中，可应用于设计、电商、教育 等多个领域。 https://mp.weixin.qq.com/s/l6y2q2fZtAKje7fWyKW9kw 二、 微软AI同事来了，Researcher、Analyst等智能体强势登场 1. 微软推出AI智能体同事功能，主要包括研究员(Researcher)、分析师(Analyst)等智能体， 打造全新工作流程系统； 2. Microsoft 365 Copilot更新整合了网页、工作内容和Pages，支持全天候专家咨询、新型 工作流、综合搜索等功能； 3. 微软2025工作趋势报告预测，未来2-5年内所有公司都将转型为"前沿公司"，由人类和AI 智能体组成混合团队。 https: ...

Core Viewpoint - Open-source AI is rapidly becoming a core component for enterprises to build AI capabilities, drive innovation, and seek competitive advantages, moving beyond being a supplementary option [3][17]. Group 1: Current State of Open-source AI - Open-source AI technology has penetrated various levels of the AI tech stack, with over half of respondents utilizing open-source in data, models, and tools [4][5]. - The adoption of open-source technology is uneven across different areas, with lower rates in model modification and hosting/inference computing [6]. - The most commonly used models are "partially open," reflecting the current market landscape where many well-known models have limitations on data transparency or usage licenses [7]. Group 2: Value Perception and Trade-offs - Cost-effectiveness is a significant driver for adopting open-source AI, with 60% of respondents believing implementation costs are lower than proprietary solutions [8]. - Performance and ease of use are also critical factors, with a majority of users satisfied with their open-source AI models [8][9]. - However, proprietary tools are perceived to deliver value more quickly, with 48% of respondents favoring them for faster returns [9]. Group 3: Future Outlook and Risk Management - There is a strong expectation for growth in open-source AI usage, with 75% of respondents anticipating increased adoption in the coming years [11]. - A hybrid approach combining open-source and proprietary technologies is likely to emerge, as over 70% of respondents are open to using both [12]. - Key risks associated with open-source AI include cybersecurity, regulatory compliance, and intellectual property issues, with 62% of respondents expressing concerns about cybersecurity [13][14]. Group 4: Community Contribution and Strategic Integration - Only 13% of respondents reported contributing to open-source projects, indicating a need for greater community engagement [16]. - Companies should integrate open-source AI into their core strategies, balancing the benefits of open-source with the need for risk management and community involvement [17][18].

因为 OpenAI 30 亿美元的收购报价，Windsurf 成为近期最受关注的 AI 编程公司。 2021 年成立的 Windsurf（前身为 Codeium），最初是一家 ToB 的 GPU 虚拟化平台，并且已经实现了百万美元级别的收入。但在 见识到大模型的能力后，创始人 Varun Mohan 意识到，大模型让基础设施端的优势不再明显，应用端才是未来。 切入 AI 编程赛道后，一开始是作为 VS Code 的插件，到后来自己做 IDE，Windsurf 走上了一条和 Cursor 不同的 AI Coding 之 路。 今年 4 月，公司 ARR 收入约为 1 亿美元，相比一月份的 4000 万，收入翻倍。而在本轮筹集超过 2.43 亿美元的资金后，Windsurf 的估值达到了 28.5 亿美元。 为什么在有现金流之后还果断转向 AI 编程赛道，为什么要自己做 IDE，以及 Windsurf 和其他 AI 编程产品的差异在哪里？ 关于 这些问题，创始人 Varun Mohan 最近在接受科技播客 Lenny's Podcast 采访时，就这些问题进行了深聊。 TLDR Founder Park 正在搭建 ...

其他人对此表示赞同。"我们最近重组或重新激活了一个跨公司的 AI 团队，"Cadence 验证软件产 品管理高级部门主管 Matt Graham 说，"我们仍然需要基础引擎，工程师需要理解所有这些的要 求。但我们也需要这些总体工程团队。以前，这些可能是市场营销团队和产品工程团队——上市类 型的团队，如果我们以某种方式将它们结合起来使用，我们可以解决诸如低功耗混合之类的问题。 但我们越来越发现这实际上是一个工程问题，而不仅仅是一个上市解决方案。我们可能需要在工具 中构建特定的功能，或者在代码级别（而不仅仅是在脚本级别）将特定的流程拼接在一起，以实现 这些不同的解决方案。这不是一个完全统一的单一流程，但它是一个接一个地流动的。" 一个巨大的挑战是如何集成各种 AI 实现，这实际上可以在设计过程开始时收集的数据与芯片制造 前后显示的结果之间架起一座桥梁。 "我们的应用工程师团队和产品工程团队越来越开始构建这种跨职能的知识，"Graham 说，"我们 的客户也在寻找这类人才并组建这类团队。验证工程师非常擅长使用 UVM、SystemVerilog 和运 行各种调试工具来找到仿真过程中发现的逻辑错误的根本原因。但他们也 ...

Core Viewpoint - The article discusses how AI is fundamentally transforming the chip industry, particularly in the design, packaging, and manufacturing processes, with a focus on the integration of chiplets and the evolution of EDA tools [1][6][11]. Group 1: AI's Impact on Chip Design - AI is reshaping EDA (Electronic Design Automation) by enhancing the design possibilities and requiring a more integrated approach to chip specifications, verification, and manufacturing [1][3]. - The traditional silos in semiconductor design are breaking down, prompting a reorganization of design teams and their interactions with other teams [1][2]. - There is a growing need for cross-functional teams that combine expertise from various engineering disciplines to address complex design challenges [2][3]. Group 2: Challenges in AI Integration - Integrating various AI implementations poses significant challenges, particularly in bridging the gap between data collected during the design process and the results observed before and after chip manufacturing [2][6]. - The complexity of AI models necessitates trade-offs, such as balancing the prediction of component interactions with the reliability of control loops [2][9]. - As chip manufacturers begin to stack chips, the intricacies of interconnections increase, making the design process more complex than traditional 2D packaging [8][9]. Group 3: Industry Trends and Future Directions - The surge in interest in generative AI, particularly following the launch of ChatGPT, has led to substantial investments in high-performance AI architectures and data centers [6][11]. - The shift towards advanced packaging and multi-chip components is driven by the limitations of scaling single-plane chips, with a focus on improving yield and reusability of chiplets [6][8]. - The industry is witnessing a transition where packaging design is becoming a critical factor in the overall chip design process, reversing the traditional approach where it was often the final step [7][8]. Group 4: Concerns and Risks - There are concerns regarding the reliability of AI-driven processes, including issues related to hardware incompatibility, silent data errors, and security vulnerabilities in multi-chip systems [11]. - The black-box nature of many AI implementations limits traceability and raises questions about the predictability of outcomes in the semiconductor industry [11].