Core Viewpoint - The article discusses the advancements and innovations in the VLA (Vision Language Architecture) and its comparison with DeepSeek's MoE (Mixture of Experts), highlighting the unique approaches and improvements in model architecture and training processes. Group 1: VLA and MoE Comparison - Both VLA and MoE have been previously proposed concepts but are now being fully realized in new domains with significant innovations and positive outcomes [2] - DeepSeek's MoE has improved upon traditional models by increasing the number of specialized experts and enhancing parameter utilization through Fine-Grained Expert Segmentation and Shared Expert Isolation [2] Group 2: Key Technical Challenges for VLA - The VLA needs to address six critical technical points, including the design and training processes, 3D spatial understanding, and real-time inference capabilities [4] - The design of the VLA base model requires a focus on sparsity to expand parameter capacity without significantly increasing inference load [6] Group 3: Model Training and Efficiency - The training process incorporates a significant amount of 3D data and driving-related information while reducing the proportion of historical data [7] - The model is designed to learn human thought processes, utilizing both fast and slow reasoning methods to balance parameter scale and real-time performance [8] Group 4: Diffusion and Trajectory Generation - Diffusion techniques are employed to decode action tokens into driving trajectories, enhancing the model's ability to predict complex traffic scenarios [9] - The use of an ODE sampler accelerates the diffusion generation process, allowing for stable trajectory generation in just 2-3 steps [11] Group 5: Reinforcement Learning and Model Training - The system aims to surpass human driving capabilities through reinforcement learning, addressing previous limitations related to training environments and information transfer [12] - The model has achieved end-to-end trainability, enhancing its ability to generate realistic 3D environments for training [12] Group 6: Positioning Against Competitors - The company is no longer seen as merely following Tesla in the autonomous driving space, especially since the introduction of V12, which marks a shift in its approach [13] - The VLM (Vision Language Model) consists of fast and slow systems, with the fast system being comparable to Tesla's capabilities, while the slow system represents a unique approach due to resource constraints [14] Group 7: Evolution of VLM to VLA - The development of VLM is viewed as a natural evolution towards VLA, indicating that the company is not just imitating competitors but innovating based on its own insights [15]

Core Insights - The rise of AI-native organizations represents a dual revolution in technology and organizational structure, posing significant challenges to traditional internet giants [1][2] - The competition is not only about technological capabilities but also about organizational forms, cultural genes, and talent strategies [2][3] Group 1: Characteristics of AI-native Organizations - AI-native organizations integrate AI as a core driver of products, services, and business processes, rather than as an added feature [2] - They possess self-developed core technologies, with rapid iteration speeds that outpace traditional companies, exemplified by OpenAI's swift transition from GPT-3 to GPT-4 within two years [2] - Product design inherently relies on AI capabilities, making it impossible for products to exist independently of AI [3] - The focus has shifted from "data and computing power" to "algorithms and community," emphasizing algorithm breakthroughs and scenario innovations as keys to market recognition [4] - Organizational structures are fluid, with flat, self-organizing teams that enable rapid decision-making and resource responsiveness [5] - A geek culture and strong founder cohesion drive these organizations, emphasizing technical idealism and long-term value [6] Group 2: Challenges for Traditional Internet Giants - Traditional tech giants face a core issue: how to evolve their organizations to maintain competitiveness in the AI-native wave [2][9] - Despite having significantly more resources, traditional companies struggle to replicate the technical sharpness of AI-native organizations like DeepSeek [1][9] - The lack of visionary leadership and a clear pursuit of algorithmic efficiency hampers traditional firms' ability to compete effectively [9] - The user engagement battle is intensifying, with AI-native applications rapidly gaining traction and threatening traditional applications' user time [10] Group 3: Strategic Responses from Major Companies - Major companies are attempting to integrate AI-native capabilities into their core businesses, recognizing the potential for scalable applications [11][21] - ByteDance is restructuring its AI organization to enhance agility and innovation, with a focus on AI-native talent [19][20] - Tencent is migrating its AI product lines to a more integrated structure, emphasizing collaboration with AI-native models [21] - Alibaba plans to invest over 380 billion yuan in AI infrastructure and aims for a comprehensive transformation across its core businesses [22] Group 4: Future Directions and Organizational Evolution - The evolution of organizational forms will be crucial as companies transition from traditional data-algorithm-traffic models to a model-data-agent framework [27] - Companies must focus on enhancing their organizational learning speed to convert technological breakthroughs into business cycles effectively [27] - The historical challenges of organizational inertia must be addressed to facilitate meaningful transformation in response to AI-native competition [25][26]

| © 企查查 企业主页 | | --- | | 杭州深度求索人工智能基础技术研 存续 | | 究有限公司 | | 21万+ 91330105MACPN4X08Y ¥ 发票抬头 | | 简介:DeepSeek成立于2023年,是一家通用人工智能模... 展开 | | 法定代表人 注册资本 成立日期 | | 製作 1000万元 2023-07-17 | | 企查查行业 规模 品丁 2023年 | | 信息系统集成服务 微型 XS 4人 | | & 0571-85377238 | | 9 浙江省杭州市拱墅区环城北路169号汇金国际大厦西1幢120 | | 1室 | | 宁波程图个业管理 | | 梁文章 服 咨询合伙 ... 大股东 | | 东 | | 持股比例 99.00% 持股比例 1.00% 2 | | 投资企业2家 关联企业15家 2 | | 裴活 王南军 | | 퀘 + 등 执行董事兼. 监事 | | 2 关联企业3家 关联企业2家 | 文/梧桐晓驴 DeepSeek爆火，晓驴好奇地去查了一下开发、运营DeepSeek的公司情况。 "企查查"显示：杭州深度求索人工智能基础技术研究有限公司，英文名Hangz ...