美股三大指数集体低开，道指跌0.08%，标普500指数跌0.1%，纳指跌0.1%。英伟达涨0.08%，公司将于 盘后公布最新财报；热门中概股多数下跌，理想汽车跌逾6%。 (本文来自第一财经) ...

Core Viewpoint - The use of NOA (Navigation on Autopilot) has been increasingly restricted on highways due to concerns over safety and 5G signal limitations, leading to a significant reduction in its availability for drivers [2][5][6]. Group 1: NOA Usage and Restrictions - Many drivers have reported a decrease in the availability of NOA on highways, with one driver noting that out of a 438-kilometer journey, only 68 kilometers allowed for NOA usage [2][4]. - The limitations on NOA usage are attributed to 5G signal issues rather than direct restrictions from car manufacturers or highway management [5][6]. - There is a growing concern among drivers regarding the safety of using NOA, especially after several accidents involving improper use of the system [6][7]. Group 2: Driver Behavior and Safety Concerns - Reports indicate that many drivers misuse NOA by becoming overly reliant on the system, leading to dangerous situations such as not paying attention while driving [6][7]. - A significant increase in accidents related to smart driving systems has been noted, with a 217% year-on-year rise in incidents involving driver over-reliance on these systems [7]. - Experts emphasize the importance of driver responsibility and the need for a balance between convenience and safety in the use of autonomous driving features [8][9]. Group 3: Industry Perspectives and Future Directions - Industry experts suggest that the current state of NOA and similar systems requires significant improvement, with calls for better integration of safety measures and technology advancements [8][9]. - The development of autonomous driving technology is seen as a complex challenge, requiring a combination of data-driven approaches and regulatory frameworks to ensure safety [8][9]. - There is a consensus that future advancements in smart driving must prioritize safety and the understanding of human behavior in conjunction with technological capabilities [9].

Group 1 - The Nasdaq Golden Dragon China Index fell over 2% on Wednesday, indicating a broad decline in popular Chinese concept stocks [1] - JD.com and Futu Holdings dropped over 3%, while Alibaba, Pinduoduo, NetEase, and Tencent Music saw declines of over 1% [1] - Li Auto experienced a significant drop of over 8%, with Xpeng down over 6% and NIO falling over 5% [1]

Core Insights - The company aims to achieve a target of over 4000 supercharging stations by the end of 2025, with a current count of 3144 stations, leaving 856 stations to be built [1] - The progress for new stations this year has increased from 61.94% to 62.34%, with 126 days remaining in the year [1] - To meet the year-end target, the company needs to build an average of 6.79 stations per day [1] Summary by Sections New Supercharging Stations - Nine new supercharging stations have been completed, including locations in Beijing, Guangdong, Guangxi, Ningxia, Shandong, Shaanxi, Tianjin, and Yunnan [1][2] - The specifications for the new stations vary, with some being 4C and others 5C, indicating different charging capabilities [1][2] Current Progress - The current progress towards the target of 4000+ stations is at 62.34%, with a time progress value of 65.48% for the year [1] - The company is on track but needs to accelerate the pace of new station construction to meet its goals [1]

Market Overview - The three major US stock indices opened lower, with the Dow Jones down 0.04%, the S&P 500 down 0.06%, and the Nasdaq down 0.18% [1] - The Nasdaq Golden Dragon China Index fell by 1.9% [1] Company Performance - Intel shares decreased by 0.66% [1] - Qualcomm shares dropped by 0.51% [1] - Li Auto's stock fell by 5.11% [1] - Xpeng Motors' stock declined by 3.82% [1]

Core Viewpoint - The article discusses the contrasting sales performance of electric vehicle manufacturers, highlighting how Leap Motor has achieved significant sales while Li Auto lags behind Xiaomi, despite claiming a "low start, high rise" strategy [1] Group 1: Sales Performance - Leap Motor has become the top seller among new energy vehicle manufacturers, indicating strong market acceptance and demand for its products [1] - Li Auto's sales have fallen behind Xiaomi, raising questions about its growth strategy and market positioning [1] Group 2: Market Strategy - Li Auto describes its current performance as a "low start, high rise," suggesting a long-term growth strategy despite short-term sales challenges [1] - The article implies that the competitive landscape is shifting, with companies like Leap Motor and Xiaomi gaining traction at the expense of Li Auto [1]

国金证券指出，由于9月降息或已成定局，考虑到港股今年相对A股的超额已经大幅回吐，A-H市场将重 回统一起跑线，企业盈利变化将成为两地市场表现差异的驱动。 华泰证券发布港股策略研报称，往后看，外资依然有继续增配中国市场空间：1)海外流动性易松难紧， 不仅因为货币政策，还有金融监管、发债久期调整等，美元流动性可能趋势偏松；2)国内基本面预期改 善，人民币汇率仍有升值空间。但也需要注意的是，当前外资在港股市场重要性已经有所下降，南向资 金在互联互通标的中成交占比已经超过40%，其未来流入的持续性同样甚至更加值得关注。 恒生指数高开0.4%，恒生科技指数涨0.55%。蔚来(09866)涨超8%，理想汽车(02015)涨超2%，比亚迪 (01211)股份涨超1%。 关于港股后市 中泰国际指出，尽管港股估值已经大幅修复，恒生指数预测PE已修复至过去七年接近80%分位数，但考 虑到外部货币政策不确定性有所下降，内部"反内卷"、产业扶持等政策仍在加码，市场仍具备上行支 撑。 本文转载自"腾讯自选股"，智通财经编辑：徐文强。 ...

Core Viewpoint - The resurgence of stock trading among listed companies in the A-share market, with significant investments planned for securities trading and wealth management products, despite some companies terminating their plans shortly after announcement [1][2][8]. Group 1: Investment Plans - Jiangsu Guotai planned to use up to 138.3 billion yuan for wealth management and securities investment but terminated part of the plan shortly after [1][2]. - Other companies like Liou Co., Fangda Carbon, and Qipilang also announced plans to invest over 10 billion yuan in stock trading [1][2][3]. - Yidian Tianxia increased its planned investment from 1 million yuan to 5 million yuan for securities trading [3]. Group 2: Historical Context - Jiangsu Guotai has a history of stock trading for over ten years, with significant investments recorded in previous years [5][6]. - Liou Co. has also been involved in stock investments since 2016, with a notable increase in investment amounts over the years [6][7]. Group 3: Performance and Returns - Jiangsu Guotai reported a cumulative fair value change loss of 71.96 million yuan in the first half of the year, with total losses exceeding 200 million yuan over recent years [8][10]. - Liou Co.'s investment in Li Auto resulted in significant fluctuations in returns, with a peak profit of 60 billion yuan in 2020, but a loss of 4.41 billion yuan in 2022 [10]. - Seven Wolves reported that their securities investment accounted for over 70% of their total profit, despite facing a decline in overall performance [10][11].

Core Viewpoint - The article discusses the advanced technologies used in Li Auto's autonomous driving solutions, specifically focusing on the "MoE + Sparse Attention" efficient structure that enhances the performance and efficiency of large models in 3D spatial understanding and reasoning [3][6]. Group 1: Introduction to Technologies - The article introduces a series of posts that delve deeper into the advanced technologies involved in Li Auto's VLM and VLA solutions, which were only briefly discussed in previous articles [3]. - The focus is on the "MoE + Sparse Attention" structure, which is crucial for improving the efficiency and performance of large models [3][6]. Group 2: Sparse Attention - Sparse Attention limits the complexity of the attention mechanism by focusing only on key input parts, rather than computing globally, which is particularly beneficial in 3D scenarios [6][10]. - The structure combines local attention and strided attention to create a sparse yet effective attention mechanism, ensuring that each token can quickly propagate information while maintaining local modeling capabilities [10][11]. Group 3: MoE (Mixture of Experts) - MoE architecture divides computations into multiple expert sub-networks, allowing only a subset of experts to be activated for each input, thus enhancing computational efficiency without significantly increasing inference costs [22][24]. - The article outlines the core components of MoE, including the Gate module for selecting experts, the Experts module as independent networks, and the Dispatcher for optimizing computation [24][25]. Group 4: Implementation and Communication - The article provides insights into the implementation of MoE using DeepSpeed, highlighting its flexibility and efficiency in handling large models [27][29]. - It discusses the communication mechanisms required for efficient data distribution across multiple GPUs, emphasizing the importance of the all-to-all communication strategy in distributed training [34][37].

Core Insights - The article emphasizes that the capabilities of Li Auto's MindGPT 3.1 are significantly underestimated, highlighting three main anchors of value [1] - MindGPT 3.1's ASPO incorporates innovative optimizations from DeepSeek R1's GRPO, showcasing Li Auto's ability to rapidly learn and internalize the best practices in AI [1][8] - There is a lack of in-depth discussion about Li Auto's technology in the information ecosystem, indicating a potential undervaluation of its advancements [1] Performance Metrics - MindGPT 3.1 is a fast reasoning language model, achieving speeds of up to 200 tokens per second, nearly five times faster than MindGPT 3.0, which is a significant improvement compared to GPT-4's maximum of 79.87 tokens per second [2][4] - The model shows notable enhancements in tool invocation accuracy, complex task completion rates, and response richness compared to its predecessor [4] Benchmarking Results - MindGPT 3.1 outperforms other models in various benchmark tests, achieving high scores in both deep and non-deep thinking modes across multiple assessments [4][5] - In deep thinking mode, MindGPT 3.1 scored 0.8625 in AIME 2024, indicating strong performance relative to competitors [4] Learning Methodology - The ASPO method addresses the issue of data sampling precision, focusing on filtering low-quality learning signals to enhance model training [8][9] - Unlike GRPO, which operates at the output stage, ASPO manages the training pool at the input stage, ensuring that only samples that match the model's capability are used [8][9] Strategic Focus - Li Auto's leadership emphasizes that the primary focus is on enhancing model capabilities rather than artificially inflating benchmark scores, which they consider a waste of resources [5][6] - The company is committed to improving user experience by reducing reasoning time and enhancing the overall quality of responses from the model [5] Collaborative Initiatives - Li Auto has initiated a joint fund with local scientific committees to engage with academic professionals, aiming to gather the latest research insights without specific deliverable requirements [10]