Healthcare 是美国最大的行业之一，支出占 GDP 的 17%，雇佣 1/10 的美国劳动力。它也极其低效，美国每年 4.5 万亿美元的医疗支出中，有高达 25%，也 就是 1.1 万亿美元被视为无效或可避免的浪费。在某些情况下，healthcare 从业者用于保险账单处理的时间成本可能占账单收入的 1/7。碎片化的系统、低 效的运营流程和人力密集的环节，是 AI Agent 的天然切入点。如果 AI 能切掉哪怕一小部分，就可能创造几千亿级别的新市场空间。 过去一年，GenAI 在医疗行业的渗透加速了。本文我们系统梳理了 Healthcare AI 的产业结构，重点聚焦了在哪些环节 AI 能够真正创造价值、并且拥有明 确预算，识别了值得优先关注和投资的细分方向和公司。 医疗行业的独特性决定了 AI 的扩散路径和传统软件或消费级应用并不同。目前 Healthcare AI 的高价值切入口，集中在 高频刚需、非临床环节 ，一方面是 前台任务 ，如提升医生效率的 Patient Copilot ( Case Study 包括 Abridge、Ambience、OpenEvidence )；另一方面是 后台基 ...

作者：Alex Singla、Alexander Sukharevsky、Lareina Yee、Michael Chui、Bryce Hall 企业如何完成生成式AI部署，由谁主导？ AI治理工作涉及构建一系列政策、流程和技术，以确保负责任地开发与部署AI系统。麦肯锡的调查分 析表明，由CEO亲自监督这项工作，是企业借助生成式AI提升财务表现的关键因素之一【1】。尤其在 大型企业中，CEO的直接参与对息税前利润（EBIT）的拉动效果最为显著。在已部署AI的企业中， 28%的受访者称CEO负责AI治理工作，但在年收入超5亿美元的大型企业中，这一比例略低。同时， 17%的受访者称董事会负责AI治理工作。整体来看，这项工作通常由多人共同负责，平均由两位领导牵 头。 AI的真正价值在于重塑企业运作方式。最新调查显示，在针对各类规模企业的25个要素里，重构工作 流程对于组织通过应用生成式AI实现息税前利润增长的作用最为显著。企业正逐步在AI部署的同时调 整流程。在已部署生成式AI的企业中，21%的受访者表示其组织已对部分工作流程进行了彻底重构。 集中管理AI部署的关键环节 生成式人工智能的发展，正推动企业着手构建相应 ...

Market Overview - The A-share market saw a positive start after the May Day holiday, with the Shanghai Composite Index rising by 1.13% to 3,316.11 points, and the Shenzhen Component Index increasing by 1.84% to 10,082.34 points[16] - The international oil price rebounded over 3% from a four-year low, with NYMEX crude oil rising 3.43% to $59.09 per barrel, driven by signs of a potential decline in U.S. production[27] Economic Indicators - During the May Day holiday, 314 million domestic trips were made in China, with total spending exceeding 180 billion RMB, reflecting an 8.0% year-on-year increase[16] - The U.S. trade deficit reached a record high in March, raising concerns about economic stability[6] Stock Market Performance - U.S. stock markets experienced declines, with the Dow Jones falling 389 points (0.95%) to 40,829 points, and the S&P 500 dropping 0.77% to 5,606 points[9] - European markets showed mixed results, with the UK FTSE 100 index rising for the 16th consecutive day, closing at 8,597 points, while the German DAX index fell 0.41% to 23,249 points[9] Commodity Trends - The commodity market outlook remains weak, with expectations of "oversupply" if OPEC+ continues to increase production, potentially leading to downward pressure on oil prices[6] - Gold is expected to outperform copper and oil in the commodity market, maintaining a strong position amid economic uncertainties[6] Currency Movements - The U.S. dollar index fell by 0.59% to 99.238, influenced by low U.S. Treasury yields and market expectations of a slower pace of interest rate cuts by the Federal Reserve[27] - The euro appreciated against the dollar, with the exchange rate at 1.137, reflecting a 0.5% increase[26] Fixed Income Market - The U.S. Treasury market showed strong demand, particularly for the 10-year Treasury auction, which had an indirect bid ratio of 91.1%, the highest since February 2023[29] - The yield on the 10-year U.S. Treasury note was reported at 4.29%, down 4.9 basis points from the previous day[29]

Core Insights - The AI industry continues to grow rapidly, with Fourth Paradigm successfully listing on the Hong Kong Stock Exchange after overcoming previous setbacks [1][3] - The company is focusing on customized AI solutions through a "platform + scenario" model and is expanding into consumer electronics with its Phancy brand [1][11] - Despite significant investments from top-tier firms, Fourth Paradigm faces challenges such as high R&D costs and ongoing losses [1][13] Company Performance - Fourth Paradigm's revenue grew from 942 million yuan in 2020 to 3.083 billion yuan in 2022, with Q1 2023 revenue at 644 million yuan [5][11] - The company reported losses of 750 million yuan, 1.802 billion yuan, 1.653 billion yuan, and 304 million yuan from 2020 to Q1 2023 [5][14] - R&D expenses accounted for a significant portion of revenue, with percentages of 57.3%, 55.3%, 48.8%, and 37.5% from 2020 to Q1 2023 [5][14] Market Position and Strategy - Fourth Paradigm is positioned as a leader in the Chinese intelligent decision-making market, maintaining the largest market share as of mid-2023 [12] - The company is diversifying its offerings beyond decision-making AI, launching generative AI models and entering the edge AI market [7][9] - The AI market in China is projected to grow from 225.5 billion yuan in 2022 to 691 billion yuan by 2027, with a compound annual growth rate of 25.1% [11] Investment and Valuation - Fourth Paradigm has attracted significant investment from firms like Sequoia Capital and Tencent, with its valuation reaching nearly 3 billion USD by 2021 [15] - The company's stock performance has been volatile post-IPO, with a decline of over 10% from its initial market value [15] - The company is viewed as undervalued compared to peers, with a price-to-sales ratio significantly lower than that of Palantir [11][15] Future Outlook - Fourth Paradigm aims to achieve profitability as it scales, with expectations that R&D expenses will decrease relative to revenue [6][11] - The company is optimistic about the potential of generative AI and its impact on productivity and efficiency across industries [7] - Despite the challenges, Fourth Paradigm's ongoing investments in R&D and market expansion position it for potential growth in the evolving AI landscape [1][6]

Group 1 - Microsoft's recent focus on artificial intelligence technology in the gaming sector has garnered significant attention, but not all of its game studios are required to adopt this technology [1] - Compulsion Games, the developer of the popular game "Midnight South," stated that the decision to use AI technology is entirely voluntary and not mandated by Microsoft [1][3] - The CEO of Compulsion Games emphasized the importance of handcrafted game content and artistic creation, indicating that the studio does not feel pressured to implement emerging technologies like AI [1][3] Group 2 - While AI technology may assist in certain areas such as storyboarding during the early creative phase, Compulsion Games will not incorporate AI into the actual game production process [3] - The "Muse" AI model developed by Microsoft has broad potential applications in game development, capable of generating high-quality game visuals and commands [3] - There is a diverse range of attitudes towards AI technology within the gaming industry, with some teams recognizing its potential to enhance efficiency and quality, while others prioritize the uniqueness and artistry of handcrafted content [3]

生成式AI 一、 刚刚，OpenAI放弃营利性转型！奥特曼：非营利组织继续掌控 1. OpenAI放弃完全营利性转型，将由非营利组织继续控制，同时营利性机构转为公益公司（PBC）； 2. 公司架构调整后取消利润上限制度，采用常规股权结构，非营利组织将成为PBC主要股东； 3. 承诺继续专注AGI发展造福人类使命，并计划开源部分高性能模型。 https://mp.weixin.qq.com/s/Z1bl0zfwNXeEcoDZFtpWmQ 二、 公开一切，优于DeepSeek-R1？英伟达开源Llama-Nemotron家族 1. 英伟达发布Llama-Nemotron开源模型家族，包含8B到253B三种规格，支持动态切换推理模式，遵循 开放商业许可； 2. LN-Ultra运用Puzzle框架和FFN融合技术优化部署效率，在推理性能和吞吐量上超越DeepSeek-R1； 3. 通过Qwen和DeepSeek-R1教师模型支持，结合多阶段训练和强化学习，全面提升模型推理与通用对话 能力。 https://mp.weixin.qq.com/s/Ofw7l6XPNNinXvFReGI3vw 三、 Grok 增加PD ...

◼ FY2025Q3 PBP、IC 和 MPC 业务均好于预期： 证券研究报告·海外公司点评·计算机 微软（MSFT） FY25Q3 业绩点评：业绩好于预期，继续看 好云业务快速增长 买入（维持） | Table_EPS] [盈利预测与估值 | FY2023A | FY2024A | FY2025E | FY2026E | FY2027E | | --- | --- | --- | --- | --- | --- | | 营业总收入（百万美元） | 211,915 | 245,122 | 278,960 | 314,772 | 355,067 | | 同比（%） | 6.88 | 15.67 | 13.80 | 12.84 | 12.80 | | 归母净利润（百万美元） | 72,361 | 88,136 | 99,117 | 117,414 | 135,259 | | 同比（%） | -0.52 | 21.80 | 12.46 | 18.46 | 15.20 | | EPS-最新摊薄（美元/股） | 9.73 | 11.86 | 13.33 | 15.79 | 18.19 | | P/E（现价&最新摊薄） ...

AI在工作：解锁全 球机遇 人工智能在全球商业格局中 G-P与一家第三方研究机构合作，对来自澳大利亚、英国 和美国的超过1500名商业高管进行了调查，以了解人工智 能在全球商业复杂性中的作用，这项技术如何推动增长和 创新，以及最终人工智能将如何重塑工作未来。 章节 01 了解当前格局 人工智能在全球商业领域的采用 第二章 在困境中导航并克服人工智能 挑战 第3节 拥抱利用人工智能实现全球增长的机会 13 g-p.com 2 与聊天室是互联网早期应用之一类似，ChatGPT已成为人们在提 及人工智能（AI）时最常想到的标准用例之一。 然而，与聊天室不能代表互联网的价值一样，ChatGPT也不能 代表人工智能能力的真正广度和深度。 人工智能正在以极快的速度发展，但人工智能带来变革并在全球 商业领域产生更大影响的机会仍然很多。 3 7 目录表 01 理解 当前格局 人工智能采用率在 全球商业 g-p.com AI在工作：解锁全球机遇 全球商业正仅是 触及人工智能的表面。 新一代人正以更响亮、更勇敢的观点和全球抱负欢迎这一变化 进入职场。我们将这种现象称为"全球一代"。他们渴望增长机会 、多样化的经历、与全球公司相关 ...

"恭喜！""当之无愧！" AISTATS官宣其获奖的推文下面，业界大佬齐聚，一片祝贺之声。 当初，这篇论文被AISTATS接收。 然而在谢赛宁本人的转发推文中，我们知道另一重内幕—— 衡宇 发自 凹非寺 量子位 | 公众号 QbitAI 谢赛宁十年前被NeurIPS （当时还叫NIPS） 拒收的论文，刚在今年获得了AISTATS 2025年度时间检验奖。 这篇论文就是《Deeply-Supervised Nets》 （DSN，深度监督网络） ，2014年9月挂上arXiv。 时间匆匆，十一年过去，属于是真·时间检验了。 它提出的中间层监督思想被谢赛宁后续作品REPA （Representation Alignment） 和U-REPA （U-Net Representation Alignment） 等继 承并发展，展示出从单一模型优化到跨模型知识迁移的演进。 而后两者在深度学习、扩散模型深化发展的这两年间，影响颇深。 这篇论文最初投稿给NeurIPS。虽然拿下8/8/7高分，但仍然被该顶会拒绝了。 他表示： 那次挫折一直萦绕在我心头，困扰着我…… 十一年前，拿到8/8/7高分却被拒 补充下背景信息—— 《D ...

Sander Dielman 自 2015 年加入 DeepMind 以来，参与了包括 WaveNet、AlphaGo、Imagen 3 和 Veo 在内的多个项目，涵盖深度学习、生成模型及表征学习 （Representation Learning）。 在这篇文章中，他将潜在变量比喻为「数据的精髓」—— 通过压缩复杂信息实现图像、语音等生成。他还深入对比变分自编码器（VAEs）、生成对抗网络 （GANs）和扩散模型，展示了潜在变量如何支持这些模型生成逼真内容。 选自 Sander Dieleman博客 机器之心编译 编辑：刘欣 上个月， GPT-4o 的图像生成功能爆火，掀起了以吉卜力风为代表的广泛讨论，生成式 AI 的热潮再次席卷网络。 而在这股浪潮背后，潜在空间（Latent Space）作为生成模型的核心驱动力，点燃了图像与视频创作的无限想象。 知名研究者 Andrej Karpathy 最近转发了一篇来自 Google DeepMind 研究科学家 Sander Dielman 的博客文章，探讨了生成模型（如图像、音频和视频生成模型）如 何通过利用潜在空间来提高生成效率和质量。 博客链接：https: ...