DeepSeek
Search documents
尴尬:韩国押注“主权 AI”,却发现本土大模型用了中国开源代码
Xin Lang Cai Jing· 2026-01-14 14:03
Core Insights - South Korea aims to develop its own AI large models but has faced challenges as domestic models have been found to use code from Chinese companies, highlighting the difficulty of reducing reliance on major tech giants from China and the US [1][11]. Group 1: Competition and Development - The South Korean government initiated a competition in June last year to create a new, independent AI model using domestic technology, which is crucial for ensuring technological autonomy amid a global landscape dominated by the US and China [3][13]. - In this three-year competition, three out of five finalist companies were found to have used parts of foreign AI models' open-source code, including Chinese models, raising concerns about the feasibility of developing entirely independent models [3][13]. - Experts argue that avoiding existing AI models and attempting to build everything from scratch is impractical, while critics warn that using foreign tools poses potential security risks and undermines the hope of nurturing truly domestic AI models [3][13]. Group 2: Controversies and Reactions - Upstage, one of the finalist companies, faced controversy when it was claimed that parts of its AI model were similar to the open-source model from Chinese company Zhizhu AI, with allegations that some code retained Zhizhu AI's copyright markings [5][16]. - Upstage held a live verification session to demonstrate that its model was developed from scratch, although it acknowledged that its inference code used elements derived from Zhizhu AI's open-source components, which are widely adopted globally [8][18]. - The controversy has led to stricter scrutiny of other finalist models, with Naver's AI model being criticized for similarities with products from Alibaba and OpenAI, and SK Telecom facing criticism for its inference code being similar to that of Chinese company DeepSeek [8][18]. Group 3: Government and Regulatory Response - The competition rules did not clearly state whether the use of foreign companies' open-source code was allowed, and the South Korean Ministry of Science has not issued new guidelines since the controversy arose [10][19]. - The Minister of Science welcomed the intense discussions surrounding the technology debate, viewing it as a sign of a bright future for South Korea's AI industry [10][19]. - The Ministry plans to eliminate one of the five finalist companies as originally scheduled, despite the ongoing scrutiny and debate [10][19].
梁文锋的幻方量化去年收益57%,跻身百亿级量化基金业绩榜第二!
21世纪经济报道· 2026-01-14 08:38
Core Viewpoint - The article highlights the impressive performance of Fantom Quantitative, which achieved an average return of 56.55% in 2025, ranking second among quantitative private equity firms in China, and emphasizes the financial support it provides to DeepSeek for AI model development [1][2]. Group 1: Company Performance - Fantom Quantitative's average return over the past three years is 85.15%, and over the past five years, it is 114.35% [1]. - The company currently manages over 700 billion yuan, maintaining its position in the top tier of China's private quantitative investment sector [1]. - Estimated revenue from management fees and performance commissions for the previous year could exceed 700 million USD, based on a 1% management fee and 20% performance commission [2]. Group 2: DeepSeek Development - DeepSeek, founded in July 2023, is focused on general artificial intelligence and is primarily funded by the research budget of Fantom Quantitative [2]. - The V4 model, an iteration of the V3 model set to be released around the Spring Festival in February, is reported to surpass current leading models in programming capabilities [3]. - DeepSeek's V3 model had a total training cost budget of 5.57 million USD [2]. Group 3: Industry Context - Competitors in the AI model space, such as Zhizhu and MiniMax, have reported significant R&D expenditures, with Zhizhu's cumulative investment reaching approximately 4.4 billion yuan and MiniMax's around 316 million yuan [3]. - The Italian antitrust authority concluded an investigation into DeepSeek regarding user warnings about potential misinformation, indicating regulatory scrutiny in the AI sector [4].
中美AI巨头都在描述哪种AGI叙事?
腾讯研究院· 2026-01-14 08:33
Core Insights - The article discusses the evolution of artificial intelligence (AI) in 2025, highlighting a shift from merely increasing model parameters to enhancing model intelligence through foundational research in four key areas: Fluid Reasoning, Long-term Memory, Spatial Intelligence, and Meta-learning [6][10]. Group 1: Key Areas of Technological Advancement - In 2025, technological progress focused on Fluid Reasoning, Long-term Memory, Spatial Intelligence, and Meta-learning due to diminishing returns from merely scaling model parameters [6]. - The current technological bottleneck is that models need to be knowledgeable, capable of reasoning, and able to retain information, addressing the previous imbalance in AI capabilities [6][10]. - The advancements in reasoning capabilities were driven by Test-Time Compute, allowing AI to engage in deeper reasoning processes [11][12]. Group 2: Memory and Learning Enhancements - The introduction of Titans architecture and Nested Learning significantly improved memory capabilities, enabling models to update parameters in real-time during inference [28][30]. - The Titans architecture allows for dynamic memory updates based on the surprise metric, enhancing the model's ability to retain important information [29][30]. - Nested Learning introduced a hierarchical structure that enables continuous learning and memory retention, addressing the issue of catastrophic forgetting [33][34]. Group 3: Reinforcement Learning Innovations - The rise of Reinforcement Learning with Verified Rewards (RLVR) and sparse reward metrics (ORM) has led to significant improvements in AI capabilities, particularly in structured domains like mathematics and coding [16][17]. - The GPRO algorithm emerged as a cost-effective alternative to traditional reinforcement learning methods, reducing memory usage while maintaining performance [19][20]. - The exploration of RL's limitations revealed that while it can enhance existing capabilities, it cannot infinitely increase model intelligence without further foundational innovations [23]. Group 4: Spatial Intelligence and World Models - The development of spatial intelligence was marked by advancements in video generation models, such as Genie 3, which demonstrated improved understanding of physical laws through self-supervised learning [46][49]. - The World Labs initiative aims to create large-scale world models that generate interactive 3D environments, enhancing the stability and controllability of generated content [53][55]. - The introduction of V-JEPA 2 emphasizes the importance of prediction in learning physical rules, showcasing a shift towards models that can understand and predict environmental interactions [57][59]. Group 5: Meta-learning and Continuous Learning - The concept of meta-learning gained traction, emphasizing the need for models to learn how to learn and adapt to new tasks with minimal examples [62][63]. - Recent research has explored the potential for implicit meta-learning through context-based frameworks, allowing models to reflect on past experiences to form new strategies [66][69]. - The integration of reinforcement learning with meta-learning principles has shown promise in enhancing models' ability to explore and learn from their environments effectively [70][72].
大模型时代小公司,怎么走出OpenAI的路
新财富· 2026-01-14 08:05
Core Insights - The article discusses the recent IPOs of AI companies, highlighting the significant oversubscription rates and initial stock price surges, indicating strong market interest in AI ventures [3][5] - It emphasizes the challenges faced by AI startups in a landscape dominated by major tech firms like Tencent, ByteDance, and Alibaba, suggesting that these giants create a difficult environment for smaller companies to thrive [7][15] Group 1: Market Dynamics - The IPO of Zhihua Huazhang on January 8, 2026, had an issue price of HKD 116.2 per share, with a subscription rate of approximately 1,159 times, and a first-day price increase of 13.17%, leading to a market cap of nearly HKD 90 billion [3] - MiniMax, established only four years prior, went public on January 9, 2026, at HKD 165 per share, with an oversubscription of over 1,800 times and a first-day price increase of 109.1%, resulting in a market cap exceeding HKD 100 billion shortly thereafter [5] Group 2: Technological Paradigms - The article argues that the current AI landscape is shaped by the "Scaling Law," which suggests that increasing model size, data, and computational power leads to predictable improvements in performance [9][10] - It notes that the success of OpenAI is seen as a unique historical occurrence that may not be replicable, as the current environment is characterized by concentrated computational resources and homogenized model capabilities [12][13] Group 3: Competitive Landscape - The emergence of DeepSeek has altered industry perceptions by significantly reducing training and inference costs, challenging the narrative that only large investments can yield viable models [19][22] - Major companies are now treating models as foundational infrastructure rather than profit centers, which complicates the ability of startups to justify their value propositions to clients [22][23] Group 4: Strategies for Startups - Startups like MiniMax and Zhihua Huazhang are finding sustainable paths by avoiding direct competition with large firms, focusing instead on niche markets or specific applications [26][30] - MiniMax is targeting overseas markets with products centered on companionship and interaction, while Zhihua focuses on complex enterprise applications that larger firms may overlook [28][31] - The article suggests that successful startups must carve out unique positions within existing paradigms rather than attempting to replicate the success of giants like OpenAI [42]
韩国AI之困:国产大模型使用中国代码引发争议
Feng Huang Wang· 2026-01-14 06:54
Core Viewpoint - The article discusses the challenges faced by South Korean companies, particularly Naver, in developing indigenous AI models, highlighting the reliance on foreign code, particularly from Chinese sources, which undermines the goal of technological independence [1][2]. Group 1: Competition and Development - The South Korean government initiated a competition to create a new, independent AI model using local technology, aiming to reduce reliance on US and Chinese tech giants [1][2]. - The competition, lasting three years, has seen three out of five finalist companies using parts of foreign AI models' open-source code, including Chinese models [1][2]. Group 2: Controversies and Allegations - Upstage, one of the finalist companies, faced allegations from a competitor that its AI model contained modules similar to those of Chinese company Zhizhu AI, even retaining copyright markings [3]. - The controversy has led to increased scrutiny of other finalist models, with Naver's visual and audio encoders being compared to products from Alibaba and OpenAI, and SK Telecom's model being criticized for similarities to DeepSeek's code [4]. Group 3: Responses and Clarifications - Upstage held a live verification session to demonstrate that its model was developed from scratch, although it acknowledged using open-source elements widely adopted globally [4]. - Naver and SK Telecom defended their use of external encoders as a strategic decision, emphasizing that the core engine of their models was developed independently [4]. Group 4: Regulatory Environment - The competition rules did not clearly state whether the use of foreign open-source code was permissible, and the Korean Ministry of Science has not issued new guidelines since the controversy arose [5]. - The Minister of Science welcomed the debate surrounding AI technology, viewing it as a positive sign for the future of South Korea's AI industry [5].
阿里千问官宣:1月15日召开APP发布会,AI将“开启办事时代”
Hua Er Jie Jian Wen· 2026-01-14 06:45
Core Insights - Alibaba officially announced that its large model product, Qianwen, will hold a product launch event titled "You Ask, We Answer" on January 15, marking a key transition in AI applications from Q&A to actionable execution [1][4] Group 1: Product Development and Strategy - The upcoming launch event signals Qianwen's evolution from a simple Q&A tool to an intelligent agent capable of executing specific tasks, aligning with Alibaba Cloud's expansion goals in the AI cloud market [7] - The positioning of "From Question To Action" indicates that Qianwen aims to break the functional boundaries of traditional AI assistants, with the launch expected to introduce independent applications for end-users rather than just tools or API services for developers [8] - The recent updates to Qianwen Code, including the release of the v0.5.0 version with VSCode plugins and TypeScript SDK, provide foundational support for extending Qianwen's capabilities into execution-level functions [8] Group 2: Market Position and Competition - According to market research firm Omdia, the overall AI cloud market in China is projected to reach 22.3 billion yuan by the first half of 2025, with Alibaba Cloud holding a 35.8% market share, surpassing the combined share of the second to fourth players [7] - Alibaba Cloud's leading position in the AI cloud market supports the strategic upgrade of Qianwen products, with the company aiming to capture 80% of the incremental growth in the Chinese AI cloud market by 2026 [7] - The competitive landscape in the AI cloud market is intensifying, with other domestic models like DeepSeek preparing to launch next-generation AI models that may surpass current top models in programming capabilities [8][9]
AI应用板块再度拉升,指数涨超5%,软件ETF易方达(562930)昨日净流入资金近4亿元
Mei Ri Jing Ji Xin Wen· 2026-01-14 06:33
Core Viewpoint - The AI application sector is experiencing significant growth, with the Zhongzheng Software Service Index rising by 5.2% as of January 14, indicating strong investor interest in AI-related stocks [1] Group 1: Market Performance - The Zhongzheng Software Service Index, which includes 30 stocks involved in software development and services, has seen substantial gains, with key stocks like Shiji Information hitting the daily limit, and others like Hehe Information and Weining Health rising over 13% and 9% respectively [1] - The software ETF E Fund (562930) attracted nearly 400 million yuan in net inflows recently, reflecting heightened investor attention towards AI applications [1] Group 2: Industry Developments - Alibaba is set to release Qwen 3.5 soon, which is expected to show significant advancements in multimodal understanding, agent capabilities, and coding skills [1] - Major companies such as ByteDance, Tencent, and DeepSeek are anticipated to launch new models and products around the Chinese New Year, further driving innovation in the AI sector [1] - A report from Galaxy Securities highlights that the AI industry is continuously catalyzing, with vast commercial development potential in AI applications, particularly in generative search (GEO) [1] Group 3: Investment Opportunities - The Zhongzheng Software Service Index covers various AI application scenarios, including AI in office, finance, and education, with the top ten weighted stocks accounting for over 60% of the index [1] - The E Fund software ETF tracks this index, providing investors with opportunities to capitalize on the growth in the AI application field [1]
AI人工智能ETF(512930)涨超1.6%,国内大模型进展不断
Xin Lang Cai Jing· 2026-01-14 06:10
Core Viewpoint - The AI sector is experiencing significant growth, with the Zhongzheng AI Theme Index showing a strong increase, driven by advancements in AI technologies and upcoming product releases from major companies like Alibaba, ByteDance, and Tencent [1][2]. Group 1: Market Performance - As of January 14, 2026, the Zhongzheng AI Theme Index (930713) rose by 2.09%, with notable gains from stocks such as Yongyou Network (up 10.01%), Guangxun Technology (up 9.93%), and Runze Technology (up 9.32%) [1]. - The AI Artificial Intelligence ETF (512930) increased by 1.66%, with the latest price reported at 2.4 yuan [1]. Group 2: Upcoming Developments - Alibaba is set to release Qwen 3.5, which is expected to show significant improvements in multimodal understanding, agent capabilities, and coding abilities [1]. - Major companies like ByteDance, Tencent, and DeepSeek are anticipated to launch new models and products around the Spring Festival [1]. Group 3: Technological Advancements - Zheshang Securities believes that DeepSeek's Engram module offers a new direction for optimizing large language model architectures, enhancing model inference efficiency while reducing computational costs [1]. - The Engram model demonstrated improvements of 3.4 points and 4.0 points in knowledge tasks MMLU and CMMLU, respectively, and a 5.0 point increase in complex reasoning tasks compared to baseline models [1].
市场交投活跃!创业板人工智能ETF大成(159242)量价齐升涨超2%,机构判断AI应用商业化具备广阔发展空间
Xin Lang Cai Jing· 2026-01-14 05:23
Group 1 - The AI-focused ETF, Dachen (159242), has seen a 2.02% increase, with a trading volume of 1.33 billion yuan and a turnover rate of 39.71%, indicating active market participation [1] - The underlying index, the ChiNext AI Index (970070), rose by 2.29%, with significant gains from constituent stocks such as Yidian Tianxia (up 14.27%) and Yihua Lu (up 13.31%) [1] - The ChiNext AI Index emphasizes the engineering and industrialization of AI, focusing on foundational technologies like optical modules, computing chips, edge computing, and operating systems, distinguishing it from other indices that prioritize algorithm models [1] Group 2 - The Ministry of Industry and Information Technology has issued an action plan for the high-quality development of industrial internet platforms, aiming for over 450 influential platforms by 2028 and promoting AI technology across the industrial chain [2] - DeepSeek has released a new research result called "Engram," which introduces a scalable memory module to enhance knowledge storage and retrieval efficiency in large language models, significantly improving performance in various tasks [2] - The AI sector is expected to see a new wave of innovation driven by generative AI, with traditional consumer electronics like AI smartphones and PCs entering an upward cycle due to consumer upgrades [2] Group 3 - Recent events in the AI application field, such as the listings of Zhiyu and MiniMax on the Hong Kong Stock Exchange, are believed to transition the industry from technology validation to commercial value realization [3] - The AI application landscape is expanding, with generative search (GEO) emerging as a key area of exploration, while content interaction is becoming a significant breakthrough point, enhancing user engagement in gaming and other content sectors [3] - The Dachen ChiNext AI ETF and related funds are positioned to benefit from the ongoing developments in the AI sector [3]
2025年超1万家银行网点关闭,净减少超2000家|首席资讯日报
首席商业评论· 2026-01-14 04:34
Group 1 - In 2025, over 11,000 bank branches will be approved for closure, resulting in a net decrease of more than 2,000 branches, indicating an acceleration in digital transformation and optimization of physical banking channels [2] - SK Hynix announced an investment of 19 trillion KRW in its advanced packaging factory in Cheongju, South Korea, aiming to enhance production efficiency, with construction expected to start in April 2026 and completion by the end of 2027 [3] - The Ministry of Industry and Information Technology will focus on promoting the large-scale application of humanoid robots and health monitoring devices in various settings, emphasizing technological empowerment in the elderly care sector [4] Group 2 - Citigroup plans to lay off approximately 1,000 employees as part of a broader strategy to reduce 20,000 jobs by the end of 2026, reflecting adjustments to align workforce and skills with current business needs and technological advancements [5] - Ctrip clarified that a recent message about a mass layoff was a mistake, confirming that there is no plan for a full staff departure [6] - The U.S. Defense Secretary announced that Elon Musk's AI chatbot "Grok" will be integrated into the Pentagon's systems, alongside Google's generative AI, to enhance military operations [7] Group 3 - The launch of a 4,199 RMB bottle of Moutai sold out immediately on the iMoutai app, indicating strong demand for premium products [8] - Shanghai has introduced measures to optimize auto loan processes, including relaxing application conditions and determining reasonable loan issuance ratios, terms, and interest rates to stimulate consumption [9] - Apple responded to rumors regarding Google taking over iPhone control, clarifying that there has been no transfer of control over Siri or Apple Intelligence to Google [10] Group 4 - Lianchuang Electronics has begun supplying optical products to the robotics sector, although the industry is still in its early stages with relatively low sales [11] - DeepSeek published a new paper on conditional memory for large language models, co-authored with Peking University, contributing to advancements in AI research [12]