Core Viewpoint - The article discusses the importance of world models in intelligent driving, emphasizing that true understanding of the environment requires a high-bandwidth cognitive system that goes beyond language models [2][3][5]. Summary by Sections World Model vs. Language Model - The world model focuses on spatiotemporal cognition, while the language model addresses conceptual cognition. Language models have low bandwidth and sparsity, making them ineffective for modeling the real world's four-dimensional space-time [2][3]. - The world model aims to establish capabilities directly at the video level rather than converting information into language first [3][5]. VLA and WA - VLA (Vision-Language Architecture) is essentially an extension of language models, adding new modalities but still rooted in language. In contrast, the world model is not merely an addition of language but a comprehensive cognitive system [3][5]. - The ultimate goal of autonomous driving is to achieve open-set interactions, allowing users to express commands freely without being limited to a fixed set of instructions [3][4]. Importance of Language - Language remains crucial for three main reasons: 1. Incorporation of physical laws such as gravity and inertia into the model [6]. 2. Ability to understand and predict object movements in three-dimensional space over time [6]. 3. The vast amount of data absorbed by language models from the internet aids in training autonomous driving systems [7]. Industry Trends - The autonomous driving industry is experiencing intense competition, with many professionals considering transitioning to other fields. The ongoing debate between VLA and WA represents a larger industry transformation [9]. - The article suggests that those who remain in the industry must be versatile talents with rich technical backgrounds, as the market is expected to undergo significant changes [9]. Community and Learning Resources - A community platform has been established to provide resources for learning and sharing knowledge about autonomous driving, including video tutorials, technical discussions, and job opportunities [11][12][24]. - The community aims to gather individuals from various academic and industrial backgrounds to foster collaboration and knowledge sharing [25].

以下文章来源于晚点LatePost ，作者晚点团队 晚点LatePost . 晚一点，好一点 留在智能驾驶，不是因为容易，而是因为更难。 文 丨 魏冰 宋玮 编辑 丨 宋玮 任少卿的头发很有辨识度，浓密、微卷，刘海盖住额头。走进会议室，第一次见他的人把他当成了实习生，知道身 份后调侃说，只有在 AI 创业公司才能看到这么年轻的技术 leader。 "我们就是 AI 公司"——任少卿一本正经的回答。 但他身处的是蔚来，一家还在血海中搏杀的汽车制造商，而他的战场，是智能驾驶。这个反常回答，和他的人生轨 迹相似：总在别人以为答案已定的时候，他偏要走向另一个方向。 2007 年他考入中科大，2016 年博士毕业。期间他提出了 Faster R-CNN（一种基于深度学习的目标检测框架），又 和当时微软亚研院视觉计算组的孙剑、何恺明，博士生张祥雨一起研究 ResNet（残差网络）。后者解决了神经网络 越深越 "失忆" 的难题，让模型可以无限叠加层数，被视为深度学习史上的里程碑。当时任少卿 27 岁。 2016 年，他与曹旭东共同创立自动驾驶公司 Momenta，亲历了自动驾驶最热的创业年代。4 年后，他离开一手创立 的公 ...

Investment Rating - The report indicates a cautious outlook on the market, with a focus on potential interest rate cuts by the Federal Reserve, which may influence stock trading strategies [1][2]. Core Insights - The Federal Reserve's cautious stance is influenced by high inflation and mixed employment data, leading to uncertainty about future interest rate cuts [1][2]. - There is a notable volatility in artificial intelligence stocks, with concerns about the sustainability of growth in language model adoption [1][4]. - Economic data shows a bifurcated landscape, with a slow recovery in the labor market but optimistic projections for early 2026 due to fiscal expansion and potential interest rate cuts [1][5]. - Retail and consumer stocks have performed better than last year, with positive indicators from companies like Walmart, although negative changes in non-farm employment data could lead to market turbulence [1][6]. - Investors are advised to adjust their portfolios towards early-cycle and more cyclical stocks rather than being overly concentrated in artificial intelligence-related stocks [1][7]. Economic Data and Labor Market - Current economic data is polarized, with some indicators suggesting a potential recession while others indicate a possible acceleration in recovery by early 2026 [5]. - The labor market is recovering slowly, with employment growth not returning to previous levels, and factors like immigration and AI potentially impacting job growth [5]. Retail and Consumer Sector - Retail and consumer stocks have shown significant improvement compared to last summer, with Walmart's strong back-to-school performance signaling a positive holiday sales season [6]. - However, there is a risk of market volatility if non-farm employment data shows negative trends [6]. Investment Strategy Recommendations - Investors should focus on a non-recessionary rate-cutting cycle and the anticipated strong recovery in 2026, adjusting portfolios to favor early-cycle and cyclical stocks [7]. - Maintaining flexibility and closely monitoring economic data changes is crucial for investment strategy formulation [7]. Market Volatility and Opportunities - The report highlights that the current financial environment is very loose, with potential market volatility even from small interest rate cuts [13]. - There are significant investment opportunities in the U.S. front-end supply volatility, with events like Oracle's earnings showing substantial price movements [15]. - Emerging markets, particularly in Asia, are showing improved trading performance, driven by AI themes and favorable conditions for non-dollar denominated assets [16][19].

Core Viewpoint - The article emphasizes the importance of developing new methodologies for large model experiments, focusing on key indicators, identifying true bottlenecks, balancing large and small experiments, and enhancing team collaboration [1]. Group 1: Key Indicators - Identifying key indicators is crucial as they should clearly differentiate between state-of-the-art (SoTA) models and others, guiding the direction of model iterations [4]. - Good indicators must objectively reflect performance levels and accurately indicate the direction for model improvements, avoiding the pitfalls of focusing on misleading metrics [4]. Group 2: Experimentation Methodologies - The cost of experiments has increased significantly, making it essential to conduct meaningful experiments rather than low-value ones [5]. - It is advised to conduct large experiments to identify significant issues while using small experiments to filter out incorrect ideas [6]. Group 3: Team Collaboration - Given the complexity of large model experiments, it is important for team members to understand their comparative advantages and roles within the team [8]. - Effective collaboration can be enhanced by finding ways to observe and document experiments together, increasing communication frequency [8].

Core Viewpoint - The article discusses the rapid evolution of AI technology and its applications, highlighting the challenges of energy consumption, model size, and learning mechanisms. Aizip, a company focused on on-device AI models, aims to overcome these efficiency bottlenecks and drive the integration of AI into everyday life [1]. Group 1: AI Efficiency and Innovation - Aizip's mission is to enhance energy efficiency, model efficiency, and learning efficiency in AI systems, moving from "usable" to "efficiently usable" AI [3][10]. - The company emphasizes creating the "smallest and most efficient" AI systems, contrasting with the mainstream focus on general artificial intelligence (AGI) [3][14]. - Aizip's approach is to support businesses that require AI capabilities but lack full-stack AI expertise, allowing them to focus on application development [3][32]. Group 2: Founder's Background and Vision - The founder, Chen Yubei, has a strong academic background in AI and has shifted from theoretical research to practical applications, driven by a desire to see AI implemented in real-world products [4][16]. - The founding of Aizip was catalyzed by the COVID-19 pandemic, which disrupted initial plans for postdoctoral research and prompted discussions about entrepreneurship [6][16]. - Aizip's team comprises experienced individuals with diverse backgrounds, emphasizing a culture of collaboration and long-term value over short-term gains [17][18]. Group 3: On-Device AI Revolution - The article predicts that over 50% of AI reasoning will occur on-device in the near future, driven by advancements in hardware and user demand for low-latency, privacy-focused AI products [30][31]. - Aizip's product line includes multi-modal perception models and language models, focusing on seamless integration into various devices to enhance user experience without overtly displaying AI functionality [22][23]. - The company aims to create a comprehensive AI model ecosystem compatible with mainstream hardware, facilitating easier integration for clients [34][36]. Group 4: Market Position and Future Outlook - Aizip positions itself as a foundational support for companies lacking the resources to build their own on-device AI teams, anticipating a growing market for such capabilities [32][34]. - The company has established partnerships with leading hardware manufacturers and has achieved recognition for its innovative AI products [38]. - Aizip's strategy focuses on gradual commercialization, prioritizing technology validation and model stability before scaling operations [35][36].

本文主要作者是吕昂和谢若冰。吕昂，中国人民大学博士生，研究方向为语言模型结构优化，导师为严睿教授；谢若冰，腾讯高级研究员，研究方向为大语言模 型、推荐系统。 最近的一篇论文中，来自人大和腾讯的研究者们的研究表明，语言模型对强化学习中的奖励噪音具有鲁棒性，即使翻转相当一部分的奖励（例如，正确答案得 0 分，错误答案得 1 分），也不会显著影响下游任务的表现。 研究者解释道，强化学习对下游任务的提升，关键不仅在于奖励的准确性，而更在于模型是否能够产生高质量的思考过程。仅通过奖励模型输出中关键思考词的 出现频率，而非基于答案正确性的奖励，语言模型依然能够在下游任务中取得非常高的峰值表现。这表明，强化学习对下游任务的提升，更多来源于让模型学会 采用恰当的思考路径接近正确答案。而相关的解题基础能力，模型已在预训练阶段获得。因此，预训练阶段的能力提升依然至关重要。 研究者还展示了基于思考模式的极简奖励如何有效校准奖励模型，从而在开放性 NLP 任务中增强语言模型的表现，并使较小的模型也能通过强化学习成功获得思 考能力。 论文地址：https://huggingface.co/papers/2505.22653 代码链接： ...

Core Insights - The article emphasizes a shift from traditional demand research to a dynamic sequence thinking approach, suggesting that true needs are triggered by context rather than predefined lists [1][5] - It critiques the conventional methodology of focusing solely on customer needs, arguing that this perspective assumes the existence of objectively measurable problems [2][5] Group 1: Demand Research Methodology - Traditional demand research operates under the assumption that there are objective hidden desires that can be discovered and enumerated [5] - The article suggests that effective demand arises from specific behavioral sequences, and appropriate prompts can stimulate these needs [5][6] - Observational techniques are highlighted as essential, where experienced founders act like ethnographers to create a comprehensive profile of their target customers [2][3] Group 2: Dynamic Sequence Thinking - The article advocates for abandoning objectivity in demand research, proposing that presenting one's worldview can crystallize customers' chaotic thoughts into concrete needs [3][4] - It illustrates that demand is context-dependent and not an isolated entity, emphasizing the importance of situational triggers in generating needs [4][5] - The process of creating demand is likened to a learning experience, where trial and error lead to the ability to generate responses to various situations [8][9] Group 3: Practical Application - The article encourages individuals to observe and respond to their environment actively, akin to how a language model learns to generate appropriate responses based on context [7][8] - It describes the journey of mastering the ability to create demand as a developmental process, similar to learning to walk, where repeated attempts lead to proficiency [8][9] - Ultimately, the goal is to become adept at generating the right sequences that evoke desired responses from others [6][8]