Core Insights - The research introduces a new task called Humanoid Visual Search (HVS), enabling models to perform 360-degree visual searches in real-world environments like train stations and shopping malls [6][10][12] - A new benchmark test, H*Bench, has been developed to evaluate the search capabilities of intelligent agents in complex environments, moving beyond traditional simple household scenarios [7][8][9] - The study aims to transition visual spatial reasoning from a "disembodied passive paradigm" to an "embodied active paradigm," enhancing the model's ability to integrate physical actions with visual reasoning [9][12] Group 1: Humanoid Visual Search - HVS allows intelligent agents to autonomously rotate their heads to search for target objects or paths in immersive environments [6][12] - The task focuses on two main search problems: Humanoid Object Search (HOS) and Humanoid Path Search (HPS), each with varying levels of difficulty based on visibility and environmental cues [12][16] - HOS involves locating and focusing on target objects, while HPS requires identifying navigable paths and adjusting body orientation [16][12] Group 2: Benchmark and Dataset - The H* dataset consists of approximately 3,000 labeled task instances derived from diverse high-resolution panoramic videos, providing a comprehensive geographical coverage [21][22] - The benchmark includes scenes from six main categories: retail environments, transportation hubs, urban streets, public institutions, offices, and entertainment venues [24] - The dataset allows for 12,000 search rounds by initializing agents from four different starting directions [22] Group 3: Model Training and Performance - The research utilizes a multi-modal reasoning task, employing a strategy network to integrate tool usage and head rotation, enhancing the model's decision-making capabilities [17][28] - Training results show significant improvements in search accuracy for target search (from 14.83% to 47.38%) and path search (from 6.44% to 24.94%) after model training [28] - The study highlights that larger model sizes do not necessarily guarantee better performance, with smaller models outperforming larger counterparts in certain tasks [33][34] Group 4: Challenges and Insights - The research identifies fundamental bottlenecks in advanced reasoning that require physical, spatial, and social common sense, despite improvements in low-level perception and motion capabilities [34][36] - Errors in HOS primarily stem from insufficient perception in cluttered environments, while HPS errors are more complex, involving a lack of physical and social common sense [36] - The study emphasizes that active visual search (rotating in panoramic views) is more intuitive and effective than passive analysis of static images [36]

Group 1 - The core viewpoint of the news highlights the performance of the Sci-Tech Innovation Artificial Intelligence ETF (589010), which experienced a decline of 1.59% in early trading, indicating ongoing short-term adjustment pressure despite strong support from lower moving averages [1] - The ETF's component stocks showed mixed performance, with only 4 out of 30 stocks rising, while the majority faced downward pressure, particularly companies like Hehe Information and Foxit Software, which fell over 5% [1] - Google announced a partnership with India's largest telecom operator, Reliance Jio, to provide AI services to over 500 million users, including access to the Gemini AI model and cloud storage, indicating a significant move in the AI service market [1] Group 2 - Galaxy Securities reported that global AI giants like Nvidia, OpenAI, and Alibaba are accelerating investments in computing power and ecosystem development, reflecting strong market demand and the rapid commercialization of AI [2] - The Sci-Tech Innovation Artificial Intelligence ETF closely tracks the Shanghai Stock Exchange Sci-Tech Innovation Board AI Index, covering high-quality enterprises across the entire industry chain, benefiting from high R&D investment and policy support [2] - The ETF aims to capture the "singularity moment" in the AI industry, supported by a 20% price fluctuation range and the elasticity of small and mid-cap stocks [2]

Group 1 - Google DeepMind's MoR architecture achieves two times inference speed by combining parameter sharing and adaptive computation, resulting in fewer parameters while maintaining large model performance [1] - The dynamic routing mechanism allocates different recursive depths based on token complexity, reducing redundant computations and optimizing KV cache [1] - Experimental results show that MoR improves inference throughput by 2.06 times, reduces training time by 19%, and decreases peak memory usage by 25% [1] Group 2 - Amazon launches Bedrock AgentCore preview, offering seven core AI agent services including runtime, memory, and authentication [2] - The introduction of Nova customization options and Strands Agents V1.0 simplifies agent development and enables multi-agent collaboration [2] - Amazon S3 Vectors cloud object storage is released, reducing vector storage costs by 90%, along with Kiro AI IDE to enhance developer experience [2] Group 3 - Elon Musk is seeking names for the male AI companion Grok, with suggestions like "Draven" that align with characters from "Twilight" and "Fifty Shades of Grey" [3] - A user named Jackywine has created an open-source 3D digital companion "Bella," which retains only the visual aspect without large language model capabilities [3] - The "Bella" project follows an "AI native" development path in three phases: perception core, generative self, and proactive companionship, with plans to incorporate voice recognition and affinity systems [3] Group 4 - Google Search introduces an AI feature that can make phone calls to book local services for users, such as pet grooming [4] - The search integrates the Gemini 2.5 Pro model and Deep Search functionality, capable of handling complex queries and generating in-depth reports [4] - This new feature has launched in the U.S. and will be gradually rolled out globally, sparking discussions about the effectiveness of AI automated calls and merchant experiences [4] Group 5 - The AI programming platform Windsurf reintroduces the Claude Sonnet 4 model, allowing Pro users 250 free calls per month [6] - Claude Sonnet 4 offers advantages such as cross-file intelligent refactoring, a 200,000 token context window, and precise code completion [6] - This renewed partnership follows OpenAI's acquisition failure and executive team changes, representing Windsurf's strategic move to regain user trust [6] Group 6 - Anthropic successfully rehires core programming leaders Boris Cherny and Cat Wu from Cursor within two weeks [7] - Anthropic reveals that direct sales of models and Claude yield a gross margin of 60%, while sales through AWS and Google Cloud result in a negative 30% margin [7] - Claude Code has become a new asset for Anthropic, with weekly downloads increasing sixfold to 3 million since June, contributing over $200 million in annualized revenue [7] Group 7 - CrePal launches the first AI video creation agent, allowing users to produce videos through a single command that orchestrates multiple models [8] - The system can automatically plan scripts, select appropriate models, generate visuals, and add sound effects, addressing high barriers in traditional AI video creation [8] - The innovation lies in transforming the creative process, enabling users to focus on creative expression rather than technical operations by integrating dispersed tools into a unified intelligent task [8] Group 8 - Apple's MLX framework adds CUDA support, enabling developers to train models using NVIDIA GPUs and deploy them back to Apple devices [9] - This move is seen as Apple's concession to the NVIDIA ecosystem, which dominates AI development with 5 million developers [9] - Despite past tensions over NVIDIA support, Apple opts to leverage NVIDIA's ecosystem for compliance and to expand its influence [9] Group 9 - HeShan Technology, founded by alumni from Tsinghua and Beihang University, focuses on AI tactile sensing technology and has developed the world's first AI tactile perception chip [10] - Utilizing capacitive tomography technology, HeShan achieves "sensing and control integration," addressing the tactile feedback needs in robotic precision operations [10] - The company has completed four rounds of financing and serves over 70% of domestic robot manufacturers, transitioning from a hardware provider to a comprehensive tactile solution provider [10] Group 10 - Nobel laureate John Jumper discusses the journey of AlphaFold, highlighting that the value of algorithm research is 100 times that of data [11] - AlphaFold predicts protein structures with atomic-level precision and has been cited 35,000 times, accelerating scientific discoveries [11] - Jumper predicts that AI4Science will become more generalized in the future, with AlphaFold enhancing the pace of structural biology development by 5-10%, leading to widespread advancements across scientific fields [11]

Group 1: Market Performance - As of July 1, 2025, the Sci-Tech Innovation Board 100 Index (000698) increased by 0.30%, with notable gains from Rongchang Bio (688331) at 10.76%, Xinmai Medical (688016) at 8.22%, and Zai Lab (688266) at 7.26% [3] - The Sci-Tech 100 Index ETF (588030) also rose by 0.30%, with a latest price of 1.02 yuan, and a cumulative increase of 3.99% over the past week as of June 30, 2025 [3] - The trading volume for the Sci-Tech 100 Index ETF reached 2.22 billion yuan, with a turnover rate of 3.52% [3] Group 2: Fund Growth and Leverage - The Sci-Tech 100 Index ETF saw a significant growth of 256 million yuan in scale over the past two weeks, ranking second among comparable funds [4] - The latest margin buying amount for the Sci-Tech 100 Index ETF reached 12.50 million yuan, with a margin balance of 217 million yuan [4] Group 3: Performance Metrics - As of June 30, 2025, the net value of the Sci-Tech 100 Index ETF increased by 13.69% over the past six months, ranking 416 out of 3427 index equity funds, placing it in the top 12.14% [5] - The ETF achieved a maximum monthly return of 27.67% since inception, with the longest consecutive monthly gain of 3 months and an average monthly return of 8.57% [5] - The ETF's Sharpe ratio for the past year was 1.03, indicating a favorable risk-adjusted return [5] Group 4: Fee Structure and Tracking Accuracy - The management fee for the Sci-Tech 100 Index ETF is 0.15%, and the custody fee is 0.05%, which are among the lowest in comparable funds [5] - The tracking error for the ETF over the past six months was 0.021%, demonstrating high tracking precision compared to similar funds [5] Group 5: Index Composition - The Sci-Tech 100 Index is composed of 100 securities selected from the Shanghai Stock Exchange's Sci-Tech Innovation Board, focusing on medium market capitalization and good liquidity [6] - As of June 30, 2025, the top ten weighted stocks in the index accounted for 22.99% of the total index weight, including companies like BeiGene (688235) and Huahong Semiconductor (688347) [6]