当 OpenAI 敲定与微软合作的新条款时，该公司已告知部分股东，未来几年其向微软分成的收入比例将大幅下降。 根据双方最初达成的合作协议，微软有权在 2030 年前获得该初创公司收入的 20% 。然而， OpenAI 预计到本十年末，其与商业伙伴——主要是微软，分享 的收入比例将从今年的近 20% 降至约 8% 。 这一比例差异意味着到 2030 年， OpenAI 将额外保留超过 500 亿美元的收入，这是其迫切需要的，因为公司预计在此之前将因计算能力支出创下历史新 高。 两家公司还在协商当 OpenAI 实现所谓的人工通用智能（即与人类智力相当的 AI ）时会发生什么。现有合同规定， 一旦这家初创公司证明其技术能够超越 某些财务里程碑 ，微软将失去对 OpenAI 技术的独家使用权。但微软一直在寻求修改或删除合同中的 AGI 条款。 据另一位了解谈判内情的消息人士透露，双方还在协商 OpenAI 租用微软服务器的费用金额。 虽然协议的许多细节尚未敲定，但知情人士表示，部分条款已基本达成一致。 具体而言， OpenAI 的非营利机构和微软，将各自获得新公司约三分之一的 股份。目前该公司允许员工以 5000 ...

Core Viewpoint - The article discusses the reasoning capabilities of large language models (LLMs) and emphasizes the importance of intermediate reasoning steps in enhancing model confidence and accuracy in problem-solving [5][10][34]. Group 1: Importance of Reasoning in LLMs - Reasoning in LLMs refers to the intermediate thought processes that occur before arriving at a final answer, which can significantly improve the model's ability to solve complex problems [5][11]. - Introducing a chain of thought (CoT) allows LLMs to tackle inherently serial problems without needing to expand the model size, thus bridging the gap between Transformers and Turing machines [12][13]. - The presence of reasoning steps increases the accuracy and reliability of answers, reducing the likelihood of random guessing [14][17]. Group 2: Enhancing Model Confidence - Answers derived from reasoning processes lead to greater confidence in the model's outputs, as they are based on logical deductions rather than mere guesses [19][20]. - Denny Zhou highlights that pre-trained models possess reasoning capabilities even without fine-tuning, although these outputs may not be prioritized in greedy decoding [21][24]. Group 3: Methods to Improve Reasoning - The CoT-decoding method selects reasoning paths from top-k alternatives, enhancing performance on reasoning tasks and approaching the effectiveness of instruction-tuned models [26]. - Supervised fine-tuning (SFT) involves training models on human-written step-by-step problems, but it may lack generalization across new scenarios [27][28]. - Reinforcement learning fine-tuning has emerged as a powerful method for eliciting reasoning, focusing on generating longer responses and improving model performance through iterative training [31]. Group 4: Future Directions - Denny Zhou identifies key areas for future breakthroughs, including addressing tasks with non-unique verifiable answers and developing practical applications beyond benchmark testing [35][40].

Core Insights - The competition for AI talent among tech giants is intensifying, with OpenAI successfully recruiting at least four senior engineers from companies like Tesla, xAI, and Meta [1][2] Group 1: Recruitment Details - OpenAI has hired four senior engineers, including Tesla's VP of Software Engineering David Lau, xAI and X's infrastructure lead Uday Ruddarraju, xAI infrastructure engineer Mike Dalton, and Meta's AI researcher Angela Fan [1] - David Lau has been with Tesla since 2017, previously working in firmware, platform, and systems integration [1] Group 2: Statements from New Hires - David Lau expressed that accelerating the realization of safe and well-aligned artificial general intelligence is the most meaningful mission for the next chapter of his career [2] - Uday Ruddarraju highlighted the importance of infrastructure in turning research into reality, noting OpenAI's success in this area [2]

Group 1 - The National Development and Reform Commission and the National Bureau of Statistics have issued the "2025 Digital Economy Development Work Points," focusing on high-quality development of the digital economy, which is expected to boost capital expenditure related to computing power [1] - Companies are accelerating the procurement of high-performance computing servers, with a notable shortage in supply, as seen in the case of Xiechuang Data, which plans to invest up to 3 billion yuan in high-performance computing servers this year [1] - Xiechuang Data's major shareholder has announced a loan of up to 1.5 billion yuan to support the company's investment in computing power, indicating strong financial backing for its operations [1] Group 2 - The domestic AI boom, driven by DeepSeek, has led major companies to increase capital expenditure in the AI sector, significantly boosting order volumes for upstream enterprises [2] - Xiechuang Data has established a comprehensive intelligent computing power ecosystem and has built a solid technical foundation for computing power services across multiple cities [2] - The company has made significant progress in the computing power service sector, securing various orders from leading firms such as Tencent and China Unicom, highlighting its strong capabilities in this area [2] Group 3 - Xiechuang Data is focused on the transition from AI to AGI (Artificial General Intelligence) and anticipates the need for larger computing power clusters in the future [3] - The company aims to become a global leader in comprehensive cloud services while also expanding into the service robot sector, including advanced manufacturing capabilities [3] - Xiechuang Data has completed its annual equity distribution for 2024, increasing its total share capital to 343,357,204 shares [3]

Group 1: Business Performance and Financials - In Q1 2025, the company achieved revenue of 2,077.32 million yuan, a year-on-year increase of 18.11% [7] - The net profit attributable to shareholders in Q1 2025 was 169.21 million yuan, up 4.29% year-on-year [7] - Total assets as of March 31, 2025, reached 9,996.56 million yuan, reflecting a growth of 36.85% since the beginning of the year [7] Group 2: Strategic Initiatives and Market Position - The company plans to enhance its position in the AI and AGI sectors by increasing its investment in computing power services and cloud services [2] - The company has established a comprehensive platform for service robots, including advanced manufacturing capabilities such as robotic dogs and humanoid robots [2][4] - The company has obtained NVIDIA CLOUD PARTNER certification, indicating its capability in AI computing rental and cloud services [3] Group 3: Research and Development - R&D expenses have increased by 53.84%, with an expected revenue generation timeline of approximately 18 months from the start of investment [2] - The company is focusing on innovation and talent development, with increased investments in R&D and employee compensation [7] Group 4: Market Demand and Competition - The company is actively engaged in the computing power rental market, with a stable pricing strategy aimed at high-quality clients [8] - The expected return on investment for computing power rental services is reported to be between 28% and 35% [8] Group 5: Product Development and Future Outlook - The FCloud intelligent training platform was launched on January 24, 2025, and is expected to generate revenue in the future [9] - The company’s IoT smart terminal products achieved revenue of 2.259 billion yuan in 2024, marking a 60.82% year-on-year growth [9]