以下文章来源于理想TOP2 ，作者理想TOP2 理想TOP2 . 找对社群，深度交流理想长期基本面 作者 | 理想TOP2 来源 | 理想TOP2 点击下方 卡片 ，关注" 自动驾驶之心 "公众号 戳我-> 领取 自动驾驶近30个 方向 学习 路线 以下为正文： Ilya是前OpenAI首席科学家，目前在做超级对齐的工作（如果不认为超级对齐非常重要，本质是不信AGI。） 最近十余年AI界多项最重要的变化由其推动。包括但不限于2012年和Hinton/Alex >>自动驾驶前沿信息获取 → 自动驾驶之心知识星球 本文核心分享四条逻辑链： 本文只做学术分享，如有侵权，联系删文 1. 对predict the next token不同的理解本质是对LLM或AI的潜力与实质有不同的理解。 本文架构： 2. 越认为predict the next token不只是概率分布/统计学的人，越容易认可LLM潜力很大/AI潜力很大/推理过程就是意识雏形甚至就是意识/超级对齐非常重要。 3. 不同时真正的深入思考AI与理想，很容易对理想所做之事含金量低估。 4. 理想的VLA实质是在强化学习占主导的连续predict the n ...

Core Viewpoints - The article presents four logical chains regarding the understanding of "predict the next token," which reflects different perceptions of the potential and essence of LLMs or AI [1] - Those who believe that predicting the next token is more than just probability distributions are more likely to recognize the significant potential of LLMs and AI [1] - A deeper consideration of AI and ideals can lead to an underestimation of the value of what ideals accomplish [1] - The ideal VLA essentially focuses on reinforcement learning dominating the continuous prediction of the next action token, similar to OpenAI's O1O3, with auxiliary driving being more suitable for reinforcement learning than chatbots [1] Summary by Sections Introduction - The article emphasizes the importance of Ilya's viewpoints, highlighting his significant contributions to the AI field over the past decade [2][3] - Ilya's background includes pivotal roles in major AI advancements, such as the development of AlexNet, AlphaGo, and TensorFlow [3] Q&A Insights - Ilya challenges the notion that next token prediction cannot surpass human performance, suggesting that a sufficiently advanced neural network could extrapolate behaviors of an idealized person [4][5] - He argues that predicting the next token well involves understanding the underlying reality that leads to the creation of that token, which goes beyond mere statistics [6][7] Ideal VLA and Reinforcement Learning - The ideal VLA operates by continuously predicting the next action token based on sensor information, indicating a real understanding of the physical world rather than just statistical probabilities [10] - Ilya posits that the reasoning process in the ideal VLA can be seen as a form of consciousness, differing from human consciousness in significant ways [11] Comparisons and Controversial Points - The article asserts that auxiliary driving is more suited for reinforcement learning compared to chatbots due to clearer reward functions [12][13] - It highlights the fundamental differences in the skills required for developing AI software versus hardware, emphasizing the unique challenges and innovations in AI software development [13]

Core Viewpoint - The VLA architecture has significant technical potential and is seen as a long-term framework for autonomous driving, evolving from end-to-end systems to a more robust model that can support urban driving scenarios [1][4]. Group 1: VLA Architecture and Technical Potential - The VLA architecture is derived from robotics and embodied intelligence, emphasizing the need for visual and action capabilities, and is expected to evolve alongside advancements in robotics [1]. - VLA's ability to generalize is not solely dependent on data input but is enhanced through reinforcement learning, allowing it to autonomously address new challenges [5]. - The VLA model is designed to support various platforms without differentiation, ensuring consistent performance across different hardware [2][3]. Group 2: Performance Metrics and Future Enhancements - The current operational speed of the Thor-U chip is 10Hz, with potential upgrades to 20Hz and 30Hz through optimizations in data and algorithm architecture [2]. - The VLA model's upgrade cycle includes both pre-training and post-training updates, allowing for continuous improvement in capabilities such as spatial understanding and language processing [6]. - The VLA architecture aims to achieve L4 autonomous driving capabilities within a year, with a focus on rapid iteration and simulation-based testing [12]. Group 3: User Experience and Interaction - Language understanding is deemed essential for future autonomous driving, enhancing the model's ability to handle complex scenarios and improving overall driving experience [4]. - The VLA system is designed to adapt to user preferences, allowing for different driving styles based on individual needs and enhancing user trust in the technology [19]. - Features such as remote vehicle summoning and real-time monitoring of the vehicle's surroundings are being developed to improve user interaction and experience [13]. Group 4: Competitive Landscape and Strategic Decisions - The company is currently utilizing NVIDIA chips for model deployment, focusing on maintaining versatility and avoiding being locked into specific architectures [3]. - The company is closely monitoring competitors like Tesla, aiming to learn from their advancements while prioritizing a gradual and comprehensive approach to achieving full autonomous driving capabilities [12]. - The VLA architecture is positioned as a differentiating factor in the market, leveraging reinforcement learning to enhance driving logic and user experience [20].

Core Viewpoint - The discussion revolves around the implications of Artificial General Intelligence (AGI) and its potential to reflect human limitations and desires, urging a reevaluation of human identity in the face of advanced AI technologies [5][7][24]. Group 1: AGI and Human Identity - AGI is described as a mirror that reveals human limitations and desires, prompting a need for self-reflection as humans create entities capable of understanding complex emotions like "regret" [5][7]. - The evolution of AI from traditional tools to a new species capable of self-evolution raises questions about the future of human-AI relationships and the ethical implications of such advancements [11][21]. - The potential for AGI to amplify human intelligence while also posing risks to cognitive freedom is highlighted, suggesting a duality in its impact on society [5][7]. Group 2: Educational Implications - The emergence of AGI presents an opportunity to reshape educational paradigms, emphasizing the need for individuals to learn how to learn rather than merely accumulating knowledge [24][30]. - AI can enhance educational equity by providing access to knowledge and resources that were previously unavailable to underprivileged students, thus transforming traditional learning environments [28][30]. - The focus shifts from rote learning to developing critical thinking and creativity, as AI can handle knowledge-based tasks, allowing humans to engage in more innovative pursuits [26][30]. Group 3: Industry and Innovation - The current landscape of AI development in China is characterized by "follow-up innovation," which may hinder the emergence of groundbreaking original ideas [35][36]. - Strategic support from national resources and a shift in investment culture are necessary to foster an environment conducive to original innovation in AI [36][37]. - The integration of brain science and cognitive science into AI development is proposed as a pathway to break free from existing paradigms and create more advanced AI systems [34][38].

Core Insights - The article discusses the transformative impact of artificial intelligence (AI) on the automotive industry, particularly through the lens of Li Auto's VLA driver model, which represents a significant evolution in AI applications within the sector [1][3][10] Group 1: AI Development and Evolution - Li Auto categorizes AI tools into three levels: information tools, auxiliary tools, and production tools, emphasizing that the true explosion of AI will occur when it becomes a production tool [3][5] - The development of the VLA driver model follows a clear evolutionary trajectory, moving from basic rule-based systems to advanced models that can understand and interact with the physical world like humans [5][9] - The company views the VLA model as an evolutionary process rather than a sudden leap, highlighting the importance of foundational algorithms and end-to-end technology in its development [5][10] Group 2: Human-Centric AI and Safety - Li Auto emphasizes the importance of aligning AI behavior with human values through techniques like Reinforcement Learning from Human Feedback (RLHF), ensuring that the AI adheres to traffic rules and societal driving norms [7][9] - The company aims to create an AI that embodies human values, establishing ethical boundaries for its operation, which is crucial for user trust and safety [7][10] - Li Auto's approach to AI development reflects a commitment to enhancing safety standards, addressing the inherent contradictions in automated driving capabilities [7][9] Group 3: Strategic Vision and Market Position - Li Auto positions itself as a pioneer in the AI space, claiming to explore "unmanned areas" in both automotive and AI sectors, which have not been traversed by major competitors like DeepSeek or OpenAI [9][10] - The company is focused on building a robust technological foundation, leveraging its past experiences in the automotive field to drive innovation in AI [9][10] - Li Auto's strategic vision includes redefining the essence of smart vehicles, moving beyond mere parameter accumulation to a deeper understanding of productivity [10]