原文 发布于2017年1月27日，原标题为《新春招聘：先来了解一下车和家 》 以下为原文： 至于时间的进度，我们第一批硬模组装的 EP1 测试车已经在春节前完成了。 【关于质量】 我们尊重汽车工业所有的质量和生产准则。 作为一个初创企业，质量是我们的命根子，我们为产品制定了两倍于国内普遍标准里程的实验和路 试，以及严格的供应链质量控制和生产质量控制体系。 我们为消费者打造了两款产品，一个全能型的 SUV，一个专注于短途出行的 SEV。 SUV 是一款 7 座/6 座版的中大型豪华 SUV，同级别的产品有：沃尔沃的 XC90、奥迪的Q7、宝马的 X5 等。我们将陆续提供三种不同的动力，分别为：纯电动四驱、增程电动四驱、串联式混合动力。 满足不同消费者实际能源环境的需求，让任何人都可以享受到电驱动的平顺性和低能耗。 SEV 是按照欧洲 2016 年最新 L7e 规范（同时符合美国的 NEV 和日本相应的规范）打造的四轮电动 摩托车，专注于短途出行。类似的产品以概念车为主，如丰田的 i-Road、本田的MC-Beta、雷诺日产 的 Twizy 等。用户可以用任何 220V/10A 的标准插头直接充电，充电功率低于电 ...

原文 发布于2016年12月11日，原标题为《 安静、高能、易用，airx A7使用报告 》 以下为原文： 我是从 2010 年入住新房开始真正明白空气净化器的作用，之前也买过净化器，但基本上都在乱用， 比如打开窗户开净化器这种脑残的行为都做的出来。前前后后买了 20 多个净化器，主要是因为我们 家的新风管道因为劣质的开地产发商施工（风管埋在水泥地面下面，基本没法改造）问题，让新风系 统基本上荒废掉了，所以对净化器的依赖度很高。 怎么衡量净化器是不是真的有用？2010 年以前，每到春天、秋天、冬天，生活在北京的我一觉醒来 都会嗓子疼（还有大量的鼻屎，好恶心，弹出去），还有永远不好的慢性鼻炎，过去我总以为是干燥 的问题，可是你加装了加湿器以后，发现并没有什么改善。自从用了净化器以后，除了真正感冒发 烧，嗓子疼的问题再也没有发生过，我的慢性鼻炎也基本上好了，而湿度还是那个湿度，没有改变。 我也真正明白了，过去那么多年的困扰，核心都是雾霾造成的。这也是我 2010 年后，坚定不移的使 用净化器的原因（家里、车里、办公室里都在用，这是我们日常超过 80%的生活空间），虽然很多 人今天还认为净化器没有用。总之，谁用谁知 ...

Core Viewpoint - The article emphasizes the user-centric approach of the company, highlighting how it integrates user feedback into its product features and services, which enhances its competitive edge in the market. Group 1: User-Centric Features - The company offers free service fees for its self-operated charging stations during holidays and special routes, along with discounts on charging cards for users [1] - The pricing strategy for charging shows a significant discount for company car owners compared to non-owners, indicating a thoughtful approach to user benefits [1] - The company has implemented numerous small details that reflect a strong user understanding, such as adjusting features based on user feedback, which contributes to long-term competitiveness [2] Group 2: Functionality and Usability - Common functionalities are designed to be highly user-friendly, allowing for seamless interactions such as voice commands for music and navigation [3] - The company has optimized various features to ensure ease of use, such as reminders and automated settings for climate control, which are not commonly found in other vehicles [4] - The overall satisfaction with less common scenarios indicates that the company has successfully addressed user needs, outperforming competitors in user experience [5] Group 3: Company Culture and Future Outlook - The emphasis on continuous improvement and attention to detail reflects a strong corporate culture that prioritizes user experience and innovation [5] - The company is expected to gradually build a positive reputation over time, even in the face of initial skepticism, as evidenced by past product performance trends [5] - Upcoming events, such as the AI Talk series, are positioned as opportunities for deeper engagement with the community and to showcase the company's commitment to innovation [5]

来源： 北北自律机 25年05月02日星期五 理想超充 1 新增。 超充建成数：2270→2271座 需每日 2.54 座，达到 i8发布 目标值 基于2025年底4000+座目标 今年新增数进度值：23.89%→23.93% 基于i8发布日期 2500+座目标 新增数进度值：70.25%→70.38% i8发布剩余90天（按7月31假设） i8发布剩余时间进度值：57.35% ———————————————————— 今年剩余243天 今年时间进度值：33.42% 需每日 7.12 座，达到年底目标值 【附】1 座新增建成 河北省 石家庄市 石家庄尚东绿洲 为城市4C站，规格：4C × 4 加微信，进群深度交流理想长期基本面。不是车友群。 李想将在25年5月7日 20:00播出 AI Talk第二季，推荐读者预约一下。AI Talk 第一季很有含金 量，详见《 理想 AI TALK 》。 ...

来源： 北北自律机 25年05月01日星期四 理想超充 47 新增。 超充建成数：2223→2270座 基于i8发布日期 2500+座目标 新增数进度值：64.17%→70.25% i8发布剩余91天（按7月31假设） i8发布剩余时间进度值：56.87% 需每日 2.53 座，达到 i8发布 目标值 基于2025年底4000+座目标 ———————————————————— 今年新增数进度值：21.82%→23.89% 今年剩余244天 今年时间进度值：33.15% 需每日 7.09 座，达到年底目标值 【附】47 座新增建成 北京市 大兴区 北京金日文化科技园 福建省 泉州 为城市4C站，规格：4C × 6 北京市 海淀区 北京e世界财富中心⭐️ 为城市5C站，规格：4C × 6 5C × 2 福建省 泉州 泉州晋江中国鞋都 为城市4C站，规格：4C × 6 泉州晋江机场油料停车场 为城市4C站，规格：4C × 8 福建省 泉州 泉州鲤城浮桥街 为城市4C站，规格：4C × 6 福建省 泉州 泉州石狮长辉路 为城市4C站，规格：4C × 6 福建省 厦门市 厦门加藤中骏 为城市5C站，规格：2C × 3 ...

Core Viewpoint - Li Auto excels in practical user experience without emotional value, focusing on continuous iteration based on user feedback [1] Group 1: Background and Challenges - Li Auto held a winter driving technology day to address winter driving issues, particularly the discomfort caused by cold cabin temperatures [1][2] - The main challenge in enhancing winter driving comfort lies in the slow heating and uneven temperature distribution within the cabin [1][2] Group 2: Innovative Solutions - The company has developed a self-researched multi-source heat pump system with 51 sensors to ensure effective heating across various conditions [3] - The dual-layer air conditioning box design helps mitigate the common issue of fogging on windows while maintaining cabin warmth [6][7] - Li Auto's heat management system includes a dedicated design for recovering electric drive waste heat, improving energy efficiency by 12% [8] - The design allows for the storage of excess heat in the battery during high-speed driving, which can be utilized later in congested conditions [10] Group 3: Efficiency and Integration - The highly integrated design of the heat management system reduces component count and thermal losses, achieving significant efficiency improvements [11] - The intelligent pre-cooling and pre-heating algorithm ensures optimal battery temperature before charging, with a precision of less than 1°C [13][14] - The innovative cooling design of the battery enhances both cooling efficiency and heating capabilities, improving user experience in cold environments [14]

Core Viewpoint - The article discusses the performance of various AI models, particularly focusing on their capabilities in mathematics and coding, highlighting the strengths and weaknesses of models like Qwen, MiMo, and MindGPT. Group 1: Model Performance - Qwen-7B outperforms MiMo in elementary mathematics tasks, which is unusual given that Qwen is a lower-tier model compared to MiMo [2] - The performance of models in the AIME (American high school mathematics competition) shows a significant disparity, with MiMo scoring high while struggling in other areas [2][5] - The results indicate that the pre-training of models like MiMo is heavily focused on mathematics and coding, potentially at the expense of other capabilities [1] Group 2: Model Comparison - MindGPT is noted to have a much larger parameter size compared to MiMo, making direct comparisons challenging [3] - The strategy of using smaller parameter models for specific metrics is seen as a way to showcase capabilities, although it may not reflect overall performance [3] - There is speculation that MiMo may have utilized distillation techniques for training, which could explain its performance discrepancies [4] Group 3: Community Insights - Discussions within the community suggest that the strategies employed by various teams, including the use of distillation, are common across the industry [7] - The community expresses a desire for genuine performance and capabilities rather than just marketing hype [3]

Core Viewpoint - The article discusses advancements in reinforcement learning (RL) techniques for large language models (LLMs), emphasizing the need for improved algorithms, reward design, and training strategies to enhance reasoning capabilities and model performance. Group 1: Algorithm Improvements - Current algorithms have significant room for improvement, with the introduction of Dr. GRPO addressing issues in GRPO related to response length bias and problem difficulty bias, leading to better token efficiency and reasoning performance [3][4]. - The DAPO method is proposed to tackle entropy collapse and sample efficiency issues in GRPO and PPO, enhancing training stability and efficiency through techniques like Clip-Higher and dynamic sampling [6]. Group 2: Training Strategies - Larger training batch sizes (e.g., TBS = 1024) enhance training efficiency and stability, while on-policy strategies are more advantageous than off-policy ones for model exploration [6]. - Increasing rollout times (e.g., n = 64) improves training outcomes, encouraging longer responses, and a dynamic annealing strategy for KL penalty is recommended to balance exploration and stability [6]. Group 3: Reward Design - Early reward design flaws led to various reward hacking behaviors, necessitating a refined reward system that includes format and answer rewards to constrain model behavior and avoid cheating [6]. - The relationship between response length and reasoning ability is not causal; longer responses may provide more exploration space but do not directly enhance reasoning performance [6]. Group 4: Generalization and Learning - RL is more effective than supervised fine-tuning (SFT) in promoting generalization across tasks, suggesting that reasoning can be a universal capability stimulated by specific tasks [7][9]. - Combining rule-based rewards with reward model-based rewards is beneficial, especially in tasks without clear answers, to enhance learning and mitigate reward hacking [9].

来源： 北北自律机 25年04月30日星期三 理想超充 67 新增。 超充建成数：2182→2249座 （北北记到2243） 318川藏线打通目标，已初步达成 基于i8发布日期 2500+座目标 新增数进度值：58.86%→64.17% i8发布剩余92天（按7月31假设） i8发布剩余时间进度值：56.40% （提示：建设进度优于时间进度） ———————————————————— 需每日 3.01 座，达到 i8发布 目标值 基于2025年底4000+座目标 今年新增数进度值：20.02%→21.82% 今年剩余245天 今年时间进度值：32.88% 需每日 7.25 座，达到年底目标值 【附】41 座新增建成 安徽省 阜阳市 阜阳红星美凯龙(颍州商场) 为城市4C站，规格：4C × 4 福建省 厦门市 厦门集美同集南路 为城市5C站，规格：2C × 3 5C × 1 广东省 广州 广州黄埔智造谷创新园 为城市4C站，规格：4C × 6 广东省 广州 广州立白科技园·IA谷 为城市4C站，规格：4C × 6 广东省 深圳市 深圳市大运体育中心 为城市4C站，规格：4C × 6 广东省 中山市 中山海港城广 ...

方案2：现有方案尾部缩短30-40公分，留出一定的行李空间。 加微信，进群深度交流理想长期基本面。不是车友群。 李想将在25年5月7日 20:00播出 AI Talk第二季，推荐读者预约一下。AI Talk 第一季很有含金量，详 见《 理想 AI TALK 》。 核心需求是：希望既能躺平，又能有空间放东西且不放在床上，厚度也不要太厚，价格也不要太贵。 目前官方商城的L7床垫是直接从前排到后备箱，东西只能放床垫或车外。 第三方床垫有两种，一种是二排与后备箱形成一张床，缺点是不平。 另一种是可以纯平，但太厚，且价格要翻三四倍。 这位车主想到两种方案： 方案1：设计一款厚度薄一些的二排与后备箱纯平的床垫。 来自一位L7车主 ...