点击下方 卡片 ，关注" 具身智能 之心 "公众号 作者丨具身智能之心 编辑丨具身智能之心 本文只做学术分享，如有侵权，联系删文 >> 点击进入→ 具身智能之心 技术交流群 更多干货，欢迎加入国内首个具身智能全栈学习社区 ： 具身智能之心知识星球 (戳我) ， 这里包含所有你想要 的。 最近很多同学询问π0、π0.5、A0都是啥？用在什么地方？能实现哪些功能？方法论有啥不同？前面 刚开始听到这些，也一头雾水，今天为大家梳理下。 π₀模型结构 原文：π0: A Vision-Language-Action Flow Model for General Robot Control π₀的核心架构基于 预训练视觉语言模型（VLM） 和 Flow Matching 技术，具体包含以下关键组件： VLM backbone 动作专家（Action Expert） 跨具身训练（Cross-Embodiment Training） 整合 7 种机器人、68 项任务、超 10,000 小时数据（含开源 OXE 数据集），通过权重调整处理不 同机器人的动作空间差异（如零填充低维动作向量）。 训练流程 基于 PaliGemma V ...

点击下方 卡片 ，关注" 具身智能 之心 "公众号 作者丨具身智能之心 编辑丨具身智能之心 本文只做学术分享，如有侵权，联系删文 >> 点击进入→ 具身智能之心 技术交流群 更多干货，欢迎加入国内首个具身智能全栈学习社区 ： 具身智能之心知识星球 (戳我) ， 这里包含所有你想要 的。 最近很多同学询问π0、π0.5、A0都是啥？用在什么地方？能实现哪些功能？方法论有啥不同？前面 刚开始听到这些，也一头雾水，今天为大家梳理下。 π₀模型结构 原文：π0: A Vision-Language-Action Flow Model for General Robot Control π₀的核心架构基于 预训练视觉语言模型（VLM） 和 Flow Matching 技术，具体包含以下关键组件： VLM backbone 动作专家（Action Expert） 跨具身训练（Cross-Embodiment Training） 整合 7 种机器人、68 项任务、超 10,000 小时数据（含开源 OXE 数据集），通过权重调整处理不 同机器人的动作空间差异（如零填充低维动作向量）。 训练流程 基于 PaliGemma V ...

Core Viewpoint - The article emphasizes the importance of systematic thinking training for children's development, highlighting the educational approach of Huohua Siwei in enhancing logical thinking and comprehensive quality improvement for youth [1][2]. Group 1: Educational Philosophy - Huohua Siwei adheres to the educational philosophy of "learning through fun," focusing on youth thinking training and comprehensive quality enhancement since its establishment [1]. - The company offers a variety of products, including logical thinking, Chinese literacy, programming, Go, chess, and public speaking [1]. Group 2: Teaching Methodology - The company employs a "teaching according to aptitude" and "layered education" approach, tailoring learning content to match each child's developmental level [2]. - Interactive participation is emphasized in Huohua Siwei's courses, with activities like "Thinking Sports Meeting" and "Little Teacher Videos" being particularly favored by students [2]. Group 3: Impact on Development - The courses not only enhance children's logical and expressive abilities but also improve parent-child communication and social integration skills [2]. - Systematic thinking training is believed to lay a solid foundation for children's future academic and career development, benefiting them throughout their lives [2].

Core Viewpoint - Huawei announced the release of the Pangu Model 5.5 at the Huawei Developer Conference 2025, highlighting significant upgrades in predictive capabilities through a novel triplet transformer architecture [1] Group 1: Model Features - The Pangu Model 5.5 utilizes a unique triplet transformer architecture for unified pre-training, allowing for the integration of diverse data types from various industries [1] - The model can efficiently process and pre-train tabular data, time-series data from equipment logs, and image data from product inspections within the same framework [1] - This approach is expected to significantly enhance the accuracy of predictive models and improve generalization across different industries and scenarios [1]

Core Viewpoint - The article discusses the differences between precise interventions in psychological counseling and universal guidance in public communication, highlighting the importance of tailored approaches for individual psychological issues while advocating for more accessible language in public discourse [1][4]. Group 1: Psychological Counseling - Psychological counseling aims to help individuals, like the case of a 12-year-old boy, establish healthy psychological boundaries and express their needs [1]. - Counselors use precise language and Socratic questioning to challenge irrational beliefs and help clients differentiate between real consequences and imagined threats [2]. - Experimental behavior training is employed to encourage clients to practice expressing their needs in a safe environment, thereby reshaping their understanding of self-expression [2]. Group 2: Public Communication - Directly applying counseling language in public spaces can lead to misunderstandings and negative outcomes, such as creating crises in parent-child relationships [3]. - Public communication should utilize universal language that is empathetic and relatable, helping parents understand their children's emotional needs without labeling them negatively [3]. - The article emphasizes the need for public discourse to focus on basic psychological knowledge and positive guidance, avoiding absolute suggestions and allowing for flexibility [4].

本文由半导体产业纵横（ID：ICVIEWS）综合 国产GPU"四小龙"冲击上市，摩尔线程率先完成上市辅导。 中国证监会官网显示，国产GPU"四小龙"之一的摩尔线程已于6月10日率先完成上市辅导，目前进入"辅 导验收"阶段。 | 全国一体化在线政务服务平台 | | | | | | | | | | --- | --- | --- | --- | --- | --- | --- | --- | --- | | | 中国证券监督管理委员会网上办事服务平台(试运行) | | | | | | | | | 公开发行辅导公示 | | 니아있었 | 储导机构 | 餐家时间 国际状态 | | 最出肌肉 | 报告买型 | 擬告标設 | | II 北京 河北 山西 | 川 天津 | 鄉分液程智能科技《北班》 股份有限公司 | 中国证券股份有限公司 | 2024-11-12 辅导治农 | 北京证监局 辅导工作完成报告 关于雕尔成程智能- | | | | 2024年11月，摩尔线程在北京证监局办理辅导备案登记，由中信证券股份有限公司（下称"中信证券"） 开展辅导。上市辅导是企业IPO前的必经阶段，由辅导机构协助企业进行全面规范、整改。 ...

Core Viewpoint - The article discusses the development of LucaOne, a generalized biological foundation model that can simultaneously understand and process nucleic acids (DNA and RNA) and protein sequences, marking a significant advancement in the field of life sciences [4][26]. Group 1: Introduction to LucaOne - LucaOne is the world's first foundational model capable of unifying the understanding of nucleic acids and protein sequences, likened to a "DeepSeek" for life sciences [4]. - The model was pre-trained on sequences from 169,861 species, showcasing its ability to comprehend key biological principles such as the translation of DNA into proteins [4][16]. Group 2: Technical Aspects of LucaOne - The model utilizes a vocabulary of 39 "characters" to encode nucleotides and amino acids, allowing it to read both nucleic acids and proteins [13]. - It employs semi-supervised learning, integrating known biological annotations to enhance its understanding [14]. - LucaOne has 1.8 billion parameters and has been trained on 36.95 billion biological sequence "words," enabling it to extract deep, universal patterns from nucleic acid and protein sequences [16]. Group 3: Performance and Capabilities - LucaOne demonstrated an impressive ability to understand the central dogma of molecular biology without explicit instruction, outperforming specialized models in tasks involving DNA and protein sequence matching [18]. - The model excels in generating embeddings that accurately capture the biological significance of sequences, outperforming other models in clustering similar sequences [19]. - It has shown strong performance across seven challenging bioinformatics tasks, including species classification and protein stability prediction, often using simpler downstream networks compared to specialized models [20][24]. Group 4: Significance and Future Outlook - LucaOne provides a unified framework for understanding the two core molecular carriers of life, breaking down barriers between different molecular types [26]. - The model exemplifies the potential of foundational models in bioinformatics, allowing researchers to develop various biological computational tools efficiently [26]. - It paves the way for deeper and more automated analysis of complex biological systems, such as gene regulatory networks and disease mechanisms [26].

Core Insights - The article highlights the significant contributions of Chinese scholars in the field of research, particularly in the publication of papers in the prestigious journal Nature, with 9 out of 19 papers authored by Chinese researchers [2][3][5][7][9][11][13][15][17][18]. Group 1: Research Contributions - On June 18, 2025, a paper titled "Strategies for climate-resilient global wind and solar power systems" was published by Tsinghua University professors Zhang Qiang and Song Dan as co-corresponding authors [2]. - A paper titled "Unsupervised pretraining in biological neural networks" was authored by postdoctoral researcher Zhong Lin from the Howard Hughes Medical Institute [3]. - The National University of Singapore's Liu Xiaogang and Xiamen University's Liang Liang published a paper on "Optical nonlinearities in excess of 500 through sublattice reconstruction" [5]. - A study titled "Bimodal centromeres in pentaploid dogroses shed light on their unique meiosis" was published by M. Zhang from the Max Planck Institute for Plant Breeding Research [7]. - A paper titled "Kupffer cell programming by maternal obesity triggers fatty liver disease" was authored by postdoctoral researcher Huang Hao from Bonn University [9]. - A research paper on "Allosteric modulation and biased signaling at free fatty acid receptor 2" was published by Professor Zhang Cheng from the University of Pittsburgh [11]. - A study on "Machine-learning design of ductile FeNiCoAlTa alloys with high strength" was co-authored by professors Ma En, Sun Jun, and Zhang Jinyu from Xi'an Jiaotong University [13]. - A paper titled "Single-cell transcriptomic and chromatin dynamics of the human brain in PTSD" was published by Jing Zhang from the University of California, Irvine [15]. - A study titled "R9AP is a common receptor for EBV infection in epithelial cells and B cells" was co-authored by professors Zeng Muxing and Zhong Qian from Sun Yat-sen University [17]. - A paper on "Cryo-EM structure of a natural RNA nanocage" was published by Fudan University's Ma Jinbiao and collaborators [18].

随着大型模型需要处理的序列长度不断增加，注意力运算（Attention）的时间开销逐渐成为主要开销。此前，清华大学陈键飞团队提出的即插即用的 SageAttention 和 SageAttention2 已经被业界及社区广泛的使用于各种开源及商业的大模型中，比如 Vidu，CogvideoX，Mochi，Wan，HunyuanVideo，Flux，Llama3，Qwen 等。 近日，清华大学陈键飞团队进一步提出了针对 BlackWell 架构的首个全 FP4 量化的即插即用注意力算子（ SageAttention3 ）。实现了 5 倍相比于 FlashAttention 的 即插即用的推理加速 （此前的 SageAttention V1/V2/V2++ 分别达到了 2.1，3，3.9 倍的加速效果），比如在 RTX 5090 上，SageAttention3 达到了 1040 TOPS 的速 度，甚至是比 RTX 5090 昂贵十几倍的 H100 上使用 Hopper 独有的 FlashAttention3 还要快 1.65 倍！SageAttention3 在多种视频和图像生成等大模型上（包括 Hunyua ...

Core Insights - Huawei has launched the "Genius Challenge" to attract global talent in five key areas: intelligent connectivity & computing, fundamental research and innovation, intelligent terminals, cloud computing, and intelligent vehicles [3][4][5][6] Group 1: Intelligent Connectivity & Computing - The challenge includes research on autonomous intelligent wireless communication architecture and key technologies to meet future communication demands [3] - It also focuses on the key technologies of the Ascend reinforcement learning system to enhance performance [3] - Research on AI cluster all-optical switching networks aims to improve data transmission speed and efficiency for large-scale AI computing [3] Group 2: Fundamental Research & Innovation - Key technologies for large model security are being explored to address safety risks in current applications [4] - Research on intelligent imaging/editing technology aims to achieve breakthroughs for enhanced user visual experiences [4] - The design and optimization of training cluster architecture will improve the efficiency and quality of model training [4] Group 3: Intelligent Terminals - The challenge includes research on world models to help intelligent terminals better understand and simulate physical laws [5] - It aims to enhance personalization and memory capabilities for intelligent terminals [5] - Research on multimedia algorithms based on computer vision and multimodal understanding is also included [5] Group 4: Cloud Computing - Research on generalizable embodied intelligent operation technology seeks to enable cloud AI to control physical devices [6] - The challenge includes exploring core technologies for the digital-native era [6] - AI-based next-generation cloud network infrastructure research aims to build advanced cloud network systems [6] Group 5: Intelligent Vehicles - The challenge focuses on training and optimizing large models for intelligent vehicles [6] - Research on advanced autonomous driving models is part of the initiative [6] - The development of collaborative control technologies for vehicle chassis aims to enhance safety and comfort [6] Group 6: R&D Investment and Talent Development - Huawei's R&D expenditure for 2024 is projected to reach 179.7 billion yuan, accounting for approximately 20.8% of total revenue [7] - Over the past decade, Huawei has invested more than 1.249 trillion yuan in R&D [7] - The "Genius Challenge" reflects Huawei's commitment to fundamental research and innovation, emphasizing the importance of active participation in basic research [7]