Macro and Strategy - In March 2026, net capital outflow totaled 33.9 billion yuan, contrasting with a net inflow of 13 billion yuan in the previous week [7] - Short-term sentiment indicators are at a mid-high level since 2005, while long-term sentiment indicators are at a mid-low level [8] - The highest trading volume sectors in the past week were power equipment, communication, and semiconductors, while textiles, real estate, and food processing had the lowest [8] Computer Industry - Alibaba Cloud reported Q3 FY2026 revenue of 43.284 billion yuan, a year-on-year increase of 36%, with external commercialization revenue growing by 35% [9] - AI-related product revenue has seen triple-digit growth for ten consecutive quarters, with a target of exceeding 100 billion USD in annual revenue from cloud and AI commercialization over the next five years [9] - The growth driver for Alibaba Cloud has shifted from internal support to a resonance of internal and external demand, confirming a structural upward trend driven by AI [9] Internet Industry - Tencent's QClaw has officially entered public testing, and Xiaomi has launched three large models aimed at the Agent era [10] - The AI sector is witnessing significant investments, with major companies increasing their capital expenditures, talent recruitment, and marketing expenses related to AI [11] - The report suggests maintaining observation on internet giants, particularly those leading in large models and computing power supply chains [11] Gold Mining Industry - Zijin Mining International reported a revenue of 5.383 billion USD for 2025, a year-on-year increase of 80.05%, with a net profit of 1.602 billion USD, up 232.71% [12] - The company plans to produce approximately 59.2 tons of gold in 2026, a 26% increase from 2025, not accounting for potential acquisitions [13] - The report highlights the company's focus on both organic growth and external acquisitions to sustain rapid growth in gold production [13] Real Estate and Asset Management - China Merchants Jinling reported a revenue of 19.27 billion yuan for 2025, a 12.2% increase, but a net profit decline of 22.1% due to one-time impairments [15] - The property management segment achieved a revenue of 18.6 billion yuan, growing by 12.8%, while asset management revenue slightly decreased [15] - The company secured new contracts worth 4.48 billion yuan in the residential sector, marking a 59.6% increase [16] Chemical Industry - Yuntianhua's Q4 2025 revenue was 10.82 billion yuan, down 27% year-on-year, with a net profit decline of 53% due to reduced demand and high raw material costs [18] - The company reported a significant increase in sulfur prices, which pressured profitability, while maintaining a strong cost control capability across its supply chain [19] - The report indicates that Yuntianhua's phosphate rock supply remains tight, supporting its competitive position in the market [19]

Core Insights - Xiaomi's AI team, in collaboration with Peking University, has recently published a paper focusing on MoE (Mixture of Experts) and reinforcement learning, revealing new advancements in large model training [2][8]. Group 1: Research Findings - The paper proposes a novel approach to enhance the stability and efficiency of large model reinforcement learning within the MoE framework [8][10]. - Current reinforcement learning methods face challenges in balancing efficiency and stability, often leading to catastrophic failures during training [14][24]. - The research introduces a method called Rollout Routing Replay (R3), which locks the routing distribution during inference and reuses it during training, ensuring consistency between the two phases [30][31]. Group 2: Experimental Results - Experiments conducted on the Qwen3-30B-A3B model demonstrate that R3 consistently outperforms other methods across various metrics, achieving higher scores in multiple scenarios [41][42]. - The introduction of R3 significantly reduces the occurrence of training crashes, maintaining a stable performance curve even after extended training periods [44][48]. - R3 not only stabilizes the model but also accelerates the optimization process, allowing for quicker identification of effective strategies [50]. Group 3: Team and Contributors - The research team includes notable contributors such as Wenhan Ma, a researcher from Xiaomi's LLM-Core team, and Luo Fuli, who has a strong academic background and has previously worked on significant AI projects [52][59]. - The paper also acknowledges the contributions of Professor Sui Zhifang from Peking University, who has extensive experience in computational linguistics and AI research [62][66].

Core Insights - Xiaomi's large model team had 10 research papers accepted at the 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025), showcasing the company's commitment to advancements in natural language processing and computational linguistics [1][2] Group 1: Conference Details - The ACL conference is a prestigious annual event organized by the International Association for Computational Linguistics, recognized as an A-level conference by the China Computer Federation (CCF) [1] - The 63rd ACL meeting will take place from July 27 to August 1, 2025, in Vienna, Austria [1] Group 2: Research Contributions - The accepted papers cover various cutting-edge topics, including new position encoding methods that enhance context awareness and extrapolation capabilities, and a hybrid framework that optimizes long-context reasoning through customized KV caching [2] - The research team also proposed a dynamic prompt update method to address fixed thinking patterns during instruction expansion [2] - Significant advancements were made in visual language models for multi-image scenarios, KV cache compression, and web agents, including a focus-centered visual chain paradigm that improves performance in multi-image contexts [2]