Core Viewpoint - The article discusses the breakthrough of One-Shot Critique Fine-Tuning (One-Shot CFT) in enhancing reasoning capabilities of large language models (LLMs) with minimal data and computational resources, outperforming traditional reinforcement learning (RL) methods and small-scale supervised fine-tuning (SFT) approaches [1][3][14]. Group 1: One-Shot CFT Methodology - One-Shot CFT is a new method that allows models to learn reasoning by analyzing the quality of answers rather than merely imitating them, thus providing a deeper learning signal [3][12]. - The process involves selecting a representative task, generating multiple answers using various models, and then having a more powerful model critique these answers, which serves as the supervision signal for training [4][5]. - The entire training process requires only one question, multiple answers, and critiques, taking approximately 5 GPU hours, significantly less than RL methods [5][14]. Group 2: Performance and Results - In experiments, Qwen2.5-Math-7B achieved a 15% accuracy increase after One-Shot CFT fine-tuning on a single question, surpassing both RL and full supervised fine-tuning models that used tens of thousands of training samples [9][10]. - The method demonstrated strong performance across various mathematical and logical reasoning tasks, with accuracy improvements ranging from 10% to 16% in specific sub-tasks [10][11]. - One-Shot CFT showed stability and reproducibility across different tasks and model configurations, indicating its robustness [11][13]. Group 3: Advantages of One-Shot CFT - The method emphasizes critical learning, allowing models to understand why answers are correct or incorrect, which enhances the depth of learning compared to traditional SFT [12]. - It introduces multi-perspective inputs by generating multiple answers and critiques for a single task, closely mimicking human learning processes [12]. - The training signals from critiques are highly generalizable, reducing the risk of overfitting and allowing for easier transfer to new tasks [12]. Group 4: Accessibility and Practical Implications - One-Shot CFT's low computational cost makes it accessible for individual researchers, resource-limited labs, and startups, providing a cost-effective solution for enhancing reasoning capabilities [14][15]. - The entire process is open-source, including training scripts, model parameters, and datasets, which significantly lowers the barrier for replication and experimentation [17].

Core Insights - Apple and Meta are intensifying their efforts in AI, realizing its potential to disrupt device experiences and advertising models [1][2] - Both companies face challenges in talent acquisition and strategic direction, risking marginalization in the AI landscape [3][12] Group 1: AI Competition and Acquisitions - Apple and Meta are competing against AI giants like Microsoft, Amazon, Google, and OpenAI, with significant valuations for potential acquisition targets such as Perplexity at $14 billion and Thinking Machines Lab at $10 billion [2][23] - Meta has acquired nearly half of Scale AI for $14.3 billion and is considering other acquisitions like SSI, valued at $32 billion, and several other AI companies with valuations ranging from $4.5 billion to $62 billion [2][21] Group 2: Strategic Challenges - Both companies are struggling with a lack of direction and talent, leading to confusion in strategic execution [3][12] - Apple has not delivered substantial AI innovations at its recent developer conference, raising concerns about its future in the AI ecosystem [6][13] Group 3: Market Position and Threats - Apple is losing its dominance in the smartphone market, with competitors like Huawei and Xiaomi advancing rapidly in AI capabilities [8][22] - Google is solidifying its position in AI search and video, posing a direct threat to Meta's advertising market, particularly in short videos [7][10] Group 4: Talent Acquisition Efforts - Zuckerberg is actively recruiting top talent in AI, emphasizing the importance of building a strong team to drive Meta's AI initiatives [15][18] - Apple is also seeking to enhance its AI capabilities by potentially acquiring or collaborating with companies like Mistral and Thinking Machines Lab [19][21] Group 5: Future Outlook - The competition for AI talent and technology is intensifying, with both Apple and Meta needing to adapt quickly to avoid being left behind [12][23] - The ongoing mergers and acquisitions in Silicon Valley signal a new wave of consolidation in the AI sector, with both companies needing to act decisively [23]

Core Insights - Xiaomi's Q1 2025 financial report shows significant growth, with revenue reaching 111.29 billion yuan, a 47.4% year-on-year increase, and adjusted net profit of 10.68 billion yuan, up 64.5% [1][2] - The company regained its position as the top smartphone vendor in China with a 40% increase in domestic market shipments and an 18.8% market share [1][3] - Xiaomi's smart home appliances and IoT business also saw substantial growth, with revenue from smart appliances increasing by 113.8% and IoT revenue rising by 58.7% [1][5] Business Segments - **Smartphone Business**: The average selling price (ASP) of smartphones reached 1211 yuan, a 5.8% increase, with high-end smartphone shipments accounting for 25% of total shipments in mainland China, up 3.3 percentage points [3][5] - **Smart Electric Vehicles**: Revenue from the smart electric vehicle segment reached 18.1 billion yuan, representing 55% of last year's total revenue for this segment, with losses narrowing from 1.8 billion yuan to 500 million yuan [2][3] - **IoT and Consumer Products**: The IoT and consumer products segment generated 32.34 billion yuan in revenue, accounting for 29.1% of total revenue, highlighting its importance as a core business [5][9] Market Dynamics - Xiaomi benefited from subsidy policies that stimulated demand, particularly in the domestic market, leading to a "volume and price increase" [5][9] - The company is preparing for potential challenges post-subsidy, focusing on building its own smart appliance factory to reduce costs and maintain inventory levels [10][12] - Future growth may depend on Xiaomi's ability to transition from IoT hardware to AI services, with ongoing research in AI technology [12][13]

Investment Rating - The report assigns a "Buy" rating for Xiaomi Corp. (1810.HK) with a 12-month target price of HK$62.00, representing an upside potential of 21.6% from the current price of HK$51.00 [1]. Core Insights - Strong sales in the AIoT segment are expected to drive higher profits in 1Q25, with significant growth in various product categories [1][2]. - The upcoming strategic product launch event is anticipated to unveil key innovations, including the XRING O1 chip and new premium smartphone models, which could enhance Xiaomi's competitive position [2][3]. - The report highlights Xiaomi's structural market share gains in China, particularly against competitors like Apple and Honor, despite a less optimistic overseas shipment outlook [3]. Financial Performance - Revenue forecasts for 2025-2027 remain largely unchanged, while adjusted net profit forecasts have been raised by 3-6% due to stronger IoT sales and gross profit outlook [17]. - For 1Q25, revenue is projected to grow by 45% year-on-year to RMB 109.5 billion, with adjusted net profit expected to increase by 70% year-on-year to RMB 9.4 billion [17]. Market Position and Growth - In the AIoT segment, Xiaomi's domestic sales of air conditioners, washing machines, and refrigerators saw year-on-year growth of 103%, 184%, and 145%, respectively, in 1Q25 [16]. - Xiaomi's tablet shipments grew by 57% year-on-year in 1Q25, achieving a No.3 market share globally and in China [16]. - The report anticipates that sales from large appliances and tablets will contribute approximately 40% of AIoT sales by 2027, up from around 30% in 2024 [16][37]. Valuation and Price Target - The 12-month SOTP-based target price for Xiaomi has been adjusted to HK$62, based on a 23x 2026E EV/NOPAT for Xiaomi core and a DCF-based valuation for Xiaomi EV at US$74 billion [18]. - The report indicates multiple share price catalysts in the coming months, including the strategic product launch event and 1Q25 results [19].

5月港股以"科技领涨+全市场普涨"的强劲姿态迎来开门红，$港股科技50ETF(SZ159750)$今天盘中一度涨至2.8%，临近午间收盘成交额突破1亿，换手率超 18%，两融品种交投十分活跃。 从K线来看，恒生科技5月2日大阳线正好触及关税大跌缺口上沿，包括今日的震荡，表明在此处依然承压。 港股科技指数更强一些，2号大涨已经突破了大跌缺口上沿，今天盘中的下探也回踩了这一位置。 港股行情本质是"钱潮"驱动的估值修复行情。 可以看到两只指数午盘已经双双翻红，这个位置有阻力，但势头很猛，率先突破的港股科技指数后续弹性可能更大。 4月南向大幅涌入港股将近2000亿，今年已超6000亿元，差不多是去年同期的三倍，市场预计全年仍将突破万亿元。这说明内地资金很清楚：港股科技现在 还趴在估值洼地里。恒生科技市净率在近十年18%左右的历史分位，相当于过去十年里只有10%的时间比现在便宜。 不止南向，外资买港股买得也很猛，把港元都买贵了。今天上午，香港金管局在市场卖出605.43亿港元，因为港元汇价触及强方兑换保证。这是自2020年10 月28日以来，港元首次触发联系汇率机制下的强方兑换保证。 金管局表示，近期港元偏强主要由于股 ...

【主编观市】 四月最后一天恒指往上，给五月行情带来指引。 一般放长假海外市场上涨的概率偏大，美股有几个催化： 1，海外AI巨头数据超预期，假期内大涨。如微软、mate等。 2，美国4月非农数据超预期。新增17.7万，大幅超出预估的13.8万增量。 3，特朗普做预期管理，不断释放各种签署协议的所谓利好。 优必选(09880) 2024 年公司实现营收 13.05 亿元，同比+23.7%；毛利润 3.74 亿元，同比+12.4%。主要得益于教育智能 机器人和定制智能机器人产品收入增长。 但这依然只能作为短期来看，看下伯克希尔的现金储备从2024年底的约3340亿美元上升至创纪录的3477 亿美元，显示巴菲特仍在等待合适的投资机会。 当地时间5月7日，美联储将公布最新利率决议。目前市场一致预期，美联储将按兵不动。 对国内而言，汇率走强才是关键，5月5日，离岸人民币盘中一度升穿7.20关口，为去年11月以来首次， 创近半年以来新高。亚洲其它货币也延续上周五的涨势，集体向上脉冲，这意味着美国经济衰退概率上 升、未来利率可能走低。市场普遍预期美元可能续贬值。 财政部今年赤字率按4%安排，比去年提高1个百分点，赤字规模达到 ...

假期海外市场复盘 地缘政治方面，乌达成矿产协议，乌克兰总统泽连斯基称之为"真正平等的协议"：哈马斯愿与以色 列达成为期5年的停火协议。 能源方面，OPEC+确认6月将增产41.1万桶1日，高盛预计2026年布油将推至 40 美元区间。 国内宏观 五一出行消费:跨区域人员流动有望创新高;出行旅游火爆;票房同比明显下滑。 政策：习近平主持召开部分省区市"十五五"时期经济社会发展座谈会;商务部回应美方愿与中方就关 税谈判，评估美方诚意与行动;财政部部长蓝佛安在《求是》上发表署名文章。 五一假期期间(亚太股市取5月1日至5月2日，其他股市/商品均取4月30日至5月2日，下同)，港股虽 只有一个交易日(5月2日)，但整体表现强势，恒生科技涨超3%领涨全球主要指数。美股三大指数均 收涨，纳指涨近 3%，道指、标普500涨幅在 2%附近，纳斯达克中国金龙指数同步走强，全球其余 股市也多数上涨;商品整体下跌，油跌幅在3%附近，黄金同步走弱，有色金属多数下跌，粮食整体上 涨;美元指数升破100点，美债收益率涨16BP;离岸人民币5月2日大幅涨近 700基点。 五一假期期间，港股仅有5月2日开市交易。恒生指数、恒生科技5月2日分 ...

[Table_Main] 行业研究|电信服务 证券研究报告 电信服务行业周报 2025 年 5 月 4 日 [Table_Summary] 报告要点： 市场整体行情及通信细分板块行情回顾 [Table_Invest]推荐|维持 [Table_Title] 小米发布首个推理模型 MiMo，Meta 上修资本开支 指引 ——通信行业周报 周行情：本周（2025.4.28-2025.5.2）上证综指回调 0.49%，深证成 指回调0.17%，创业板指上涨 0.04%。本周申万通信上涨0.59%。考 虑通信行业的高景气度延续，AI、5.5G 及卫星通信持续推动行业发 展，我们给予通信行业"推荐"评级。 细分行业：本周（2025.4.28-2025.5.2）通信板块三级子行业中，通 信应用增值服务上涨幅度最高，涨幅为 6.23%，其他通信设备回调幅 度最高，跌幅为 0.93%，本周各细分板块主要呈上涨趋势。 个股方面：本周（2025.4.28-2025.5.2）涨幅板块分析方面，博创科 技（26.63%）、平治信息（23.03%）、万隆光电（15.28%）涨幅 分列前三。 建议关注方向：算力产业链、卫星互联网 1) 算 ...

通信周观点： 1）微软与 Meta 最新财报验证 AI 商业化加速、算力投资延续高景气。微软 Azure 和其他云服务收入本季度同比增长 35%，其中 AI 贡献了 16%，2025 年资本开支维持 800 亿美元。Meta 第一季度经营利润 175.6 亿美元，同比增长 27%。 用户在其应用上的使用时长和互动频率提高，公司还上调全年资本开支至 640-720 亿美元，主投 AI 数据中心和硬件。 在北美云厂商强劲的资本开支下，我们认为上游光模块、服务器、连接器等行业需求有望保持高增长，市场此前有关 北美云厂商资本开支增速放缓的担忧得到释放。2）受益于 AI 需求驱动和国内外互联网厂商资本开支增长，服务器、 连接器等龙头公司业绩亮眼。服务器板块工业富联营收、净利均创历史新高。数据中心高密度连接需求爆发，MPO 及 AEC 成为核心增量赛道。以太网交换机市场结构性分化，AI 算力需求推动数通交换机向 800G/1.6T 高速率升级，增势 迅猛。我们看好交换机板块业绩触底回升。3）国内大模型迭代，落地应用有望加速。小米首个推理大模型 MiMo 开源， 满足端侧本地运行。阿里通义千问发布新版 Qwen3 系列模型 ...

Global Market Overview - During the May Day holiday, global stock markets mostly rose, with the Nasdaq index up by 2.96%, and the Hang Seng Technology index rising by 3.08% on May 2, driven by easing US-China tariff tensions [5][19][20] - Commodity markets experienced an overall decline, with oil prices dropping approximately 3% and gold prices also weakening, while grain prices saw an increase [16][17] - The US dollar index surpassed 100 points, and the offshore RMB appreciated significantly, gaining nearly 700 basis points [17][18] Overseas Macro - The US economy showed signs of weakness under Trump's tariff policies, with a reported GDP contraction of 0.3% in Q1, while consumer spending increased by 0.7% in March [21][22] - The April non-farm payroll data exceeded expectations, with an increase of 177,000 jobs, leading to a reduced probability of interest rate cuts in June [22][23] - A mineral agreement was reached between the US and Ukraine, establishing a joint investment fund for resource exploration [25] Domestic Macro - The May Day holiday saw a significant increase in cross-regional travel, with an estimated 1.42 billion trips, marking a 4.5% year-on-year growth [29] - The tourism market was robust, with a notable increase in long-distance travel demand, while box office revenues saw a significant decline of 51.6% year-on-year during the holiday [30][32] - Policy developments included a meeting led by Xi Jinping focusing on economic and social development strategies for the upcoming "15th Five-Year Plan" [33] Industry Dynamics - Xiaomi launched its first inference open-source model, MiMo, which surpassed larger models from OpenAI and Alibaba in performance [43] - Apple is collaborating with Anthropic to develop an AI platform for software coding, aiming to enhance internal workflows [50] - Momenta and Uber announced a strategic partnership to commercialize Robotaxi services in international markets starting in 2026 [51]