Core Viewpoint - The article discusses the limitations of existing depth estimation and completion models due to reliance on outdated datasets, highlighting the significance of the newly released LingBot-Depth-Dataset by Ant Group, which provides a large-scale, high-quality RGB-depth dataset to enhance model training and performance in real-world applications [4][5][34]. Group 1: Dataset Overview - Ant Group has open-sourced approximately 3 million pairs of high-quality RGB-depth data, making it one of the largest real-world RGB-D datasets available [5][16]. - The dataset consists of 2.71TB of data, including around 2 million pairs of real RGB-D data and 1 million pairs of high-quality rendered data, covering six mainstream depth cameras [5][6]. - The dataset is structured into four subsets: RobbyReal, RobbyVla, RobbySim, and RobbySimVal, each designed to address specific challenges in depth perception tasks [17][22][24]. Group 2: Importance of Real Data - The article emphasizes the challenges in obtaining high-quality real RGB-D data, including high costs, technical complexities, and the inherent limitations of depth sensors [12][13][14]. - The lack of large-scale real-world RGB-D datasets has created a gap in the field, which the LingBot-Depth-Dataset aims to fill, providing a critical resource for advancing depth estimation technologies [14][34]. - The dataset's design allows models to learn from diverse sensor characteristics, improving their generalization across different hardware environments [19][20]. Group 3: Impact on the Industry - The introduction of the LingBot-Depth-Dataset is expected to shift the focus from model complexity to data quality, as the performance of models is increasingly determined by the quality and quantity of training data [31][32]. - This dataset could serve as a new benchmark for depth estimation and completion, similar to how ImageNet transformed visual recognition [34][35]. - By providing a comprehensive dataset, Ant Group enables research teams to concentrate on higher-level problems without the need to collect data from scratch, fostering innovation in the field [36].

Core Insights - The article discusses the limitations of traditional Vision-Language-Action (VLA) models in accurately interpreting complex spatial instructions and proposes a new method called Point-VLA to overcome these challenges [5][27]. Group 1: Limitations of Traditional VLA Models - Language often fails to express certain spatial scenarios accurately, leading to ambiguity in communication [6][8]. - Even when detailed descriptions are provided, VLA models struggle to generalize and execute complex spatial commands, resulting in low success rates [7][20]. - Advanced Visual-Language Models (VLM) can achieve 60-70% accuracy in locating targets based on complex text descriptions, but text-only VLA models have a success rate of only around 25% [14][9]. Group 2: Introduction of Point-VLA - Point-VLA introduces visually grounded instructions by overlaying bounding boxes on images, allowing robots to understand commands more intuitively, similar to human pointing [10][11]. - This method combines high-level intentions expressed in language with precise spatial information encoded visually, enhancing the model's performance [12][15]. Group 3: Experimental Results - Point-VLA achieved an impressive average success rate of 92.5% across various challenging tasks, significantly outperforming the 32.4% success rate of traditional text-only VLA models [20][19]. - In specific tasks, such as cluttered scene grasping, Point-VLA's success rate improved from 43.3% to 94.3%, demonstrating its effectiveness in real-world applications [20][23]. Group 4: Data Annotation and Scalability - The development of an automated data annotation pipeline allows for efficient generation of visual grounding signals, reducing the cost of acquiring training data [18][27]. - As training data increases, Point-VLA's performance continues to improve, while traditional text-only VLA models reach a performance plateau [25][30]. Group 5: Implications for Future Development - Point-VLA addresses a fundamental issue in the VLA field by bypassing the limitations of language expression, paving the way for new advancements in VLA models [27]. - The demonstrated capabilities of Point-VLA provide a technical foundation for practical applications in industrial and service sectors, highlighting the effectiveness of human-like interaction methods in human-robot collaboration [27][29].

Market Overview - The Hang Seng Index is expected to rise to 25,700 points as geopolitical tensions ease and oil prices retreat, alleviating inflation concerns [1] - The market sentiment is supported by the resumption of shipping by certain Chinese companies in the Middle East and the anticipated visits between US and Chinese leaders [1] - However, geopolitical uncertainties may persist, and corporate earnings are likely to face challenges from rising commodity prices and shipping costs [1] Company News - Industrial robot manufacturer Huichuan (300124.SZ) plans to list in Hong Kong, aiming to raise 15.6 billion HKD [3] - Bank of China (3988) reported a 2% increase in profit for the previous year [3] - Agricultural Bank of China (1288) saw a 3% profit increase year-on-year [3] - Wallen Technology (6082) experienced a 207% surge in revenue last year [3] - Sunny Optical (2382) reported a 72% increase in net profit, with a final dividend doubling to 120.6 HKD cents [3] - China Shenhua (1088) reported a 9% decline in net profit [3] Economic Outlook - The US Federal Reserve maintained interest rates, reflecting a cautious stance on monetary policy, with economic growth forecasted to rise by 0.1 percentage points to 2.4% [4] - The unemployment rate is projected to remain at 4.4%, while inflation expectations have been adjusted upward to 2.7% due to geopolitical uncertainties [4] - The geopolitical situation in the Middle East continues to impact oil prices, with expectations of a mid-term retreat from high levels [4] Sector Focus - AI stocks are experiencing rapid growth due to intensive upgrades in large models, contributing to the semiconductor industry's expansion [6] - Electric vehicle manufacturers are expected to report improved sales figures for March, indicating a monthly recovery [6] - The innovative drug sector is seeing strong performance from CXO companies, with significant academic conferences scheduled for the second quarter to validate R&D progress [6] Regulatory Developments - The State Administration for Market Regulation is focusing on preventing "involution" competition in key sectors such as platform economy, photovoltaic, lithium batteries, and new energy vehicles [7] - Measures will be implemented to address unfair competition practices among platform operators, including price manipulation and forced sales below cost [7] - The administration is also working on a new regulation to prohibit unfair competition in the digital space, aiming to balance data protection and utilization [7]

Core Viewpoint - The article emphasizes the importance of practical learning experiences in cutting-edge technology sectors, highlighting the value of direct engagement with leading companies and industry experts to foster innovation and investment opportunities [4][13]. Schedule Overview - The schedule for 2025 includes visits to major technology companies and events such as CES, with a focus on AI, robotics, and commercial space [7][8]. - Key events include closed-door research meetings on China's AI capabilities and participation in the Hong Kong Web3 Carnival [8][9]. Learning Experience - Participants will engage in deep explorations of technology companies, gaining insights into strategic decisions, technological challenges, and industry disruption logic through direct dialogues with founders and executives [13][24]. - The program aims to empower entrepreneurs by focusing on three dimensions: cutting-edge technology trends, emerging industry ecosystems, and business model exploration [13]. Past Activities - In 2023, participants visited leading companies such as Huawei and ByteDance, while in 2024, they will explore firms like BYD and Tencent, focusing on themes like artificial intelligence and new energy [24][25]. - The program has a history of fostering connections among entrepreneurs and investors, enhancing their understanding of macroeconomic trends and investment strategies [48][49].

Core Insights - The article discusses the challenges and solutions related to KV cache compression in long-context reasoning for Vision-Language Models (VLM) and Large Language Models (LLM) [1][2][42] - It introduces MixKV, a method that combines importance and diversity in KV cache selection to enhance stability and coverage in compressed contexts [5][13][42] Group 1: KV Cache Challenges - The lengthening of context leads to linear expansion of KV cache, resulting in increased memory usage and bandwidth costs, which negatively impacts throughput [3][5] - Traditional compression methods often focus solely on "importance," neglecting the inherent "semantic redundancy" present in multimodal KV caches, which can lead to instability [5][12] Group 2: Key Findings - The research team visualized the statistical properties of KV, revealing that multimodal inputs exhibit a higher degree of semantic redundancy, indicating a larger compressible space [8][10] - There are significant differences in redundancy levels across different heads within the same model, suggesting a non-uniform distribution of redundancy [10][12] Group 3: MixKV Solution - MixKV aims to retain KV entries that are both important and diverse, thereby reducing the risk of losing semantic coverage due to redundancy [13][23] - The method consists of two scoring steps (importance and diversity) and a head-wise mixing approach to adaptively balance the two factors based on redundancy levels [14][15][16] Group 4: Experimental Results - MixKV demonstrated consistent performance improvements across various benchmarks in multimodal understanding, long-context reasoning, and GUI localization tasks [25][29][37] - The method showed significant efficiency gains, reducing inference latency and peak memory usage under extreme compression conditions [41][42] Group 5: Conclusion - MixKV represents a critical upgrade for KV cache compression in long-context reasoning, emphasizing the need to consider redundancy structures in the design paradigm for scalable deployment of VLMs and LLMs [42]

Group 1 - Huawei has recruited top photonics expert Martin Schell from Germany's Fraunhofer HHI to lead its Prague R&D center, highlighting China's attractive opportunities for top talent in certain tech fields [3] - iQIYI has submitted a listing application to the Hong Kong Stock Exchange and approved a share buyback plan worth up to $100 million [4] - Naixue's Tea reported a revenue of 4.331 billion yuan for 2025, a year-on-year decline of 11.99%, with a net loss of 239 million yuan, although the loss narrowed by 73.94% [6] - Xiaomi's founder Lei Jun announced a significant investment of 16 billion yuan in AI research and development this year, with a recruitment drive for AI talent officially launched [6] Group 2 - SF Holding reported a revenue of 308.2 billion yuan for 2025, a year-on-year increase of 8.4%, and a net profit of 11.1 billion yuan, up 9.3% [9] - Epic Games announced layoffs affecting over 1,000 positions due to declining player engagement in "Fortnite," impacting its China team [9][10] - The annual recurring revenue of "Moonlight Dark Side" has surpassed $100 million, with its valuation increasing fourfold within three months to $18 billion [12] - Faraday Future's founder Jia Yueting announced the company exceeded its delivery target for robots in March, with 22 units sold against a target of 20 [12] Group 3 - Mistral AI raised $830 million to build a data center in France, planning to purchase 13,800 GB300 chips from NVIDIA [16] - Rebellions, a Korean AI chip startup, raised an additional $400 million, bringing its total funding to $850 million and its valuation to approximately $2.34 billion [17] - This Chip Technology completed nearly 1 billion yuan in Series B financing, led by a strategic investment from a Shanghai state-owned platform [18] - Zero Gravity Aircraft Industry announced a completion of 150 million yuan in Pre-B financing, aimed at product development and innovative business model construction [18] Group 4 - BYD launched the Song Ultra EV with a starting price of 151,900 yuan, featuring the second-generation blade battery and fast-charging technology [21][22] - SAIC Volkswagen's ID.ERA9X is now available for pre-sale, with prices ranging from 329,800 to 379,800 yuan [24] - Geely's Galaxy Star 8 series was launched with prices between 142,800 and 172,800 yuan, featuring advanced driver assistance systems [26] - Toyota's global production fell for the fourth consecutive month in February, with a 3.9% year-on-year decline to 749,673 vehicles [29]

Core Insights - The article emphasizes that great companies are built over time through various critical milestones rather than a single moment, highlighting the importance of continuous support and investment in early-stage entrepreneurs [1] - The company introduces three initiatives: "Jinqiu Lab" for institutionalized joint entrepreneurship, "Jinqiu Soil" for seed investments in early-stage startups, and "Jinqiu Grow" as an AI-native post-investment empowerment platform [2][8] Group 1: Jinqiu Lab and Soil - "Jinqiu Lab" aims to provide essential resources for tech innovators, facilitating the journey from concept to market [1] - "Jinqiu Soil" focuses on seed investments, supporting early-stage entrepreneurs from idea validation to company formation [2] Group 2: Jinqiu Grow - "Jinqiu Grow" is designed to offer a scalable and sustainable post-investment service, utilizing AI to streamline resource allocation and support for portfolio companies [5][30] - The platform allows CEOs to access resources on-demand, enhancing the efficiency of talent recruitment and brand support [9][13] Group 3: AI Integration and Ecosystem - The integration of AI into the post-investment service aims to create a low-friction system for resource connection, talent acquisition, and community engagement [5][14] - The company has successfully saved portfolio companies nearly $10 million in model token costs through partnerships with major cloud service providers [10] Group 4: Testimonials and Impact - Testimonials from various CEOs highlight the effectiveness of Jinqiu's support in areas such as talent recruitment, resource connection, and overall partnership [16][22][24] - The feedback emphasizes the company's proactive approach and deep understanding of the entrepreneurial journey, fostering a collaborative environment [18][20][30]

Investment Rating - The report maintains a "Positive" investment rating for the storage industry, particularly in relation to the implications of the TurboQuant algorithm on storage demand [2]. Core Insights - The report discusses the recent Google paper on TurboQuant, which has sparked debates regarding storage demand, suggesting that the excitement may be overstated and that TurboQuant may represent a conventional academic advancement rather than a groundbreaking change in storage technology [4][12]. - It emphasizes the need for investors to understand the nuances of TurboQuant, including its operational mechanics and potential limitations, particularly in terms of its application in various scenarios [4][24]. - The report highlights that while TurboQuant claims significant performance improvements, the actual benefits may not be as pronounced as suggested, particularly when compared to existing methods [25][26]. Summary by Sections 1. Background and Context - The report outlines the context of the TurboQuant paper, noting that media coverage has often been more aggressive than the original research, which presents a more tempered view of its innovations [4][9]. - It identifies that previous algorithms like PolarQuant and RaBitQ have laid the groundwork for TurboQuant, suggesting that the latter may not be as revolutionary as portrayed [12][13]. 2. TurboQuant Overview - The report provides a detailed summary of the TurboQuant algorithm, explaining its methodology and the theoretical underpinnings that guide its design [16]. - It describes the algorithm's focus on minimizing mean squared error (MSE) and optimizing inner product calculations, which are critical for its performance [16][18]. 3. Advantages and Disadvantages - The report discusses the advantages of TurboQuant, such as its potential for significant memory compression, but also highlights critical drawbacks, including its limited applicability to certain types of processing and potential accuracy trade-offs [24][25]. - It notes that TurboQuant primarily compresses KV-Cache without addressing other components like model weights, which remain a significant factor in overall memory usage [24]. 4. Broader Implications - The report suggests that while TurboQuant may not drastically alter storage demand, it raises important questions about the alignment of interests across different segments of the storage industry [28]. - It emphasizes the importance of understanding the diverse technological approaches within the AI-native storage landscape, which may lead to varying preferences among manufacturers [29][30]. 5. Academic Contributions and Insights - The report concludes by recognizing the academic contributions of the TurboQuant paper, particularly its innovative approach to applying digital communication theory to optimize storage solutions [31][32]. - It encourages further exploration of these theoretical frameworks as they may yield significant advancements in the field [31].

Group 1 - The Core View: The Science and Technology Innovation Entrepreneurship AI Index (932456.CSI) was officially launched on May 14, 2025, to reflect the overall performance of large and mid-cap growth companies with core artificial intelligence attributes in the Sci-Tech Innovation Board and the Growth Enterprise Market [1] - The index aims to capture the performance of companies involved in AI foundational resources, technology development, and application support, highlighting the characteristics of balancing domestic and overseas computing power chains [9] - The index is composed of 50 securities selected based on liquidity and market capitalization, focusing on high-elasticity computing power targets [9] Group 2 - Highlight 1: The top-level design of the "14th Five-Year Plan" anchors long-term beta for the AI chain, addressing core constraints in AI development and promoting large-scale application [10] - Highlight 2: The AI industry cycle is transitioning from the training phase to the inference phase, with significant capital expenditure from cloud vendors continuing to expand [14][17] - Highlight 3: The anticipated reversal of "stagflation" expectations may lead to greater elasticity in technology styles, with historical data showing that tech stocks often rebound first after such expectations dissipate [26][28] Group 3 - Highlight 4: The index focuses on large and mid-cap growth styles, with a market capitalization structure dominated by companies with over 100 billion in market value, providing strong foundational support [36] - Highlight 5: The index achieves risk balance across markets, with a reasonable distribution of core technology sectors, effectively avoiding excessive concentration in a single market or sector [41][45] - Highlight 6: The index is heavily weighted towards upstream sectors while also considering downstream applications, capturing the full-cycle benefits of the AI industry [48] Group 4 - Highlight 7: The AI industry's prosperity continues to validate the index's strong earnings growth expectations, with projected net profit growth significantly outperforming mainstream broad-based indices [56] - Highlight 8: The index exhibits high return elasticity and a favorable risk-return ratio, with a historical annualized return of 50.02% and a Sharpe ratio of 1.18, indicating strong risk compensation [60]

Investment Rating - The industry investment rating is "Outperform the Market" [6][8]. Core Insights - The daily average token call volume in China has surged from 100 billion at the beginning of 2024 to over 140 trillion by March 2025, indicating a shift from "model competition" to "application competition" in the AI industry [11][12]. - Momenta, a leading provider of intelligent driving solutions, has submitted its IPO application to the Hong Kong Stock Exchange, aiming for a valuation exceeding 100 billion yuan and plans to list by 2026 [14][16]. - Tesla's Optimus Gen3 is set to begin production in the summer of 2026, showcasing significant advancements in its dexterous hand and gearbox technology [19][20]. Summary by Sections Token Call Volume Growth - The daily average token call volume in China has increased dramatically, with a growth rate exceeding 1000 times over two years, reflecting the rapid development of the AI industry and the establishment of a new value system around tokens [11][12][13]. - The increase in token consumption is driven by user demand for productivity tools like OpenClaw and emerging applications such as video generation [12][13]. Momenta's IPO and Market Position - Momenta has established itself as a top player in the high-level intelligent driving solutions market, with nearly 700,000 vehicles equipped with its technology and partnerships with over 170 vehicle models [14][16]. - The company is pursuing a dual strategy of mass production of assisted driving and fully autonomous driving, leveraging data-driven technology iterations [17][18]. Tesla's Optimus Gen3 Developments - The Optimus Gen3 features significant upgrades, including a dexterous hand with 22 degrees of freedom and a new gearbox design aimed at improving efficiency and precision [19][20]. - Tesla's CEO has confirmed that the production of Optimus Gen3 will commence in summer 2026, with large-scale production expected by 2027 [20][21].