Core Insights - The article discusses the significance of integrating Vision-Language-Action (VLA) models with Reinforcement Learning (RL) in the field of Embodied AI, emphasizing the limitations of imitation learning and the necessity for robust learning methods [1][2][4]. Group 1: Importance of VLA+RL - VLA models are being developed to apply powerful Vision-Language Models (VLM) in the control of robots, primarily through supervised fine-tuning (SFT) [2]. - Imitation learning alone is insufficient for robots to handle novel situations, necessitating the use of RL to enhance robustness and persistence in task execution [4]. Group 2: Challenges in Applying RL to VLA - The integration of RL with VLA faces three main challenges: environmental differences, model instability, and computational demands [6]. - Direct application of RL algorithms to large VLA models can lead to catastrophic forgetting and training collapse, making it difficult to maintain performance [6]. Group 3: Solutions to VLA's RL Challenges - The industry has proposed three types of solutions to address the challenges faced by VLA in RL applications, with a focus on internalizing high-value behaviors through SFT [7][13]. - The iRe-VLA model introduces a two-phase iterative learning process that alternates between online RL for exploration and supervised learning for consolidation [10][15]. Group 4: iRe-VLA Model Architecture - The iRe-VLA model consists of a VLM backbone for understanding images and instructions, and an Action Head for translating features into control signals [11]. - The use of Low-Rank Adaptation (LoRA) technology allows for efficient training without the need for full model fine-tuning [12]. Group 5: Experimental Results and Analysis - Extensive experiments in both simulated environments and real-world scenarios demonstrate the effectiveness of the iRe-VLA method, showing significant improvements in task success rates [26][30]. - The iRe-VLA model outperformed traditional methods, achieving a success rate increase from 43% to 83% in benchmark tasks [30]. Group 6: Conclusion and Future Implications - The article concludes that the iRe-VLA approach provides a viable solution to the challenges of deploying large models in robotic control, ensuring stability and continuous learning [37][42]. - Future research directions include efficient exploration and learning of new skills under sparse rewards, as well as developing scalable RL algorithms for large VLA models [40].

Core Viewpoint - Meta is shifting its strategic focus from the "metaverse" to artificial intelligence (AI), facing multiple internal challenges as a result [1]. Group 1: Internal Conflicts - A newly formed top AI team at Meta is experiencing friction with existing core business departments over resource allocation, development goals, and cultural integration [2]. - Internal conflicts have escalated due to differences in priorities regarding AI development, with long-term executives advocating for using data from Instagram and Facebook to enhance social media and advertising, while the new AI team led by Alexandr Wang aims to develop advanced AI models without immediate product application focus [5][12]. - The AI team believes that the existing executive focus on social media improvements is hindering the development of cutting-edge AI models [5]. Group 2: Resource Allocation and Financials - To support its ambitious AI goals, Meta is reallocating resources, significantly cutting the budget for Reality Labs, which oversees VR, AR, and metaverse initiatives [8]. - Reality Labs has incurred losses exceeding $70 billion since the end of 2020, and Meta plans to reduce its budget by up to 30% (approximately $4 billion to $6 billion) next year, with funds redirected to the AI team [11]. - Meta's projected spending in AI for this year is estimated to be between $66 billion and $72 billion, nearly equivalent to the total losses from its metaverse business in recent years [11]. Group 3: Strategic Challenges - Meta's current situation mirrors historical challenges faced by tech giants, such as Microsoft's failure to adapt to mobile operating systems, which resulted in a loss of market dominance [17]. - The company is simultaneously engaged in multiple costly battles across the metaverse, short video markets, and AI, leading to a dilution of strategic focus [19]. - The failure of Llama 4 has raised concerns about whether resource allocation towards the metaverse has impeded the AI team's progress at a critical time [19]. Group 4: Cultural and Organizational Dynamics - Tensions within Meta are ongoing due to differing philosophies between established executives and the new AI elite, with some employees prioritizing resources for the profitable social media business [12]. - The departure of Yann LeCun over ideological differences highlights the intense cultural shifts within the organization as it navigates its new direction [21]. - The outcome of Meta's internal struggles will determine whether it faces a collapse similar to Google+ or can reorganize effectively to achieve success akin to Google's Gemini project [22].

Core Viewpoint - The article discusses the introduction of the FeRA (Frequency-Energy Constrained Routing) framework, which addresses the limitations of existing static parameter-efficient fine-tuning (PEFT) methods in diffusion models by implementing a dynamic routing mechanism based on frequency-energy principles [3][23]. Group 1: Research Background and Limitations - The current PEFT methods, such as LoRA and AdaLoRA, utilize a static strategy that applies the same low-rank matrix across all time steps, leading to a misalignment between parameters responsible for structure and detail, resulting in wasted computational resources [8][9]. - The research team identifies a significant "low-frequency to high-frequency" evolution pattern in the denoising process of diffusion models, which is not isotropic and has distinct phase characteristics [7][23]. Group 2: FeRA Framework Components - FeRA consists of three core components: - Frequency-Energy Indicator (FEI), which extracts frequency-energy distribution features in latent space using Gaussian difference operators [11]. - Soft Frequency Router, which dynamically calculates the weights of different LoRA experts based on the energy signals provided by FEI [12]. - Frequency-Energy Consistency Loss (FECL), which ensures that the parameter updates in the frequency domain align with the model's original residual error, enhancing training stability [13]. Group 3: Experimental Validation - The research team conducted extensive testing on multiple mainstream bases, including Stable Diffusion 1.5, 2.0, 3.0, SDXL, and FLUX.1, focusing on style adaptation and subject customization tasks [19]. - In style adaptation tasks, FeRA achieved optimal or near-optimal results in FID (image quality), CLIP Score (semantic alignment), and Style (MLLM scoring) across various style datasets [20]. - In the DreamBooth task, FeRA demonstrated remarkable text controllability, allowing for specific prompts to be effectively executed [21][26]. Group 4: Conclusion and Future Implications - The FeRA framework represents a significant advancement in fine-tuning diffusion models by aligning the tuning mechanism with the physical laws of the generation process, thus providing a pathway for efficient and high-quality fine-tuning [23][27]. - This work not only sets new state-of-the-art (SOTA) benchmarks but also offers valuable insights for future fine-tuning in more complex tasks such as video and 3D generation [27].

机器之心报道 机器之心编辑部 谷歌的领先优势，只保持了不到一个月。 今天是 OpenAI 的十周年纪念日，十周年之际，来点大的。 在「红色警报」后，OpenAI 在北京时间本周五拿出了最新的顶级模型 GPT-5.2 系列 —— 迄今为止在专业知识工作上最强大的模型系列。 GPT-5.2 Thinking ，为专业级工作全面提升标准： 业界最先进的长上下文推理能力 与 GPT-5.1 一样温暖、对话自然 更清晰的讲解，把关键信息提前呈现 改进的操作指南与分步骤讲解 更强的技术写作与翻译能力 更好地支持学习与职业规划 GPT-5.2 Pro ，在面对困难问题时最聪明、最值得信赖的模型： GPT-5.2 的设计目标，就是为人们创造更多经济价值：它在制作电子表格、构建演示文稿、编写代码、理解图像、处理超长上下文、使用工具，以及执行 复杂的多步骤项目方面都有显著提升。 真正的生产力不是空口无凭，让我们来看看数据，GPT-5.2 到底有多强。 在如图所示的众多基准测试中，GPT-5.2 均刷新了最新的 SOTA 水平。 简而言之，OpenAI 本次推出： GPT-5.2 Instant ，为日常工作与学习而打造： | | ...

Core Insights - The article discusses the emergence of intelligent agents based on language models that possess reasoning, planning, and action capabilities, highlighting a new paper from Google that establishes quantitative scaling principles for these agents [1][7]. Group 1: Scaling Principles - Google defines scaling in terms of the interaction between the number of agents, collaboration structure, model capabilities, and task attributes [3]. - The research evaluated four benchmark tests: Finance-Agent, BrowseComp-Plus, PlanCraft, and Workbench, using five typical agent architectures and three LLM families [4][5]. Group 2: Experimental Findings - The study involved 180 controlled experiments across various scenarios, demonstrating that the effectiveness of multi-agent collaboration varies significantly depending on the task [10][11]. - In finance tasks, centralized architectures can enhance performance by 80.9%, while in game planning tasks, multi-agent systems can lead to performance drops of 39% to 70% due to high communication costs [14]. Group 3: Factors Affecting Agent Performance - Three core factors hindering agent scalability were identified: 1. The more tools required, the harder collaboration becomes, leading to inefficiencies [15]. 2. If a single agent is already sufficiently capable, adding more agents can yield negative returns [16]. 3. Without a centralized commander, errors can amplify significantly, highlighting the importance of architectural design [18]. Group 4: Model Characteristics - Different models exhibit distinct collaborative characteristics: - Google Gemini excels in hierarchical management, showing a 164.3% performance increase in centralized structures [19]. - OpenAI GPT performs best in hybrid architectures, leveraging complex communication effectively [20]. - Anthropic Claude is sensitive to communication complexity and performs best in simple centralized structures [20]. Group 5: Predictive Model Development - Google derived a predictive model based on efficiency, overhead, and error amplification, achieving an 87% accuracy rate in predicting the best architecture for unseen tasks [22][25]. - This marks a transition from an era of "alchemy" in agent system design to a more calculable and predictable "chemistry" era [26].

Core Insights - The article highlights the significance of the "Test of Time Award" received by the paper "Faster R-CNN," co-authored by renowned researchers, marking its impact on the field of computer vision since its publication in 2015 [1][5][25] - The presentation by He Kaiming at NeurIPS 2025 summarizes the evolution of visual object detection over the past 30 years, showcasing key milestones and influential works that have shaped the field [6][31] Historical Development - The early attempts at face detection in the 1990s relied on handcrafted features and statistical methods, which were limited in adaptability and speed [12] - The introduction of AlexNet in 2012 demonstrated the superior feature extraction capabilities of deep learning, paving the way for its application in object detection [15] - The R-CNN model, proposed in 2014, revolutionized object detection by integrating CNNs for feature extraction and classification, although it initially faced computational challenges [17][18] Technological Advancements - The development of Faster R-CNN in 2015 addressed the speed bottleneck by introducing the Region Proposal Network (RPN), allowing for end-to-end real-time detection [25] - Subsequent innovations, such as YOLO and SSD in 2016, further enhanced detection speed by enabling direct output of object locations and categories [32] - The introduction of Mask R-CNN in 2017 added instance segmentation capabilities, while DETR in 2020 redefined detection using Transformer architecture [32][34] Future Directions - The article concludes with reflections on the ongoing exploration in computer vision, emphasizing the need for innovative models to replace outdated components as bottlenecks arise [35][36]

Core Insights - The article discusses the significant advancements in achieving dexterous manipulation capabilities in robotics through the Vision-Language-Action (VLA) model, addressing the critical challenge of high-quality data acquisition for training these models [2][6]. Group 1: Key Contributions - The research introduces a Shared Autonomy framework that effectively divides control responsibilities between human operators and autonomous AI systems, significantly reducing cognitive load and data collection costs [2][12][15]. - The DexGrasp-VLA strategy is highlighted as a foundational element of the Shared Autonomy framework, integrating multimodal inputs including tactile feedback, which enhances the robot's ability to adaptively grasp objects [9][20]. - The study establishes a complete technical system composed of four core modules, achieving a closed-loop from data collection to policy optimization [5][8]. Group 2: Data Collection and Efficiency - The Shared Autonomy framework has improved the efficiency of high-quality data collection by 25%, allowing for more data to be collected per hour and compressing the development-deployment cycle to under one day [33]. - The framework has demonstrated a near-industrial standard performance with approximately 90% success rate in grasping over 50 different objects, facilitating the transition of dexterous manipulation technology from concept validation to practical deployment [33]. Group 3: Mechanisms and Enhancements - The Arm-Hand Feature Enhancement module is designed to model and integrate the kinematic differences between the arm and hand, resulting in more natural and robust coordination of macro and micro actions [16][19]. - The Corrective Human-in-the-Loop mechanism allows the robot to learn from failures by incorporating human demonstrations of correct actions, continuously improving the strategy and generalizing to edge cases [20][34]. Group 4: Future Directions - Future research directions include expanding the framework to more complex tasks such as object reorientation and precise placement, as well as exploring intelligent fusion mechanisms to address challenges in tactile feedback [36]. - The potential for autonomous error recognition and recovery through reinforcement learning is also discussed, aiming for a smooth transition from human-robot collaboration to full autonomy [36].

Core Viewpoint - The article discusses the rapid advancements in large models, particularly in the AI field, highlighting the emergence of models like ChatGPT and DeepSeek, and the anticipated release of Google's Gemini 3, which is seen as a significant step towards Artificial General Intelligence (AGI) and Artificial Super Intelligence (ASI) [2][3]. Group 1: Large Model Developments - The investment in AI in the U.S. has surpassed the GDP of many countries, indicating a booming industry [2]. - DeepSeek has achieved remarkable performance with low training costs, further pushing the boundaries of AI capabilities [2]. - Gemini 3 is expected to challenge NVIDIA's ecosystem with its TPU training paradigm [2]. Group 2: Theoretical Foundations - The research paper "Forget BIT, It is All about TOKEN" aims to combine statistical physics, signal processing, and information theory to better understand the mathematical principles behind large models [4]. - The article emphasizes the need for a comprehensive understanding of large models beyond single-dimensional theories, which have limited insights into their underlying principles [3][4]. Group 3: Memory Capacity and Generalization - The memory capacity of large models increases exponentially with the linear growth of model parameters, suggesting that smaller models can still perform effectively but are prone to collapse if over-trained [8]. - The upper bound of generalization error in large models is linked to the absolute sum of logits, necessitating careful management during model reduction techniques like pruning and distillation [8][34]. Group 4: Causality and Prediction - The article posits that the ultimate goal of large models is to predict the next token, with the Transformer architecture being effective in achieving this [14][36]. - The reasoning behind large model capabilities is tied to Granger causality, indicating that while scaling laws will continue, true logical reasoning and concept abstraction may remain out of reach for these models [36][38]. Group 5: Future Directions - The article outlines plans for a series of articles that will delve deeper into the first principles of large models, focusing on statistical physics, signal processing, and information theory [4][39].

Core Insights - The article discusses the significant advancements in AI driven by increased computational budgets and algorithmic innovations over the past decade [2][6] - It highlights that while computational growth is measurable, the quantification of algorithmic progress remains unclear, particularly regarding the efficiency improvements and their scalability [2][3] Group 1: Algorithmic Progress - Research estimates that algorithmic advancements have contributed over 4 orders of magnitude in effective compute over the past decade, while computational scale itself has increased by 7 orders of magnitude [2] - The overall efficiency of models has improved by approximately 22,000 times due to algorithmic innovations, allowing for similar performance with significantly fewer floating-point operations (FLOPs) [3][4] - Most algorithmic innovations yield only minor efficiency improvements, with less than 10 times overall efficiency gain when extrapolated to 2025's computational limits [4][11] Group 2: Scale-Dependent Innovations - Two major scale-dependent algorithmic innovations, from LSTM to Transformer and from Kaplan to Chinchilla, account for 91% of the total efficiency improvements [4][22] - The efficiency gains from algorithmic improvements are significantly larger in large-scale models compared to small-scale models, indicating that algorithmic progress is heavily reliant on computational scale [6][25] - The article suggests that the perceived rapid progress in algorithms may be more a reflection of increasing computational budgets rather than continuous algorithmic breakthroughs [22][24] Group 3: Experimental Findings - The study employed various methods, including ablation studies and scaling experiments, to analyze the impact of individual algorithms and their combinations [5][8] - The findings reveal a highly skewed distribution of efficiency improvements, with a few key innovations contributing disproportionately to overall gains [11][12] - The scaling experiments demonstrate that improvements in neural network architectures are not scale-invariant but exhibit increasing returns to scale [20][21]

Core Insights - Waymo has released a deep blog detailing its AI strategy centered around its foundational model, emphasizing the use of distillation methods to create efficient models for onboard operations [1] - Jeff Dean highlighted the significance of knowledge distillation in AI, reflecting on its initial rejection by NeurIPS 2014, which underestimated its potential impact [3][4] Group 1: Historical Context of Rejected Papers - Many foundational technologies in AI, such as optimizers for large models and computer vision techniques, were initially rejected by top conferences, showcasing a systemic lag in recognizing groundbreaking innovations [6] - Notable figures in AI, including Geoffrey Hinton and Yann LeCun, faced rejection for their pioneering work, often due to reasons that seem absurd in hindsight, such as claims of lacking theoretical basis or being overly simplistic [6] Group 2: Specific Case Studies of Rejected Innovations - LSTM, a milestone in handling sequential data, was rejected by NIPS in 1996 during a period when statistical methods were favored, only to later dominate fields like speech recognition [8] - The SIFT algorithm, which ruled the computer vision domain for 15 years, faced rejection from ICCV and CVPR due to its perceived complexity and lack of elegance, ultimately proving the value of robust engineering design [11] - Dropout, a key regularization method for deep neural networks, was rejected by NIPS in 2012 for being too radical, yet it became crucial for the success of models like AlexNet [17] - Word2Vec, despite its revolutionary impact on NLP, received a strong rejection at ICLR 2013 due to perceived lack of scientific rigor, but it quickly became a cornerstone of text representation [19][20] Group 3: Reflection on Peer Review Limitations - The peer review system often struggles to recognize disruptive innovations, leading to a "simplicity trap" where reviewers equate mathematical complexity with research contribution [40] - Reviewers tend to maintain existing paradigms, which can hinder the acceptance of novel ideas that challenge traditional metrics of success [40] - The demand for rigorous theoretical proof in an experimental field like deep learning can stifle practical breakthroughs, as seen with the initial skepticism towards methods like Adam optimizer [40] Group 4: Broader Implications - The experiences of rejected papers illustrate the nonlinear nature of scientific progress, highlighting that peer review, while essential, is limited by human cognitive biases [41] - Historical anecdotes, such as Einstein's rejection of a paper on gravitational waves, emphasize that the true measure of a research's impact is its long-term relevance rather than immediate acceptance [42][44]