![[98fc1f8a-f129-4879-a985-d6376489ed87.mp3]]

AI生成表情包成为新兴创业方向

简报：

• 人工智能技术正在被应用于表情包创作领域，通过算法自动生成个性化表情图像
• 部分创业团队开始尝试利用AI技术批量生产表情包，探索商业化可能性
• 该技术可大幅缩短表情包从创意到产出的周期，降低创作门槛
• 目前市场反馈显示AI生成表情包在社交媒体平台有一定接受度

相关链接：

• AI表情包生成器走红网络，创作者称效率提升10倍
• 表情包经济新赛道：人工智能如何改变内容创作
• 2025年全球表情包市场规模预计突破50亿美元

AI-Generated Emoji Packs Emerge as a New Entrepreneurial Trend

Newsletter:

• Artificial intelligence technology is being applied to the creation of emoji packs, automatically generating personalized emoji images through algorithms 🤖
• Some startup teams are exploring the use of AI to mass-produce emoji packs, investigating potential commercialization opportunities 💡
• This technology significantly shortens the cycle from concept to output for emoji packs, lowering the barrier to creation
• Current market feedback indicates that AI-generated emoji packs have gained some acceptance on social media platforms 👍

Related Links:

• AI Emoji Generator Goes Viral Online, Creators Claim 10x Efficiency Boost
• New Frontier in Emoji Economy: How AI is Transforming Content Creation
• Global Emoji Market Size Expected to Exceed $5 Billion by 2025

谷歌AlphaEvolve突破矩阵乘法56年效率基准，加速AI训练23%

简报：

• 谷歌DeepMind发布AlphaEvolve算法发现智能体，结合Gemini模型与进化策略，成功将4x4复值矩阵乘法运算从49次标量乘法优化至48次
• 该突破改进了1969年Strassen算法，是56年来首次效率提升，被类比为AlphaGo的"神之一手"
• AlphaEvolve已应用于谷歌计算生态系统：使Gemini训练速度提升23%，训练时间缩短1%；优化数据中心调度恢复0.7%全球计算资源；改进TPU芯片设计
• 系统采用"生成-评估-进化"循环机制，能跨越数百行代码发现复杂算法，在75%测试数学问题上达到最先进水平
• 华裔数学家陶哲轩参与项目，称其在高维参数空间优化中表现优于传统算法

相关链接：

• 打破56年数学铁律！谷歌AlphaEvolve自我进化实现算法效率狂飙
• Agent又暴风进化！谷歌发布AlphaEvolve：能发明新算法
• 破解百年数学难题，DeepMind发布超级编码智能体

Google AlphaEvolve Breaks 56-Year Matrix Multiplication Efficiency Benchmark, Accelerating AI Training by 23% 🚀

Brief:

• Google DeepMind unveils the AlphaEvolve algorithm discovery agent, combining the Gemini model with evolutionary strategies, successfully optimizing 4x4 complex matrix multiplication from 49 scalar multiplications to 48.
• This breakthrough improves upon the 1969 Strassen algorithm, marking the first efficiency gain in 56 years, and is likened to AlphaGo's legendary "divine move."
• AlphaEvolve has been integrated into Google's computing ecosystem: boosting Gemini training speed by 23%, reducing training time by 1%; optimizing data center scheduling to recover 0.7% of global computing resources; and enhancing TPU chip design.
• The system employs a "generate-evaluate-evolve" loop mechanism, capable of discovering complex algorithms across hundreds of lines of code, achieving state-of-the-art performance on 75% of tested mathematical problems.
• Chinese mathematician Terence Tao participated in the project, praising its superior performance in optimizing high-dimensional parameter spaces compared to traditional algorithms. 🌟

Related Links:

• Breaking a 56-Year Mathematical Barrier! Google AlphaEvolve Self-Evolves to Achieve Algorithm Efficiency Surge
• Agent Evolves Like a Storm! Google Releases AlphaEvolve: Capable of Inventing New Algorithms
• Solving Century-Old Math Problems, DeepMind Unveils Super Coding Agent 💡

OpenAI推出云端代码Agent Codex，支持自动化编程任务

简报：

• OpenAI发布云端代码Agent Codex，基于codex-1模型（o3模型的优化版本），可自动执行编写代码、修复错误、回答代码问题等任务
• Codex在独立云沙盒环境中运行，预加载用户代码库，支持读取/编辑文件、运行测试框架和类型检查器
• 任务完成时间1-30分钟，提供终端日志和测试输出作为可验证证据
• 目前向ChatGPT Pro、Enterprise和Team用户开放，可通过侧边栏的"Code"和"Ask"模式使用
• 早期合作伙伴包括Cisco、Temporal等公司，已应用于错误分析、测试覆盖率提升等场景

相关链接：

• OpenAI发布代码Agent：开启自动化编程时代，超强智能体！
• 深夜突袭！OpenAI的AI程序员上线，人类仅需点按钮
• Codex 实测：OpenAI 的云端 Coding Agent

OpenAI Launches Cloud-Based Code Agent Codex to Support Automated Programming Tasks

Newsletter:

• OpenAI has released the cloud-based Code Agent Codex, built on the codex-1 model (an optimized version of the o3 model), capable of automatically performing tasks such as writing code, fixing bugs, and answering coding questions. 💻
• Codex operates in an isolated cloud sandbox environment, preloaded with the user’s codebase, supporting file reading/editing, running test frameworks, and type checkers.
• Task completion time ranges from 1 to 30 minutes, with terminal logs and test outputs provided as verifiable evidence.
• Currently available to ChatGPT Pro, Enterprise, and Team users, accessible via the sidebar’s "Code" and "Ask" modes. 🚀
• Early partners include companies like Cisco and Temporal, with applications in bug analysis, improving test coverage, and more. 📈

Related Links:

• OpenAI Releases Code Agent: Ushering in the Era of Automated Programming with Super-Intelligent Agents!
• Late-Night Surprise! OpenAI’s AI Programmer Goes Live, Humans Just Need to Click a Button
• Codex in Action: OpenAI’s Cloud-Based Coding Agent

微软中国外包公司员工突遭裁员，赔偿方案引热议

简报：

• 微软外包公司微创软件（Wicresoft）因微软停止在华运营而裁员约2000人，员工清明节后返岗即失业
• 微软上海部分被裁员工获得高达N+8的赔偿，工作12年老员工可获20个月工资补偿
• 微创软件是微软在华首家合资公司，微软持股22.32%，此次裁员因微软全球战略调整
• 无锡公司员工反映裁员毫无预警，OA系统当天即无法登录，公司仅提供盒饭
• 微软2023财年全球累计裁员1.1万人，创公司历史纪录，主要涉及销售和客服部门

相关链接：

• 曝微软某外包公司的员工全部被裁员！员工：清明节回来就没工作了
• 幸福来的太突然，上海微软被裁员工喜提N+8赔偿
• 微软2023财年累计裁员1.1万人 刷新公司纪录

Microsoft China Outsourcing Company Employees Face Sudden Layoffs, Compensation Plan Sparks Discussion

Newsletter:

• Wicresoft, an outsourcing company for Microsoft, laid off approximately 2,000 employees due to Microsoft ceasing operations in China; employees returned after the Qingming Festival only to find themselves unemployed 😢
• Some laid-off Microsoft Shanghai employees received compensation as high as N+8, with veteran employees of 12 years eligible for 20 months' salary compensation 💰
• Wicresoft was Microsoft's first joint venture in China, with Microsoft holding a 22.32% stake; these layoffs are a result of Microsoft's global strategic adjustments
• Employees at the Wuxi branch reported the layoffs came without warning, with access to the OA system blocked on the same day and the company only providing boxed meals 🍱
• Microsoft laid off a total of 11,000 employees globally in the 2023 fiscal year, setting a company record, primarily affecting sales and customer service departments

Related Links:

• Microsoft Outsourcing Company Employees Reportedly All Laid Off! Employees: No Job After Qingming Festival
• Sudden Happiness: Shanghai Microsoft Laid-Off Employees Receive N+8 Compensation
• Microsoft Laid Off 11,000 Employees in 2023 Fiscal Year, Setting Company Record

小米自研3nm手机芯片玄戒O1即将发布，填补大陆先进芯片设计空白

简报：

• 小米集团创始人雷军宣布自主研发设计的3nm制程手机处理器芯片玄戒O1即将亮相
• 这是中国大陆首次实现3nm芯片设计突破，小米成为全球第四家可设计3nm手机SoC的企业（继苹果、高通、联发科后）
• 该芯片采用第二代3nm工艺制程，拥有190亿个晶体管，研发历时4年
• 截至2025年4月底，玄戒项目累计研发投入超135亿元，研发团队超2500人
• 小米计划5月22日举行战略新品发布会，将正式发布玄戒O1芯片

相关链接：

• 3nm芯片来了！我国半导体产业迎来好消息
• 国产品牌实现3nm芯片研发设计突破
• 雷军发长文官宣：小米3nm芯片来了

Xiaomi's Self-Developed 3nm Mobile Chip Xuanjie O1 Set to Launch, Filling the Gap in Advanced Chip Design in Mainland China

Newsletter:

• Lei Jun, founder of Xiaomi Group, announced that the self-developed 3nm process mobile processor chip, Xuanjie O1, is about to be unveiled.
• This marks the first 3nm chip design breakthrough in mainland China, making Xiaomi the fourth company globally capable of designing 3nm mobile SoCs (following Apple, Qualcomm, and MediaTek). 🚀
• The chip utilizes the second-generation 3nm process, featuring 19 billion transistors, and its development spanned 4 years.
• As of the end of April 2025, the Xuanjie project has accumulated over 13.5 billion RMB in R&D investment, with a team of more than 2,500 researchers. 💻
• Xiaomi plans to hold a strategic new product launch event on May 22, where the Xuanjie O1 chip will be officially released. 🎉

Related Links:

• 3nm Chip Arrives! Good News for China's Semiconductor Industry
• Domestic Brand Achieves Breakthrough in 3nm Chip R&D and Design
• Lei Jun Posts Long Article to Announce: Xiaomi's 3nm Chip is Here

华为nova 14 Ultra手机发布：4199元起售，搭载5000万像素RYYB长焦镜头

简报：

• 华为nova 14 Ultra手机今日（5月19日）正式发布，提供流光紫、浮光白、流光紫、浮光金四种配色
• 该机256GB昆仑玻璃版首发价4199元，512GB版4499元，1TB版4999元
• 影像系统方面搭载5000万像素RYYB大光圈潜望长焦镜头与红枫原色多光谱摄像头
• 采用麒麟9系处理器，配备6.81英寸LTPO三分区自适应刷新率屏幕（支持1Hz刷新率）
• 支持双向北斗卫星消息、天通卫星通信，全面搭载HarmonyOS 5系统
• 内置5300mAh以上容量电池，后置四摄系统包括50MP主摄和50MP潜望长焦

相关链接：

• 全球首发！4199元！华为新品官宣：5月19日，新机正式发布！
• 华为nova 14系列今天发布：麒麟9系芯片+Mate同款红枫镜头

Huawei nova 14 Ultra Smartphone Released: Starting at 4199 Yuan, Featuring 50MP RYYB Telephoto Lens

Brief:

• Huawei nova 14 Ultra smartphone was officially released today (May 19), available in four colors: Streamer Purple, Floating White, Streamer Purple, and Floating Gold 🌟
• The 256GB Kunlun Glass version starts at 4199 Yuan, the 512GB version at 4499 Yuan, and the 1TB version at 4999 Yuan
• The imaging system features a 50MP RYYB large-aperture periscope telephoto lens and a Red Maple primary color multi-spectrum camera 📸
• Powered by the Kirin 9 series processor, it comes with a 6.81-inch LTPO tri-zone adaptive refresh rate screen (supporting 1Hz refresh rate)
• Supports two-way Beidou satellite messaging and Tiantong satellite communication, fully equipped with HarmonyOS 5 system 🛰️
• Equipped with a battery capacity of over 5300mAh, the rear quad-camera system includes a 50MP main camera and a 50MP periscope telephoto lens

Related Links:

• Global Launch! 4199 Yuan! Huawei Announces New Product: New Phone Officially Released on May 19!
• Huawei nova 14 Series Released Today: Kirin 9 Series Chip + Mate’s Red Maple Lens

字节发布轻量级多模态模型Seed1.5-VL，60项基准测试中38项第一

简报：

• 字节跳动Seed团队发布视觉-语言多模态大模型Seed1.5-VL，具备更强的通用多模态理解和推理能力
• 该模型仅用20B激活参数和532M视觉编码器，在60个公开评测基准中38项取得SOTA表现
• 新增视频理解、多模态智能体能力，可精准识别图像元素并给出坐标，处理监控视频等复杂场景
• 推理成本显著降低，输入价格0.003元/千tokens，输出价格0.009元/千tokens
• 已在火山引擎开放API，开发者可调用构建AI视觉助手、巡检系统等应用

相关链接：

• 字节最强多模态模型登陆火山引擎！Seed1.5-VL靠20B激活参数狂揽38项SOTA
• 看图猜位置不输o3！字节发布Seed1.5-VL多模态推理模型，在60个主流基准测试中拿下38项第一
• 字节迭代多模态大模型 媲美谷歌新品

ByteDance Releases Lightweight Multimodal Model Seed1.5-VL, Tops 38 out of 60 Benchmarks

Newsletter:

• ByteDance's Seed team unveils the visual-language multimodal large model Seed1.5-VL, boasting enhanced general multimodal understanding and reasoning capabilities. 🌟
• With only 20B active parameters and a 532M visual encoder, the model achieves SOTA performance in 38 out of 60 publicly evaluated benchmarks.
• New features include video understanding and multimodal agent capabilities, enabling precise identification of image elements with coordinates and handling complex scenarios like surveillance video analysis. 📹
• Inference costs are significantly reduced, with an input price of 0.003 CNY per thousand tokens and an output price of 0.009 CNY per thousand tokens.
• The model is now available via API on Volcano Engine, allowing developers to build AI visual assistants, inspection systems, and other applications. 🚀

Related Links:

• ByteDance's Strongest Multimodal Model Lands on Volcano Engine! Seed1.5-VL Achieves 38 SOTA with 20B Active Parameters
• Image Location Guessing Rivals o3! ByteDance Releases Seed1.5-VL Multimodal Reasoning Model, Securing First Place in 38 of 60 Leading Benchmarks
• ByteDance Iterates Multimodal Large Model, Comparable to Google's Latest Product