11月20日至21日，2025世界计算大会在长沙举行。20日上午大会开幕式上发布的2025全球计算十大创新成就显示，随着全球算力水平迅速提 高，千亿参数级别的大模型在手机、汽车、智能家居等大众消费品中得到越来越广泛的应用，由此衍生的人工智能、数字孪生等前沿应用，深刻 改变了人们的生产方式和生活方式。 11月20日至21日，2025世界计算大会在长沙举行。新华社记者苏晓洲摄 "依托先进计算建设的卓越型智能工厂，产品研发周期平均缩短28.4%，生产效率提升22.3%，不良品率下降50.2%，碳排放平均减少 20.4%。"工业和信息化部总工程师钟志红在大会致辞中表示，计算是信息技术产业的核心，也是数字经济时代的根基。随着全球计算技术演进与 产业变革，先进基础计算产业的战略地位、基础地位日益凸显。当前，以先进计算为核心驱动力的科技创新，正在加快重塑全球产业格局。 世界计算大会自2019年落户长沙以来，今年已举办至第七届，逐渐发展成为计算领域国际化、专业化的交流平台。（记者苏晓洲、常竣斐） 一支"机器人乐队"在2025世界计算大会现场为与会者表演。新华社记者苏晓洲摄 在大会现场，昇腾384超节点系统、超智融合算力集群、 ...

在过去20多年时间里，金耀初一直在思考从自然演化和生物发育角度来研究、理解人的智能的可能性。 在他看来，人的神经系统和形态是协同发育，神经系统的发育过程跟环境、形态是密不可分的，这与近 期比较热的具身智能在原理上是相似的。 例如，当演化一个有不同模块、不同结构深度的脉冲大模型时，技术人员往往会基于Transformer架构加 入一些类脑的机制，实现不同的通道，有自上而下的通道，也有自下而上的通道，也有注意力机制的调 整，使得结构更灵活。这与人脑内有不同通道，也有很多神经调控机制，有着很大的关联。 大会上，金耀初还发布了西湖大学人工智能领域最近几年的研发成果。据他介绍，在具身智能领域，西 湖大学在大模型的辅助下，目前已经能够做到更加自主地感知、自主决策和自主执行。（文猛） 责任编辑：常福强 炒股就看金麒麟分析师研报，权威，专业，及时，全面，助您挖掘潜力主题机会！ 新浪科技讯 11月9日下午消息，2025年世界互联网大会乌镇峰会期间，欧洲科学院院士、西湖大学人工 智能讲席教授金耀初指出，"传统大模型有很多局限性，包括能耗大，没有自主学习的瓶颈等，基于类 脑神经网络，将类脑计算与具身智能结合，才能让类脑大模型真正具 ...

Core Insights - The development of brain-like computing is driven by the challenges faced by existing computing systems [1][2] - The demand for intelligent computing, particularly from large models in artificial intelligence, is experiencing explosive growth [1][2] Group 1: Event Overview - The 2025 Brain-like Intelligent Computing Forum and Innovation Product Launch was held on October 24, organized by the Guangdong Institute of Intelligent Science and Technology [1] - Experts from various universities and research institutions shared insights on cutting-edge technologies in brain-like computing [1] Group 2: Technological Advancements - Tsinghua University's brain-like computing center has developed a brain-like cloud brain platform with approximately 1.6 billion neurons [2] - The platform supports both brain science and brain-like science, addressing long hardware development cycles with a trillion-level end-to-end brain-like simulation platform [2] Group 3: Product Launch - The world's first ultra-compact, mobile brain-like computing device, "Wise One" BIE-1, was officially launched [2] - The device integrates 1,152 CPU cores, 4.8TB DDR5 memory, 204TB storage, and a 40G high-speed network interface, fitting into a space the size of a mini single-door refrigerator [2] Group 4: Core Technology - The product features a unique brain-like algorithm called "Intuitive Neural Network" (INN), capable of efficiently running various model training and inference, including large models [3] - The INN allows for training of 10 billion tokens in just 30 hours, with training and inference speeds reaching 100,000 tokens/second and 500,000 tokens/second, respectively [3] Group 5: Future Goals - The Guangdong Institute of Intelligent Science and Technology aims to build embodied brain-like intelligence, focusing on visual and location positioning, as well as multi-modal continuous learning [3]

Core Viewpoint - The article discusses the recent advancements in Metal-Organic Frameworks (MOFs) and their application in creating ultra-miniature fluid chips, highlighting their potential to revolutionize computing by mimicking brain-like memory functions [1][20]. Group 1: MOF Technology and Applications - MOFs, once deemed "useless," have gained recognition after winning the Nobel Prize in Chemistry, leading to innovative applications such as fluid chips [1][20]. - The newly developed fluid chips can perform conventional calculations while also retaining previous voltage changes, resembling short-term memory similar to that of brain neurons [2][3]. - The creation of advanced fluid chips using MOF materials addresses the challenges of high-precision nano-channel devices, enabling adjustable non-linear ion transport [4][5]. Group 2: Device Structure and Functionality - Researchers constructed a layered nano-fluid transistor device (h-MOFNT) using Zr-MOF-SO₃H crystals, which features heterogeneous junctions for enhanced performance [7][8]. - The device exhibits non-linear proton transport characteristics, differing from typical diode behavior, indicating a threshold-controlled transport mechanism [12][13]. - The h-MOFNT demonstrated a memory effect, capable of retaining past voltage states, which could lead to applications in liquid-based information storage and brain-like computing [18][19]. Group 3: Historical Context and Future Potential - Historically, MOFs have been viewed as having significant theoretical potential but lacking practical applications, with over 100,000 related papers published but few achieving industrial application [25][26]. - The challenges faced by MOFs include structural stability issues and complex synthesis processes, which have hindered their widespread use [27][28]. - The emergence of MOF-based chips suggests that the material may not be "useless" but rather that suitable applications have yet to be fully explored [29].

Core Insights - The recent Nobel Prize recognition of Metal-Organic Frameworks (MOFs) has sparked new interest in their practical applications, particularly in the development of advanced fluidic chips that mimic brain-like memory functions [1][14]. Group 1: MOF Technology and Applications - Scientists at Monash University have developed ultra-miniature fluidic chips using MOFs, which not only perform conventional calculations but also retain previous voltage changes, resembling short-term memory akin to brain neurons [3][4]. - The h-MOFNT device, constructed from layered Zr-MOF-SO₃H crystals, demonstrates significant potential in simulating biological mechanisms and ion separation due to its precise nanoscale channel structures [5][6]. - The device exhibits unique non-linear proton transport characteristics, which differ from traditional diode behavior, indicating a threshold-controlled transport mechanism [9][10]. Group 2: Performance Characteristics - The h-MOFNT device shows a pronounced hysteresis effect during voltage scanning, suggesting its ability to remember past voltage states and indicating a form of fluid memory [10][11]. - Experiments with multiple h-MOFNT devices in parallel have produced a series of non-linear I-V curves, simulating electronic field-effect transistor (FET) output characteristics [11][13]. - The device's ability to maintain local electric potentials during proton transport highlights its potential for liquid-based information storage and brain-like computing [13][14]. Group 3: Historical Context and Future Potential - Historically, MOFs were considered "useless" due to their theoretical nature and lack of practical applications, despite extensive research and numerous publications [14][20]. - The emergence of MOF-based chips challenges the notion of their ineffectiveness, suggesting that the right applications and contexts for MOFs have yet to be fully realized [20][21]. - The potential for MOFs to serve as customizable materials for new functionalities presents unprecedented opportunities in various fields, including energy storage, sensing, and photonic devices [15][20].

Core Viewpoint - The article highlights the recent breakthrough in using Metal-Organic Frameworks (MOFs) to create ultra-miniature fluid chips, which can perform computations and exhibit short-term memory similar to brain neurons, challenging the previous notion that MOFs were "useless" [1][20]. Group 1: MOF Technology and Applications - MOFs, once considered theoretical with limited practical applications, have now been recognized for their potential in advanced computing technologies following their Nobel Prize acknowledgment [1][21]. - The newly developed fluid chip, made from MOF materials, can overcome limitations of traditional electronic chips by enabling advanced functionalities [3][5]. - The h-MOFNT device constructed from layered Zr-MOF-SO₃H crystals demonstrates unique ion transport properties, allowing for precise control over ionic movement [7][12]. Group 2: Device Characteristics and Performance - The h-MOFNT device exhibits non-linear proton transport characteristics, which differ from typical diode behavior, indicating a threshold-controlled transport mechanism [12][13]. - Experimental results show that the device can remember past voltage states, demonstrating fluid memory and learning capabilities, akin to electronic devices [16][18]. - The ability to create a small fluid circuit using multiple h-MOFNTs showcases the potential for complex computations and memory functions in liquid systems [16][19]. Group 3: Historical Context and Future Prospects - Historically, despite extensive research (over 100,000 related papers), the practical industrial application of MOFs has been limited due to issues like structural stability and production costs [25][27]. - The emergence of MOF-based chips suggests that the material may not be "useless," but rather that suitable applications were not previously identified [29]. - Future developments may lead to the realization of liquid-based information storage and brain-like computing systems through innovative design of heterogeneous constraint systems [19].

Core Insights - Researchers at Monash University have developed a microfluidic chip that mimics the neural pathways of the brain, potentially paving the way for next-generation computing [1] Group 1: Technology Development - The chip, the size of a coin, is made from specially designed metal-organic framework (MOF) materials and transmits ions through tiny channels, simulating the switching of electronic transistors in computers [1] - Unlike traditional computer chips, this new chip can "remember" previous signals, mimicking the plasticity of brain neurons [1] Group 2: Research Findings - The research paper was published in the journal "Science Advances," highlighting the potential of engineered nanoporous materials in the development of next-generation devices [1] - Professor Wang Huanting stated that the team observed saturated nonlinear conduction of protons in nano-fluidic devices for the first time, opening new avenues for designing ionic electronic systems with memory and learning capabilities [1] Group 3: Future Applications - The research team constructed a small fluid circuit with multiple MOF channels to validate the chip's potential, which responds to voltage changes similarly to electronic transistors while also exhibiting memory functions [1] - The chip is expected to have future applications in liquid data storage or "brain-like" computing systems [1]

Core Insights - The world's first multi-center clinical trial for a neural interface targeting hydrocephalus has been launched, marking a significant advancement in brain-machine interface technology beyond traditional applications [3][4][5] - The project involves leading medical institutions in China and aims to establish Chinese standards and solutions for precision treatment in neurological disorders [3][4] - Recent breakthroughs in cognitive science and brain-machine interfaces are opening new avenues for understanding brain functions and enhancing patient care [3][4] Group 1: Neural Interface Technology - The "North Brain No. 1" intelligent brain-machine system allows patients with conditions like ALS and spinal cord injuries to control devices using their thoughts, demonstrating promising clinical outcomes [5][6] - This system is a semi-invasive product that does not penetrate brain tissue, ensuring high safety and simplicity in surgical procedures [5][6] - The initial clinical trials have shown a high success rate, with over 98% of effective channels maintained after 7 months in the first patient [6] Group 2: Neuromorphic Computing - The "Wukong" neuromorphic computing system developed by Zhejiang University aims to simulate brain functions and enhance intelligent computing capabilities [7][8] - This system features advanced hardware with over 20 billion pulse neurons and a power consumption of approximately 2000 watts, making it one of the most efficient neuromorphic computers [7][8] - It has the potential to accelerate brain science research and reduce reliance on animal testing by simulating various animal brains [8][9] Group 3: Brain Imaging Technology - The development of a wearable atomic magnetometer for brain magnetometry allows for more precise and mobile brain activity monitoring, which can aid in diagnosing brain diseases [10][11] - This technology significantly reduces the cost and complexity of brain scans, making it accessible for large-scale screening of conditions like autism and Alzheimer's disease [10][11] - The system's portability and ease of use are expected to enhance early diagnosis and intervention for various neurological disorders [11] Group 4: Olfactory Function Assessment - A localized olfactory function assessment and training system has been developed to detect early signs of neurodegenerative diseases through changes in smell [12][13] - This system can provide early warnings for conditions like Parkinson's and Alzheimer's, potentially allowing for earlier intervention [12][13] - The training program is designed to be user-friendly and can be conducted via a mobile application, creating a feedback loop for monitoring and intervention [12][13]

Core Insights - The rapid development of brain science in China over the past decade has been significantly driven by breakthroughs in interdisciplinary technologies such as imaging, molecular biology, and artificial intelligence [1] Group 1: Innovations in Brain Science - The "North Brain No. 1" intelligent brain-machine system is the first international semi-invasive brain-machine interface product that achieves over 100 channels of high-throughput, wireless implantation, aimed at helping patients with motor and speech disabilities due to spinal cord injuries, strokes, and ALS [1][3] - The "Wukong" ultra-large-scale neuromorphic brain computer, developed by Zhejiang University, features a neuron scale close to that of a monkey brain and can complete simulation tasks in one minute that traditional computers take a day to perform [4] - A wearable atomic magnetometer brain magnetometry system has been developed to overcome the high cost and low flexibility of traditional brain magnetometry, making it applicable in various fields such as brain science research and brain-machine interface [4][6] Group 2: Clinical Applications and Benefits - The "North Brain No. 1" has been implanted in five patients, primarily benefiting those with motor disabilities from spinal cord injuries and strokes, demonstrating not only functional replacement but also rehabilitation effects [3] - The system has shown a signal quality stability of over 98% six months post-implantation, indicating its effectiveness in clinical applications [1] - A localized olfactory function assessment and training system has been developed to provide early warning solutions for neurodegenerative diseases such as Alzheimer's and Parkinson's, achieving international leading levels in reliability and validity [6]

Core Insights - The article discusses the development of SpikingBrain-1.0, a brain-inspired large model that aims to reduce the computational costs associated with long sequence reasoning [1][2]. Group 1: Model Architecture and Performance - SpikingBrain-1.0 leverages a brain-like information processing mechanism, achieving linear/near-linear complexity, which significantly enhances speed for long sequences. For instance, it shows a 26.5x speed improvement on a 1M length sequence compared to mainstream models [2][18]. - The model is designed to be compatible with domestic GPU clusters, indicating the feasibility of creating a new ecosystem for non-Transformer large models in China [2][28]. - The architecture includes SpikingBrain-7B and SpikingBrain-76B, which utilize a linear (mixed) model structure and a hybrid linear MoE model, respectively [10][14]. Group 2: Theoretical Foundations - The research team has established that complex endogenous dynamics in spiking neurons can mathematically equate to combinations of simpler spiking neurons, suggesting the potential for smaller networks to replace larger ones [5][6]. - A new approach based on "endogenous complexity" is proposed, aiming to integrate the rich dynamical characteristics of biological neurons into model development [7][8]. Group 3: Efficiency and Training - SpikingBrain-1.0 demonstrates significant training efficiency for long sequences, achieving comparable performance to many open-source Transformer models with only about 2% of the data [18]. - The model supports multi-card parallel inference and can handle up to 4M length sequences, with substantial acceleration in time-to-first-token (TTFT) compared to standard attention mechanisms [21][22]. Group 4: Future Directions - The team aims to further explore the relationship between endogenous dynamics of neurons and foundational AI operators, seeking to bridge neuroscience and artificial intelligence [28]. - The model is expected to provide significant efficiency advantages in scientific tasks involving long sequences, such as complex multi-agent simulations and molecular dynamics [28].