春节主要活动：2026.2.2#千问宣布30亿春节请客，2.6开启第一波活动，2.7宣布可以买盒马，2.10接入大 麦，2.17起#为所有人带来「每日首单立减」活动 春节活动效果：①「千问请客」春节活动期间（2.6-2.17），有超过1.3亿人在千问点奶茶、囤年货、买电影票/ 门票、订机酒；#一共说了50亿次"千问帮我"。 ②"春 【国海海外】主要互联网大厂AI Agent春节核心表现梳理 【国海海外】主要互联网大厂AI Agent春节核心表现梳理 春节主要活动：2026.2.2#千问宣布30亿春节请客，2.6开启第一波活动，2.7宣布可以买盒马，2.10接入大 麦，2.17起#为所有人带来「每日首单立减」活动 春节活动效果：①「千问请客」春节活动期间（2.6-2.17），有超过1.3亿人在千问点奶茶、囤年货、买电影票/ 门票、订机酒；#一共说了50亿次"千问帮我"。 春节主要活动：2025.12.28#火山引擎正式成为总台春晚独家AI云合作伙伴，除夕当晚豆包通过总台春晚送出超 10万份科技好礼以及最高8888元现金红包 ②"春节30亿大免单"上线6天（2.6-2.12），#千问帮大家完成下单1.2亿笔。 春 ...

Core Insights - Shenzhen is positioning itself as a global leader in artificial intelligence (AI) by integrating AI into various sectors, aiming for a core industry revenue of 220 billion yuan by 2025 with over 2,600 enterprises [1][13] - The city has established a robust AI innovation ecosystem characterized by high-density innovation entities, with over 93% of R&D investment coming from enterprises [3][13] - Shenzhen's AI applications are being tested in real-world scenarios, with nearly 300 "city + AI" application scenarios addressing urban governance and industrial upgrades [6][13] Group 1: AI Industry Development - Shenzhen's AI industry has maintained double-digit growth, with significant contributions from enterprises that focus on specialized fields such as machine vision and AI chips [3] - The local humanoid robot industry has a 70% local supply rate, enhancing the resilience of the industrial chain [3] - The establishment of the Shenzhen Leading Edge Intelligent Open Research Institute aims to explore cutting-edge technologies and create a core hub for edge intelligence [5] Group 2: Policy and Infrastructure - The "Artificial Intelligence + Advanced Manufacturing Action Plan (2026-2027)" indicates a shift from isolated breakthroughs to a systematic reconstruction of AI's role in the manufacturing sector [5] - Major facilities like the National Supercomputing Center in Shenzhen are being developed to provide accessible intelligent computing power for SMEs [10] - The city is fostering a collaborative ecosystem through initiatives like the "6S" model for AI hardware, which reduces development cycles from months to weeks [11] Group 3: Real-World Applications - Shenzhen is utilizing the entire city as a testing ground for AI, with a focus on practical applications that meet real demands [6] - The introduction of 82 new low-altitude logistics routes by 2025 will enhance the efficiency of drone delivery services [6] - The establishment of the global AI application scenario center in Huaqiangbei aims to create a comprehensive model for AI application and industry integration [7]

Core Viewpoint - By 2025, the paradigm, value, and capabilities of AI will be fully confirmed, leading to significant technological investments, competitive differentiation, and market segmentation in 2026 [1][3]. Group 1: Industry Trends - The technology and internet sectors are experiencing rapid changes, with major companies competing fiercely in computing power and large model applications [2]. - Companies are shifting from a focus on technology arms races to defining scenarios for technology application, emphasizing the need to reconstruct existing business loops or create new interaction entry points [5]. Group 2: Major Company Strategies - Tencent, Alibaba, and ByteDance are heavily investing in AI, with Tencent's annual investment reaching hundreds of billions, Alibaba planning to invest 380 billion over three years, and ByteDance's capital expenditure projected to increase from 150 billion in 2025 to 160 billion in 2026 [3][4]. - Alibaba is developing its own AI chip and deploying large-scale clusters to serve over 400 clients, while Tencent is procuring GPUs and establishing AI research centers [3][4]. Group 3: Market Dynamics - The competition is intensifying, with companies like ByteDance developing their own AI chips and achieving significant daily usage metrics for their models [4]. - The narrative around computing power is shifting, with a focus on achieving greater value from lower energy costs, as exemplified by Alibaba's cloud initiatives [4]. Group 4: Future Outlook - 2026 is anticipated to be a watershed year, with the emergence of multi-modal foundational models leading to a Matthew effect, where only a few general intelligent agents will prevail [5].

Group 1 - Tencent's application "Yuanbao" has launched a live streaming feature for the NBA All-Star Game, marking its first foray into sports event broadcasting [2] - The live streaming feature is integrated with the social section "Yuanbao Pai," allowing users to interact with AI for real-time data, rule explanations, and tactical analysis during the games [2] - This initiative follows the integration of QQ Music and Tencent Video, expanding Tencent's content ecosystem to include sports broadcasting rights [2] Group 2 - The "Red Packet War" during the Spring Festival has evolved into a competition for AI application entry points among major players like Baidu, Tencent, Alibaba, and ByteDance [3] - On the first day of launching the new cash red packet feature, Tencent Yuanbao's daily active users (DAU) surged to 23.99 million, a 2.1 times increase from the previous day [3] - Tencent's mixed team has published research on the importance of context in AI applications, indicating a shift in focus from model training to providing rich and relevant context for tasks [3] Group 3 - Tencent's mixed team suggests that "how to remember" may become a core theme in the development of large models by 2026, emphasizing the need for new architectures and optimization methods [4] - The research on foundational technology may play a crucial role in retaining users in the post-red packet war phase, as AI applications extend into more complex social scenarios [4]

Core Insights - The article discusses the introduction of MoPPS, a new framework for model predictive prompt selection that aims to enhance the efficiency of reinforcement learning fine-tuning for large language models by accurately predicting question difficulty without the need for expensive evaluations from large models [5][26]. Group 1: Training Efficiency - MoPPS significantly reduces computational costs associated with training by minimizing the reliance on large model self-evaluations, achieving up to 78.46% reduction in rollouts compared to traditional methods [15][18]. - The framework accelerates training efficiency by 1.6x to 1.8x compared to conventional uniform sampling methods, ensuring that the most critical questions are selected for training [16][26]. Group 2: Methodology - MoPPS employs a lightweight Bayesian model to predict question difficulty, using a Beta distribution to estimate success rates for each question, which allows for efficient updates based on training feedback [8][9]. - The framework utilizes Thompson Sampling for active question selection, balancing exploration and exploitation to identify questions that are optimally challenging for the model [10][12]. Group 3: Performance Metrics - Experimental results indicate that MoPPS maintains a high correlation between predicted and actual question difficulty, demonstrating its reliability and effectiveness in training scenarios [19][22]. - The framework is compatible with various reinforcement learning algorithms and can adapt to different sampling strategies, enhancing its applicability across different training contexts [20][24]. Group 4: Industry Impact - The research has garnered attention from major industry players such as Alibaba, Tencent, and Ant Group, indicating its potential impact on the field of AI and machine learning [4]. - The MoPPS framework represents a significant advancement in the cost-effective fine-tuning of large models, potentially influencing future developments in reinforcement learning applications [26].

Core Insights - The article discusses the significant advancements in reasoning capabilities of large language models (LLMs) through reinforcement learning fine-tuning, particularly highlighting the high costs associated with inefficient training processes [1][2]. Group 1: Training Efficiency - Traditional training methods like "Uniform Sampling" waste computational resources by randomly selecting questions that do not provide effective learning signals [2]. - The "Dynamic Sampling" approach, while more efficient, still incurs high costs due to the need for extensive self-evaluation by the model [2][6]. - The proposed MoPPS framework aims to dynamically predict question difficulty without the expensive self-evaluation process, thus enhancing training efficiency [3][6]. Group 2: MoPPS Framework - MoPPS utilizes a lightweight Bayesian model to quickly estimate question difficulty, allowing for efficient selection of training data [8][10]. - The framework models each question as a "bandit" problem, using a Beta distribution to estimate success rates based on training feedback [9][10]. - MoPPS introduces a recursive update mechanism that improves difficulty estimation over time, adapting to the model's evolving capabilities [11][13]. Group 3: Performance Improvements - MoPPS has demonstrated a training speed increase of 1.6x to 1.8x while reducing inference costs by up to 78.46% compared to traditional methods [18][21]. - The framework has shown significant advantages across various reasoning tasks, achieving better performance with fewer computational resources [18][21]. - The correlation between predicted and actual question difficulty is high, validating the effectiveness of MoPPS in accurately estimating task challenges [25][29]. Group 4: Versatility and Future Applications - MoPPS is compatible with multiple reinforcement learning algorithms and can adapt to different sampling strategies, enhancing its applicability [26][28]. - The framework's ability to incorporate prior knowledge can further accelerate initial training phases, making it a versatile tool for large-scale model fine-tuning [28][31]. - The research indicates potential for broader applications in the reinforcement learning fine-tuning of larger models in the future [31].

Group 1 - The core theme of the magazine issue is the integration of AI into the workplace, raising the question of who benefits as AI becomes a standard tool [1][2] - AI is seen as a double-edged sword, creating super individuals who effectively utilize tools while also highlighting concerns about talent development gaps [1] - The magazine aims to provide insights into industry trends and personal strategies for navigating the evolving landscape shaped by AI [11] Group 2 - The cover story discusses the implications of AI as a baseline in various industries and identifies potential winners in this new environment [2] - The issue includes a review of significant business news from 2025, highlighting both successes and failures of major companies [7] - Articles feature diverse perspectives, including how young children can use AI and strategies for navigating market conditions [10]

Core Viewpoint - The article discusses the intense competition among major internet companies, particularly Tencent, in attracting top AI talent through high salaries and innovative scholarship programs, highlighting the industry's talent scarcity and the strategic investments being made in AI research and development [1][4]. Group 1: Talent Acquisition Strategies - Tencent is actively recruiting AI talent with high salaries for various positions, such as over 750,000 yuan for user operation roles and nearly 1,000,000 yuan for AI application engineers [1]. - The "Qingyun Plan" is Tencent's initiative aimed at attracting top technical students globally, similar to ByteDance's Top Seed talent program [1]. - The "Qingyun Scholarship" offers significant financial incentives, including 500,000 yuan per recipient, to support students in AI and computer science fields [2]. Group 2: Investment in Research and Development - Tencent's R&D expenditure reached a record high of 22.82 billion yuan in Q3 2025, with a total of 61.983 billion yuan spent in the first three quarters of 2025 [4]. - The company emphasizes the importance of computational resources for top PhD students, providing cloud heterogeneous computing resources as part of the scholarship [4]. Group 3: Recruitment of Established Talent - Tencent is also accelerating the recruitment of established AI experts, as evidenced by the hiring of prominent figures like Pang Tianyu and Yao Shunyu, who have significant academic and industry experience [5]. - The establishment of new departments within Tencent, such as AI Infra and AI Data, aims to enhance its capabilities in large model research and development [5]. Group 4: Academic Collaboration and Knowledge Sharing - Tencent launched its technical blog to share research findings, marking a step towards increasing its academic influence and transparency in AI technology [6].

Core Insights - The competition among major tech companies in the AI sector is intensifying, with a significant focus on attracting top talent from universities and overseas [1][3] - Tencent has launched the "Qingyun Scholarship" to attract top students in AI, offering substantial financial incentives and resources [2][3] Talent Acquisition - Major tech companies are offering high salaries for AI positions, with roles such as "User Operations" at Yuanbao exceeding 750,000 yuan and "AI Application Engineer" at Doubao nearing 1,000,000 yuan [1] - Tencent's "Qingyun Plan" aims to recruit top technical students globally, competing with ByteDance's Top Seed talent program [1][2] Scholarship Program - The "Qingyun Scholarship" awards 15 students a total of 500,000 yuan each, including 200,000 yuan in cash and 300,000 yuan in cloud computing resources [2][3] - The scholarship program received nearly 400 applications and focuses on students in cutting-edge fields like multimodal intelligence and AI infrastructure [2][3] Industry Challenges - The primary challenge in the AI industry is the scarcity of top-tier talent capable of breakthroughs in foundational models and multimodal fields [3] - Tencent emphasizes that the scholarship is designed to support open innovation rather than binding students to employment, aiming to expand the talent pool in the industry [3] Investment in R&D - Tencent's R&D expenditure reached a record high of 22.82 billion yuan in Q3 2025, with total spending for the first three quarters of 2025 amounting to 61.983 billion yuan [3] - The company is also accelerating the recruitment of experienced AI professionals, indicating a sense of urgency in talent acquisition [4] Recent Developments - Notable AI scientists, including former OpenAI researcher Yao Shunyu, have joined Tencent, enhancing its research capabilities [4] - Tencent's technical blog has launched, showcasing research outcomes and enhancing its academic influence in the AI field [5]

Core Insights - The core argument of the article emphasizes that the key bottleneck for models to achieve high-value applications lies in their ability to effectively utilize context [1][5][7]. Group 1: Context Learning Challenges - Recent research indicates that even when context is provided, models may still struggle to solve tasks, highlighting a significant shortfall in their learning capabilities [5][32]. - The article discusses the difference in learning abilities among models, comparing it to individuals with varying talents who learn from the same material [5]. - Current models primarily rely on "parameterized knowledge," which is static and does not adapt to new information from the context [12][34]. Group 2: CL-bench Benchmark - The CL-bench benchmark was developed to assess how well language models can learn new knowledge from context and apply it correctly [16][26]. - It includes 500 complex contexts, 1,899 tasks, and 31,607 validation standards, all designed to require models to learn from the provided context [16][27]. - The benchmark covers four main real-world context learning scenarios: domain knowledge reasoning, rule system application, procedural task execution, and empirical discovery [28][29]. Group 3: Model Performance Evaluation - Evaluation results show that even the best-performing model, GPT-5.1 (High), only solved 23.7% of tasks, indicating a significant gap in context learning capabilities [31][32]. - The majority of errors stem from models ignoring or misusing context, rather than a lack of information [34][35]. - The article notes that models struggle particularly with tasks requiring inductive reasoning from experimental data, often achieving less than 10% success [39]. Group 4: Future Directions - The article suggests that improving context learning could shift the role of humans from data providers to context providers in AI systems [43]. - It raises the challenge of how to make knowledge learned from context persistent, as current models lose this knowledge once the context window is cleared [43][46]. - The potential for models to achieve autonomous learning through effective context learning and memory consolidation is highlighted as an exciting future prospect [47][48].