这些大模型背后，是高速增长的产业规模。2025年，广东人工智能核心产业规模预计突破3000亿元，总 量占据全国四分之一，增速超40%。这一数据表明，广东不仅是传统制造业的重镇，更已成为全国人工 智能产业的核心增长极，为大模型的研发与应用提供了丰富的场景需求和雄厚的产业支撑。 广货爆款开箱 开箱！解码广货。每期一爆款，开箱见真章。 从一部智能手机到一辆新能源车，"广货"早已是国民生活中的"硬核存在"。它到底有多能打？爆款背后 藏着怎样的创新密码？南都大数据研究院"广货爆款开箱"专栏上线！用数据带您透视广货从核心零件到 全球布局的真实战力，解码它如何风靡世界。 第六期，开箱一款看不见的"新广货"——人工智能大模型。 在"广货行天下"的版图上，一项看不见却至关重要的"新广货"正崭露头角，它并非实体商品，而是驱动 千行百业智能升级的"南粤智脑"——人工智能大模型。 产业的繁荣，首先体现在合规发展的大模型数量上。截至2026年1月，广东省通过国家备案的大模型数 量已达124款，占全国总量的15%以上。这些备案模型覆盖文化、教育、工业制造、市场营销等多个领 域，标志着广东AI产业正从"单点突破"迈向"生态共振"。 ARR ...

在这次发布之前，Anthropic 内部和部分早期用户已经开始让 Opus 4.6 参与一项持续时间很长的工 程任务：从零开始，用 Rust 编写一个完整的 C 编译器，并要求它能够编译 Linux 内核。 这项实验持续了约两周时间，期间累计运行了近两千次 Claude Code 会话，最终产出了一个规模约 10 万行代码的编译器。该编译器不仅能够在多种架构上构建 Linux 6.9，还可以编译 FFmpeg、 Redis、PostgreSQL、QEMU，并通过了 GCC 自身 99% 的 torture test，甚至能够成功编译并运行 Doom。整个实验的 API 成本约为 2 万美元。 作者 | Tina Anthropic 正在升级它"最聪明的模型"。 随着新一代旗舰模型 Claude Opus 4.6 的发布，Anthropic 释放出的信号十分明确：这并不是一次常 规的性能小修小补，而是一轮围绕长任务、复杂工作，以及智能体（agent）如何真正干活展开的系 统性升级。 为了让外界更直观地理解这一成果的尺度，有网友在社交平台上给出了一个对照： GCC 的开发从 1987 年开始，历经 37 年，投入 ...

Group 1: Core Insights - The article presents a weekly roundup of the top 50 keywords in the AI sector, highlighting significant developments and trends in the industry [2] - Key players mentioned include Tencent, Anthropic, Google, and NVIDIA, indicating a competitive landscape in AI advancements [3][4] Group 2: Categories and Summaries - **Computing Power**: Tencent's HPC-Ops open-source initiative is noted as a significant development in enhancing computational capabilities [3] - **Models**: Various AI models are highlighted, including Sonnet 5 from Anthropic and Qwen3-Coder-Next from Alibaba, showcasing ongoing innovation in model development [3][4] - **Applications**: Google is actively developing AI applications such as Chrome AI and Genie 3, while Tencent is involved in the public testing of its Yuanbao platform [3][4] - **Technology**: NVIDIA's Earth-2 open model is mentioned, indicating advancements in AI technology and its applications [4] - **Events**: Significant events include SpaceX's acquisition of xAI and record revenue reported by Google, reflecting the financial impact of AI developments [4] - **Opinions**: Various viewpoints are shared, including discussions on the authenticity of AI outputs and the competitive landscape in AI innovation, particularly in China [4]

Core Viewpoint - The company Kunlun Wanwei is focusing on overseas market expansion and application ecosystem development rather than competing directly with major players in the AI field, aiming for long-term market positioning and ecological advantages despite short-term losses [3][10][22]. Financial Performance - For 2025, Kunlun Wanwei forecasts a net profit loss of between 1.35 billion to 1.95 billion yuan, continuing the trend of losses from the previous year [3][15]. - The company expects a revenue increase of over 40% year-on-year for 2025, with a growing proportion of revenue coming from overseas markets [3][15]. - In Q3 2025, the company achieved a net profit of 190 million yuan, a year-on-year increase of 180.13%, but ended Q4 with losses [4][16]. Investment Strategy - The company has adopted a "computing power - large model - application" strategy, significantly increasing its market promotion and R&D investments, which are essential for its survival in a competitive landscape [5][16]. - R&D expenditures from 2021 to 2024 were 628 million, 736 million, 1 billion, and 1.598 billion yuan respectively, with 2025's first three quarters seeing R&D spending of 1.211 billion yuan [17]. - Sales expenses from 2021 to 2024 were 1.473 billion, 1.566 billion, 1.62 billion, and 2.304 billion yuan, with 2025's first three quarters reaching 2.891 billion yuan [17]. Competitive Positioning - Kunlun Wanwei has chosen a differentiated competition strategy, focusing on vertical markets and AI application ecosystems rather than competing in the general large model arena [6][18]. - The company has developed various AI models and applications, including the Skywork series and the Skywork Super Agents, which serve multiple scenarios such as office productivity and multimedia content creation [19][20]. Market Expansion - The company has seen unexpected growth in its short drama products, with platforms like DramaWave and FreeReels achieving significant user engagement and revenue growth [21]. - Predictions indicate that the short drama business could generate 1.68 billion yuan in revenue for 2025, marking a 900% increase year-on-year [21]. Future Outlook - The company views the current investment phase in AI as a unique opportunity, with expectations of transitioning from losses to profitability by 2027 [11][23]. - The strategy emphasizes not just leading in general models but establishing advantages in specific fields and exploring global market opportunities [10][22].

转自： 新智元报道 【新智元导读】 整个硅谷又癫狂了！Anthropic深夜扔出王炸，Claude 4.6用近乎恐怖的编程能力和智能体军团，给OpenAI和谷歌上了一堂名为 「降维打击」的课。 Anthropic深夜惊雷，终于祭出了编程之王！ 刚刚， Claude Opus 4.6横空出世，一夜成为全球最强编程AI ，「无模」能敌。 它在前一代Opus 4.5的基础上，大幅提升了编码技能。 Claude Opus 4.6的规划更缜密，能更持久地执行AI Agent任务，在超大规模代码库中运行也更加可靠。 最重要的是， 它还具备更强自我纠错能力 ，比如精准的代码审查、调试。 它也是Anthropic首款在beta阶段， 上下文支持100万token的Opus级模型 。 在多项基准测试中， Claude Opus 4.6编程实力几乎全方位领先 ，Gemini 3 Pro、GPT-5.2望尘莫及。 在ARC-AGI-2上，Opus 4.6拿下68.8%高分，超过GPT-5.2-xhigh （左右滑动查看） 不仅如此，Opus 4.6一经上线，就开始革命办公了！ 今天，这款新模型同时在Excel、PPT中的Claud ...

Core Insights - The article discusses the groundbreaking AI research paper published in *Nature* by the Beijing Academy of Artificial Intelligence, introducing a multimodal model named "Emu3" that aims to unify various AI capabilities such as vision, language, and action through a single task of "next token prediction" [1][4][21]. Group 1: Emu3's Technical Innovations - Emu3 utilizes a unique "Vision Tokenizer" that compresses a 512x512 image into just 4,096 discrete symbols, achieving a compression ratio of 64:1, and further compresses video data in a time-efficient manner [8][9]. - The model architecture of Emu3 is a standard language model enhanced with 32,768 visual symbols, diverging from the complex encoder-decoder architectures used by other models [10][11]. - Emu3 demonstrates superior performance in various tasks, scoring 70.0 in human preference evaluations for image generation, 62.1 in visual language understanding, and 81.0 in video generation, surpassing established models [11]. Group 2: Scaling Laws and Multimodal Learning - Emu3's research confirms that multimodal learning adheres to predictable scaling laws, indicating that performance improves uniformly across different modalities when training data is increased [12][13]. - The findings suggest that future multimodal intelligence may not require separate training strategies for each capability, simplifying the development process [13]. Group 3: Comparison with Global Peers - Emu3 is positioned against models like Meta's Chameleon and OpenAI's Sora, showcasing its ability to bridge the performance gap between unified architectures and specialized models [17][18]. - Unlike OpenAI's approach, which requires additional models for understanding, Emu3 integrates generation and comprehension within a single framework [18]. Group 4: Commercialization Potential - Emu3's architecture allows for efficient deployment, leveraging existing infrastructure for large language models, which can reduce operational complexity and costs [19]. - The model's unified capabilities enable diverse applications, from generating instructional content to real-time video analysis, enhancing user interaction [20]. Group 5: Philosophical Implications - Emu3 challenges the notion of fragmented intelligence by proposing that intelligence can be unified through a single predictive framework, potentially reshaping the understanding of AI's capabilities [21][22]. - The success of Emu3 suggests a paradigm shift in AI development, emphasizing simplicity and unified approaches over complexity [22].

神经元如何"看见"和"移动"？ 要理解这项实验的精妙之处，首先需要了解科学家是如何让这些"无眼无手"的神经元与游戏世界互动 的。 实验的核心是一个高密度微电极阵列（HD-MEA）——一块布满数千个微型电极的硅芯片。研究人员将 从小鼠胚胎皮层或人类诱导多能干细胞（hiPSCs）分化而来的神经元接种在这块芯片上。几天后，这些 神经元彼此连接，形成一个自发放电、具有基本网络结构的微型"脑组织"。 【文/观察者网专栏作者 心智观察所】 近年来，随着大语言模型的爆发式发展和生成式人工智能的广泛应用，人们一度认为硅基计算——即由 晶体管、芯片和算法构成的传统人工智能——已经牢牢锁定了通往未来的道路。然而，就在AI系统变 得越来越庞大、能耗越来越高、对数据依赖越来越强的同时，一个曾被边缘化的疑问重新浮出水面：智 能是否必须建立在硅片之上？有没有可能，真正的下一代智能，其根基不在金属与电流，而在活生生的 细胞与突触之中？ 这一问题并非空想。在全球多个前沿实验室里，科学家正尝试将活体神经元与电子设备深度融合，构建 一种被称为"合成生物智能"（Synthetic Biological Intelligence）的新范式。它不依赖 ...

Group 1 - OpenAI launched the coding model GPT-5.3-Codex on February 5, which integrates the reasoning and expertise capabilities of GPT-5.2, achieving a significant performance leap and a 25% increase in operational speed, enabling it to handle long-term tasks involving research and complex execution [2] - Anthropic announced the release of the upgraded intelligent model Claude Opus 4.6 on February 5, which can plan more cautiously and execute tasks reliably in large codebases, generating detailed financial analysis reports that typically take days for humans to complete, leading to a significant drop in the stock price of financial services companies like FactSet by up to 10% [2] - Tencent launched the internal testing of its desktop agent tool "WorkBuddy" on February 6, marking its entry into the desktop agent market, with over 2,000 employees participating in the internal test, applying the tool in various scenarios such as data processing, content creation, and automated office tasks [2] - Zhiyuan Technology introduced the first multimodal memory platform MemoryLake, shifting AI focus from "data-centric" to "memory-centric," integrating large models and memory engines to address challenges in AI applications, serving over 15 industries and 15,000 enterprises, with significant improvements in decision-making speed [2]

(Bloomberg) — Anthropic PBC (ANTH.PVT) is ironing out the final details on a funding round likely to raise more than $20 billion and slated to close as soon as next week, according to people familiar with the matter. The OpenAI (OPAI.PVT) rival was initially seeking $10 billion but is now on track to raise more than double that amount at a $350 billion valuation, thanks in part to excess investor interest, said the people, who asked not to be identified because the details are private. Most Read from Bl ...

格隆汇2月7日｜据美国财经媒体Semafor，有知情人士透露，OpenAI正与总部位于阿布扎比的G42公司 合作，打造一个针对阿联酋定制的ChatGPT新版本，以适应当地的语言、政治观点和言论限制。OpenAI 员工透露，该定制版专为阿联酋政府设计。建成后将成为热门聊天机器人本地化的首批案例之一，预示 着美国科技公司在推进全球人工智能扩张时，可能需要推出区域化版本的技术方案。据悉，具体细节仍 在商讨中，但最终成果预计会是经过精细调整的 ChatGPT版本，它能流利地使用当地阿拉伯语，并可 能设有内容限制。参与该项目的人士透露，阿联酋希望该聊天机器人能传递与君主制一致的政治立场。 ...