图片来源： Relevance AI 随着 AI Agent 在工作场所日益普及，个人可能会与之紧密协作。波士顿咨询集团报告指出，未来五 年内， AI Agent 市场预计将以 45% 的复合年增长率扩张。 如同人类员工一样， AI Agent 可通过入职培训学习各类岗位技能，获取公司信息与业务背景，并融 入工作流程。此外，不同于传统自动化工具， AI Agent 具备持续优化与适应操作模式的潜力。 总部位于旧金山和悉尼的初创企业 Relevance AI 致力于开发 AI Agent " 操作系统 " ，帮助企业构建 AI Agent 团队，近日完成由 Bessemer Venture Partners 领投的 2400 万美元 B 轮融资。现有投资方 King River Capital 、 Insight Partners 和 Peak XV 继续跟投，使公司融资总额达 3700 万美元。本轮 估值未予披露。 此次融资距该公司完成 A 轮融资约一年半。 Relevance 表示其增长迅猛，仅 2025 年 1 月就有 4 万个 AI Agent 在其平台注册。客户包括 Qualified 、 Activ ...

图片来源： Cursor AI Anysphere 成立于 2022 年，由四位20多岁的 MIT 数学与计算机科学专业学生共同创办。今年1月，公司曾以 25亿美元估值 完成1.05亿美元融资，当时投资 方亦包括 Thrive 和 Andreessen Horowitz。 此次估值大幅提升，主要得益于其年度经常性收入（ARR）的快速增长：截至4月，ARR 已达到 2亿美元 ，使其成为软件行业增长最快的公司之一。 不过，如此激进的估值提升也再度引发部分投资者对 AI 公司估值可持续性的担忧，尤其是在近期二级市场波动背景下。 此前 OpenAI 的估值在3月飙升至 2600 亿美元 ，软银还承诺将牵头对其进行 400 亿美元 投资。 与此同时，两家由 OpenAI 前高管创立的初创公司也在筹备大额融资：Ilya Sutskever 创办的 Safe Superintelligence（SSI）已获得 300 亿美元估值。 Mira Murati 创立的 Thinking Machines Lab 正计划以 100 亿美元估值 融资 20 亿美元 。 两家公司目前尚未推出正式产品。 Cursor 因其 AI 编程工 ...

Core Insights - Recraft, a startup behind a competitive image model, has raised $30 million in Series B funding led by Accel, following a successful $12 million Series A round in 2024 [1][2] - The company has surpassed $5 million in annual recurring revenue (ARR) and has reached 4 million users [1] - Recraft's model, known as "red_panda," excelled in the Artificial Analysis benchmark, outperforming established competitors like OpenAI's DALL-E and Midjourney [1] Company Overview - Recraft builds its own AI model from scratch, competing with other image generators such as Midjourney, Adobe Firefly, and Stable Diffusion [2] - The AI is particularly effective in generating brand-specific images, allowing for precise logo placement and the creation of marketing materials that adhere to existing brand guidelines [2] - The company aims to position itself closer to design tools like Canva, which also features an AI generator for brand promotion [2] Leadership and Background - Anna Veronika Dorogush, the founder and CEO of Recraft, has a background in machine learning from Yandex and experience at Google and Microsoft [2] - Dorogush's previous career as a model and her academic background in mathematics and computer science have influenced her approach to building the AI model [2] - She emphasizes the importance of excellence in key tasks for startup success, particularly in model development [3]

Core Insights - KoBold Metals is revolutionizing mineral resource exploration using artificial intelligence, making the process more efficient, precise, and scalable [2][3] - The company invests over $100 million annually across 70 projects on four continents, focusing on metals essential for batteries and AI-driven businesses, such as lithium and copper [3][4] Group 1: Company Operations - KoBold's operations focus on mineral exploration rather than mining, as exploration offers greater value creation and economic benefits [5][6] - The company employs a comprehensive technology stack that includes sensors for data collection, data systems for integration, and predictive models to enhance exploration capabilities [4][6][15] - KoBold's exploration projects are primarily in North America, Europe, Australia, and Africa, with a strong emphasis on copper, lithium, and nickel [12][13] Group 2: Data Utilization - The company utilizes a vast array of geological data, including satellite imagery, airborne surveys, and historical geological reports, to identify potential mineral deposits [7][8][10] - KoBold combines structured and unstructured data to improve predictive modeling, allowing for better decision-making in exploration [15][19] - The integration of various data types enhances the company's ability to make informed predictions about mineral locations [19][20] Group 3: Economic Viability - The economic viability of a mining project is determined by its mineral concentration and scale, with high-grade deposits offering significantly lower operational costs [14][21] - KoBold aims to control the average discovery cost per successful project between $50 million and $100 million, which is a competitive target in the industry [22][23] - The company believes that the real scarcity lies in high-quality mineral deposits rather than funding, as good projects always attract buyers [23][24] Group 4: Industry Challenges - The exploration success rate in the industry has decreased significantly over the past 30 years, with current estimates suggesting a success rate of around 0.5% [21][22] - Regulatory challenges and community relations are critical factors in the feasibility of mining projects, requiring companies to establish strong local partnerships [11][12] - Despite the challenges, there are still underexplored regions with significant potential for mineral discovery, particularly in areas like Zambia and the Democratic Republic of the Congo [24][25] Group 5: Philosophical Approach - KoBold emphasizes a philosophical approach to exploration, focusing on making clear, falsifiable predictions and embracing uncertainty in data collection [34][35] - The company's exploration philosophy includes maintaining multiple hypotheses and using data to reduce uncertainty, which is central to its operational strategy [35][36] - The integration of philosophical thinking into business practices helps guide decision-making and enhances the company's predictive capabilities [36][37]

近日，国内首家增程式倾转旋翼 eVTOL货运无人机制造商「蓝霄航空」宣布完成天使++轮融资 ， 本轮由初心资本领投，老股东联想之星、戈壁创投超额 跟投，澜松资本担任长期独家财务顾问。 至此，蓝霄在半年内完成天使轮、天使 +轮及天使++轮三轮融资，累计融资额达近亿元。这家成立仅半年的企 业，以"军品研发体系+汽车供应链"的颠覆性模式，创下国内低空经济领域早期项目融资速度和金额新纪录，成为2025年低空经济赛道超级现象级企业。 动力系统局限：纯电 eVTOL航程普遍低于150公里，难以覆盖城际物流需求（平均运输距离800公里）； LX-1 效果图 1 行业拐点：低空经济爆发前的窗口期 在中央空管委正式批复《国家低空经济示范区建设方案》、六大试点城市启动低空改革试点的政策背景下，中国低空经济正从概念验证迈向规模化商业应 用。摩根士丹利预测， 2040年全球eVTOL市场规模将突破1.3万亿美元，其中货运场景占比超60%。然而，当前行业面临三大核心瓶颈： 适航认证风险：载人 eVTOL需满足FAA/EASA/CAAC最严苛适航标准，取证周期长达5-8年； 成本控制难题：主流机型吨公里成本高昂，是传统货运直升机的 3倍 ...

图片来源： Anthropic 英伟达发言人声明中补充道："中国拥有全球半数 AI 研究人员，在 AI 技术栈各层级都具备顶尖人 才。美国不可能通过操纵监管机构来赢得 AI 竞赛。" 出口限制将冲击英伟达的全球收入流。该公司近期表示，针对在中国销售的 H20 AI 芯片实施的新许 可要求，可能导致其 2026 财年第一季度损失 55 亿美元收入。 " Anthropic 坚持其最近提交的公开声明，"一位 Anthropic 发言人向 TechCrunch 表示。 参考资料 Anthropic 于上周 进一步强化了对美国商务部"人工智能扩散框架"的支持，该框架将从 5 月 15 日起 实施全面的 AI 芯片出口限制。 https://techcrunch.com/2025/05/01/nvidia-takes-aim-at-anthropics-support-of-chip-export-controls/ 次日，英伟达对即将实施的管控措施给出了截然不同的回应， 英伟达对 Anthropic 支持芯片出口管制 措施提出异议。 英伟达发言人向 CNBC 表示："美国企业应专注于创新并迎接挑战，而非编造大型、笨重 ...

Core Insights - Meta predicts that its generative AI products will generate between $2 billion to $3 billion in revenue by 2025, and this figure could rise to between $460 billion to $1.4 trillion by 2035 [1] - The company has signed revenue-sharing agreements with firms that host its open-source Llama model collection and has launched an API for customizing and evaluating Llama models [2] - Meta's budget for generative AI is projected to exceed $900 million in 2024 and may surpass $1 billion in 2025, excluding infrastructure costs [3] Group 1 - Meta's generative AI revenue forecast for 2025 is between $2 billion to $3 billion, with a long-term projection of $460 billion to $1.4 trillion by 2035 [1] - The company has launched an API for its Llama model and signed revenue-sharing agreements with hosting companies [2] - Meta's investment in generative AI is substantial, with a budget exceeding $900 million in 2024 and potentially over $1 billion in 2025 [3] Group 2 - Meta's planned capital expenditure for 2025 is between $60 billion to $80 billion, primarily for building large-scale data centers [3] - The company has discussed spending over $200 million to acquire training data for Llama, with around $100 million allocated for book purchases [3] - Meta has faced allegations of unauthorized use of copyrighted materials for training its AI models, which it disputes [4]

图片来源： Sesame 官网 Z Highlights 01 告别屏幕： Sesame 开启语音交互新时代 过去几十年来，人机交互的每一次革新都试图抹平数字与现实之间的鸿沟。从早期的键盘、鼠标到如今的触摸屏，人们不断追求更自然、更直观的交互体 验。然而，智能手机横空出世重塑交互逻辑的十几年来，我们被困在了玻璃屏幕筑成的围墙中，所有的操作都依赖于十指在键盘与屏幕之间穿梭。 下一个交互逻辑革命的突破口在哪儿？ —— 语音，这个根植于人类文明基因的沟通方式，理应成为下一代交互范式的破局点。然而语音在人机交互领域却 显得有些滞后，除了简单的功能性指令，我们很难体验到如同真人对话般的流畅与自然。 人工智能的飞速发展为语音交互带来了新的曙光，当强大的 AI 算法与精巧的硬件设备完美结合，语音的生成理解能力突飞猛进为真正自然的人机交互界面 揭开了神秘的面纱。在这股革命浪潮下， Sasame 公司于 2022 年正式成立，怀揣着 " 让 AI 听懂人心，让沟通回归自然 " 的愿景，致力于打造一种全新 的、自然流程的人机交互体验 ，解决现有语音交互中理解不准确、沟通不流畅、应用场景受限等问题， 通过可穿戴设备提供高品质音频， ...

此次融资距平台正式上线仅一年，印证了创始人 Steven Wang 此前向 TechCrunch 透露的扩张计划。 这位23岁的哈佛辍学生打造的 App 下载量已突破100万次，其核心模式是 允许用户一键复制顶尖交易 者的投资策略 。 "让平民投资者享有顶级资源"在一份新闻稿中，Steven Wang如此阐述愿景："超级富豪长期独占专 业基金经理资源，如今Dub要将这种优势带给普通投资者。" 这位自称"二年级开始炒股"的创始人在 近期TechCrunch播客中直言："我们的目标是成为投资界的TikTok——用极简操作消弭专业门槛。" Dub的病毒式传播得益于其特色功能：用户可订阅实时追踪美国政要公开持仓的"政治组合包"，例如 众议院前议长 南希·佩洛西 披露的交易动向。 利用政治人物持仓追踪引爆社交裂变，该平台采用订阅 制收费，月费10美元。 图片来源： Unsplash 主打"重塑Z世代投资方式"的跟单交易平台 Dub 今 日宣布完成3000万美元A轮融资，使其累计融资额 攀升至4700万美元。 这笔交易由 Notable Capital 和 Neo （ 本刊近期深度报道的加密基金）共同领投，跟投方包括 S ...

图片来源： Anthropic 据彭博社报道，苹果与 Anthropic 正联手打造一款"氛围编程"软件平台，该平台将利用生成式 AI 为 程序员编写、编辑和测试代码。 为加速 AI 布局，苹果似乎正在构建合作伙伴网络。 OpenAI 的 ChatGPT 已为其 Apple Intelligence 功能提供支持，苹果表示未来可能加入谷歌 Gemini 作为备选方案。 如今 Anthropic 或将协助苹果内 部代码编写。 在开发者群体中， Anthropic 公司最新的 Claude 模型尤其受到编程任务的青睐，特别是在 Cursor 和 Windsurf 这类氛围编程平台上。 参考资料 https://techcrunch.com/2025/05/02/apple-and-anthropic-reportedly-partner-to-build-an-ai-coding-platform/ 我们正在招募新一期的实习生 -----------END----------- 报道指出， iPhone 制造商计划内部推出该软件，但尚未决定是否公开上市 。该系统是苹果编程软件 Xcode 的新版本，并依赖于 An ...