Group 1 - Beijing is becoming a strategic high ground in the AI large model field, with significant advancements in technology and a thriving ecosystem for innovation [1][4] - The emergence of AI unicorns like DeepSeek and the development of the "Wudao" model signify China's growing capabilities in AI, aiming to compete with the US by 2025 [4][5] - The AI landscape in China is rapidly evolving, with numerous "little dragons" and "little tigers" emerging, indicating a flourishing environment for AI startups [5][6] Group 2 - The development of AI models has shifted from "large model refining" to "refining large models," with DeepSeek's success serving as a strong signal of China's position in the global AI arena [5][20] - The establishment of the Zhiyuan Research Institute has played a crucial role in fostering AI talent and innovation, acting as a "angel investor" for top scholars in the field [11][22] - The AI industry is witnessing a trend towards more efficient and capable models, with a focus on achieving higher model density and performance [20][21] Group 3 - The journey towards Artificial General Intelligence (AGI) is seen as a long-term goal for AI entrepreneurs, requiring strategic planning and patience [17][19] - The local processing capabilities of edge models provide advantages in data protection and user privacy, making them appealing in various applications [19][20] - The success of DeepSeek highlights the importance of combining financial resources with visionary leadership in the AI startup ecosystem [21][22]

大家早上壕！ 先看海外要闻： 美股涨势消减，科技股打压大盘，标普惊险五连涨，纳指回落，收盘道指涨0.28%，标普500指数涨0.06%，纳斯达克跌0.1%。 跌超2%的英伟达止步四连涨，特斯拉盘中跌超4%后转涨，和苹果五连阳；公布财报警告关税影响后，芯片股恩智浦盘后一度跌超8%。 中概股总体表现亮眼，中概指数反弹至三周高位，热门中概股中，被花旗建议30天内关注利好催化剂后，蔚来汽车大涨超7%，理想汽车涨超 3%，阿里跌超1%。 美国财政部本季度借款实际预估规模不增反降，美债收益率创近两周新低。 美元指数回落，离岸人民币一度涨近200点、涨破7.29。 黄金反弹，期金一度涨近2%。原油回落，盘中转跌超2%，美油收创近两周新低。 西葡两国大范围停电，民众抢购物资以备不时之需，目前部分地区电力供应已恢复 国内重大事件汇总： 1、外交部重申中美未就关税问题磋商或谈判。 7、网传称：大厂"抢购"算力资源，消息称腾讯今年Q1从字节购入约20亿元GPU。阿里也在今年一季度DeepSeek爆红之后，向字节跳动下了GPU 订单。字节相关负责人回复称，以上为不实信息。（财经） 点评：中信建投指出，公共数据已经成为数据要素市场最重 ...

Core Viewpoint - The concept of "domestic substitution" is evolving from a perception of being a second-best option to a strategic necessity, reflecting advancements in Chinese technology and manufacturing capabilities [1][2]. Group 1: Domestic Substitution and Technological Advancements - Domestic substitution is not merely about replacing imported components but involves a comprehensive restructuring of the industrial chain, technology standards, and market rules [2]. - In the medical device sector, companies like United Imaging Healthcare have launched innovative products that have entered the global high-end market, showcasing that domestic products are not inferior but competitive [1]. - The Chinese electric vehicle industry has established a complete ecosystem from lithium mining to vehicle manufacturing, pushing global automotive standards towards "Chinese standards" [2]. Group 2: Impact of External Pressures - The U.S. technology blockade has inadvertently accelerated the pace of domestic substitution, compelling Chinese companies to innovate and compete globally [2]. - DeepSeek, a representative AI company in China, has emerged as a competitor to OpenAI, demonstrating that external pressures can catalyze technological innovation [2]. - Following supply chain disruptions, Huawei's HarmonyOS has been installed on over 1 billion devices, indicating significant progress in domestic technology [2]. Group 3: Challenges and Opportunities - Despite challenges such as weak foundational research and a shortage of high-end talent, the environment for innovation in China is improving [3]. - In 2024, China's total R&D expenditure is projected to reach 36,130 billion yuan, an 8.3% increase from the previous year, maintaining its position as the second-largest R&D investor globally [3]. - Over 570 Chinese industrial companies are among the global top 2,500 in R&D investment, highlighting the growing strength of domestic innovation [3].

Core Insights - Alibaba has launched Qwen 3, a series of AI models that claim to match or outperform leading models from Google and OpenAI [1] - The models are available for download under an open license from platforms like Hugging Face and GitHub, with sizes ranging from 0.6 billion to 235 billion parameters [2] - The emergence of models like Qwen 3 increases competitive pressure on American AI labs and has prompted U.S. policymakers to impose restrictions on Chinese AI companies' access to necessary chips [3] Model Features - Qwen 3 models are described as "hybrid," capable of reasoning through complex problems while also providing quick responses to simpler requests, allowing users to manage the "thinking budget" [4] - The models support 119 languages and were trained on a dataset of nearly 36 trillion tokens, which includes textbooks, question-answer pairs, and code snippets [5] - Qwen 3 shows significant performance improvements over its predecessor, Qwen 2, outperforming OpenAI's o3-mini and Google's Gemini 2.5 Pro on various benchmarks [6] Benchmark Performance - The largest model, Qwen-3-235B-A22B, achieved superior results on platforms like Codeforces and AIME, indicating its advanced problem-solving capabilities [6][10] - The public model Qwen3-32B remains competitive against several proprietary and open AI models, surpassing OpenAI's o1 model in accuracy benchmarks [10] Market Position and Availability - Qwen 3 is noted for its strong tool-calling capabilities and adherence to instructions, with availability through cloud providers like Fireworks AI and Hyperbolic [11] - Despite U.S. restrictions on chip sales to China, the development of state-of-the-art models like Qwen 3 suggests a growing domestic usage of advanced AI technologies in China [12]

Group 1 - DeepSeek's new model, DeepSeek R2, is expected to launch in early May and will reduce costs by 97% compared to GPT-4, utilizing Ascend cards for training [1] - DeepSeek R2 will feature a hybrid expert model (MoE) with a total parameter count of 1.2 trillion, doubling the parameters of DeepSeek R1, which had 671 billion [1] - The model aims to achieve breakthroughs in programming capabilities, multilingual reasoning, and higher accuracy at lower costs [1] Group 2 - Jin Modern is actively expanding its standardized, general-purpose software product business centered around an "AI low-code" development platform, having developed several standardized platform software products [2] - Haoyun Technology continues to invest in low-code technology research and development, with its low-code platform "Haoyida" deeply integrated with AI and IoT to customize AI agents for enterprises [2]

以下文章来源于凤凰网科技 ，作者凤凰网科技 凤凰网科技 . 凤凰科技频道官方账号，带你直击真相。 本文来自微信公众号： 凤凰网科技 （ID：ifeng_tech） ，作者：姜凡，编辑：董雨晴，题图来自：视觉中国 五月将至，中美科技巨头或将迎来新一轮巅峰对决。 先是在4月中旬，OpenAI一口气发布了GPT-4.1 o3、o4 mini系列模型；谷歌则拿出了Gemini 2.5 Flash Preview，一个混合推理模型；与谷歌同一天，豆 包在杭州巡展中正式发布了1.5·深度思考模型，在多模态上展现出了更强的实力。凤凰网科技从行业人士处了解到，阿里的下一代大模型Qwen3也将 于本月内发布。 混战之下，那股"神秘的东方力量"似乎也在悄悄准备着新的发布。 敏感的神经之下，一点蛛丝马迹都会被放大。昨日，全球最大AI开源社区Hugging Face首席执行官Clément Delangue在社交平台发布了一条耐人寻味 的动态。这条动态仅由三个眼睛的表情符号构成，并附上了DeepSeek团队在Hugging Face平台的官方资源库入口。 一、DeepSeek R2发布已进入倒计 时？ 近半个月来，有关"DeepSe ...

Group 1 - The event "2025 Li Shanyou Opening Class and Chaos AI Innovation Institute Opening Ceremony" gathered over 3,500 entrepreneurs and CEOs, marking a significant moment in Chinese business history focused on the theme of "AI's Dawn" [1] - Li Shanyou emphasized that the current era represents a historical turning point, where entrepreneurs must navigate the new battlefield of computing power and data to avoid being consumed by AI [1][4] - The Chaos AI Innovation Institute aims to transform entrepreneurs from "technology followers" to "mission-driven" leaders, aligning business expansion with long-term human values [4][12] Group 2 - OpenAI's evolution from an idealistic startup to a commercial giant illustrates the tension between technological ideals and business interests, highlighting the need for entrepreneurs to maintain their original mission amidst growth [7][8] - DeepSeek's innovative approach, achieving a training cost of $5.57 million—one-tenth of its competitors—demonstrates a new paradigm in Chinese AI development, emphasizing efficiency and transparency [10][11] - Li Shanyou posited that DeepSeek's significance extends beyond technology, challenging the stereotype that Chinese entrepreneurs only follow trends rather than define them [12][14] Group 3 - The second day of the event focused on "Human Dignity," with discussions on how to anchor entrepreneurial missions in the context of human civilization, using Elon Musk's projects as examples [14][15] - The concept of "flow" was introduced as a unique human advantage that AI cannot replicate, emphasizing the importance of creativity and personal talent in the entrepreneurial process [15][16] - The emergence of "collective intelligence" within organizations was highlighted, showcasing how collaborative environments can lead to extraordinary outcomes [16][18] Group 4 - The final day emphasized practical applications of the theories discussed, with a focus on transforming ideas into actionable strategies for business growth [18][20] - The "AI Landing Six-Dimensional Sword" framework was introduced to help entrepreneurs identify high-efficiency opportunities across various business functions [18][20] - The Chaos AI Innovation Institute aims to bridge the gap between theory and practice, providing ongoing support and resources for entrepreneurs beyond the initial three-day course [24][25] Group 5 - Li Shanyou's closing remarks underscored the importance of believing in entrepreneurs, innovators, and China's potential in the AI landscape, encouraging a mindset of resilience and ambition [28][31][33] - The event concluded with a collective affirmation of faith in the entrepreneurial spirit, positioning participants as torchbearers of change in the AI era [33]

作者｜冬梅 字节 Top Seed 启动 2026 届招聘，瞄准顶尖博士 4 月 27 日，字节跳动 Seed 在其官微上发布了一则招聘启示，宣布正式启动 2026 届 Top Seed 大模型顶尖人才校招计划， 研究课题包括大语言模型、机器学习算法和系统、多模态生成、多模态理解、语音等方向，基本覆盖大模型研究各个领域， 计划招募约 30 位顶尖应届博士。 值得一提的是，本届 Top Seed 强调不限专业背景，更关注研究潜力，希望寻找具有极强技术信仰与热情、具备出色研究能 力、富有好奇心和驱动力的年轻研究者。 值得注意的是，字节跳动在此次招聘启事中还透露了几位刚毕业的同学已经做出了一些有影响力的研究。 比如，Z 同学构建并开源了首个多语言代码修复基准 Multi-SWE-bench，在 SWE-bench 基础上，首次覆盖 Python 之外的 Java、TypeScript、C、C++、Go、Rust 和 JavaScript 七种编程语言，1632 个真实修复任务，是真正面向"全栈工程"的评测 基准，其数据均来自 GitHub issue，历时近一年构建，以尽可能准确测评和提高大模型高阶编程智能水平。 ...

本项目由北京大学物理学院朱华星老师、曹庆宏副院长统筹指导。基准设计、项目管理以及数据整合的主要工作由学生核心团队完成，核心成员包括仇是、郭绍 阳、宋卓洋、孙韫博、蔡则宇、卫家燊、罗天宇等。项目还得到了北京计算科学研究中心罗民兴院士和人工智能研究院张牧涵老师的鼎力支持。 PHYBench 项目汇聚了来自物理学院及兄弟院系的 200 余名学生，共同承担题目编写、审核及人类基准测试等工作。这支高水平的参与者团队中，包含至少 50 位 全国中学生物理竞赛金牌得主，更有亚洲物理奥赛和国际物理奥赛的金牌获得者。这场大规模、高质量的协作，不仅充分展现了北大学子深厚的学术功底和卓越 的组织协调能力，也为 PHYBench 产出高质量成果提供了坚实保障。 在大语言模型（LLMs）飞速发展的当下，模型的推理能力俨然成为模型能力的代名词。OpenAI 的 o 系列、DeepSeek R1 等前沿模型相继发布，这些大模型凭借强 化学习技术的助力，在许多科学评测基准上频频刷新纪录，甚至声称 "超越人类专家"。 但是，随着模型能力和评测基准的军备竞赛白热化， 越来越多的基准不得不转向生僻的知识点、或者抽象的数学竞赛题。 这些题目虽然能 ...

虽然没有公开具体功能，但在已安装该系统的中国车上，开关车窗、调节车内温度和座椅位置均 可通过车内显示屏，如同智能手机一样进行操作。如果实现车辆的智能手机化，车内的舒适度将 大幅提高。 丰田在其投放的中国专用EV上首次搭载了华为的鸿蒙系统，本田也将在中国专用EV系列上搭载 DeepSeek的服务。日本车企认为仅凭一己之力难以在智能化竞争激烈中国市场取胜，因此将吸 收中国的最尖端技术…… 丰田等日本汽车厂商将与中国的IT企业合作，投放中国市场专用的纯电动汽车(EV)。丰田在其操 作系统(OS)上采用了华为的系统。在中国，汽车智能化竞争激烈，仅凭一己之力难以取胜。日本 企业将吸收中国的最尖端技术，谋求生存。 "为了在中国提供存在需求的汽车，借助中国人的大脑和技术推进汽车开发不可或缺"，在上海市 举行的上海国际车展上，丰田的当地法人总经理李晖这样强调。 丰田在全球首次披露的纯电动轿车"bZ7"正是中国专用车。为了实现在驾驶席等处显示各种信息 的"智能座舱"，首次搭载了华为的鸿蒙"HarmonyOS"。 丰田在驾驶辅助方面也在与中国企业合作。将采用与自动驾驶新兴企业北京初速度科技 （Momenta）联合开发的先进驾驶辅 ...