作者 | 王启隆 出品｜《新程序员》 2025 年的AI 领域，似乎没有哪个词比"Agent"更炙手可热。从 OpenAI 的 Operator 到"第一个通用智能体"Manus 的出圈，"智能体元年"的呼声不绝 于耳，仿佛我们距离那个能自主理解、规划、执行任务的通用人工智能（AGI）只有一步之遥。 喧嚣之下，一些根本性的问题挥之不去：究竟何为 Agent？我们真正踏上了通往通用人工智能（AGI）的那条路吗？当前主流的、依赖海量数据和算力 堆砌起来的大模型路径，是否足以孕育出真正拥有理解力、自主性甚至"灵魂"的智能？ 当许多人沉浸在狂欢之时，全球知名人工智能科学家、北京通用人工智能研究院院长、北京大学人工智能研究院院长兼智能学院院长朱松纯教授，却在 疾呼一种不同的声音——当前许多所谓的Agent，可能连真正的"智能体"都算不上。 近日，《新程序员》在北京的一场围绕其新书《通用人工智能标准、评级、测试与架构》的媒体见面会上，采访了朱松纯教授。他的观点，或许能为我 们拨开Agent 的迷雾，提供一个审视 AGI 未来更深邃的视角。 《新程序员》： 朱院长您好，今年Agent 是个热词，很多人称 2025 年是"A ...

Core Viewpoint - The Digital China Construction Summit highlighted the potential for humanoid robots to become commonplace in households by 2024, with projected financing for related projects exceeding 10 billion yuan [1]. Group 1: Industry Insights - The domestic humanoid robotics sector is expected to see significant investment, with over 10 billion yuan in financing anticipated in 2024, indicating a growing interest and market potential [1][6]. - The company, Yushu Technology, offers both consumer-grade and industrial-grade robots, with the latter primarily serving in hazardous environments such as power inspections and firefighting [2][6]. - The development of artificial intelligence (AI) is categorized into three stages: weak AI, strong AI, and AGI (Artificial General Intelligence), with the latter being a potential future goal that could be achieved through embodied intelligence [3][4]. Group 2: Technological Development - The realization of embodied intelligence is outlined in three steps: establishing a flexible cognitive system, achieving autonomous decision-making capabilities, and enabling precise physical interactions with the environment [1][7]. - Yushu Technology's humanoid robots are designed to recognize and interact with their surroundings, with ongoing research and development aimed at enhancing their decision-making and interaction abilities [7][9]. - The company emphasizes the importance of a robust industrial chain in China, suggesting that domestic firms are well-positioned to compete in the embodied intelligence space, particularly against software-focused companies in Silicon Valley [6]. Group 3: Future Outlook - The initial applications of embodied intelligence are expected to be in industrial sectors, followed by commercial applications in retail and healthcare, ultimately leading to the integration of humanoid robots into everyday households [9].

Core Insights - Alibaba has open-sourced its new generation model Qwen3, which integrates "fast thinking" and "slow thinking" capabilities, significantly reducing deployment costs compared to other large models like Deepseek [1] - The Qwen3 model employs a "Mixture of Experts (MoE)" architecture, allowing it to mimic human problem-solving by providing multi-step deep thinking for complex issues and quick responses for simpler queries, thus saving computational resources [3] - Alibaba is focusing on building its AI strategy around the Qwen series, with plans to invest over 380 billion RMB in cloud and AI hardware infrastructure over the next three years, surpassing the total investment of the past decade [4] Industry Context - Following the release of Deepseek's low-cost high-performance R1 model, domestic tech companies in China, including Baidu and iFlytek, are rapidly launching a series of cost-effective AI model services [3] - Alibaba's Qwen series has surpassed the US Llama in terms of open-source model downloads, with over 300 million downloads and more than 100,000 derivative models [4] - On the same day Alibaba announced Qwen3, OpenAI released several updates to ChatGPT, enhancing its shopping features and optimizing for various consumer categories, indicating a competitive landscape in AI model development [4]

Core Insights - Alibaba Cloud has launched the new Qwen3 model, which is the first "hybrid reasoning model" in China, integrating "fast thinking" and "slow thinking" into a single model, significantly reducing deployment costs and enhancing performance compared to previous models [1][4] Group 1: Model Performance and Architecture - Qwen3 features a total parameter count of 235 billion, with only 22 billion activated, and utilizes a mixture of experts (MoE) architecture [2][3] - The model has achieved a performance leverage of over 10 times with its 30B parameter MoE model, requiring only 3 billion to match the performance of the previous Qwen2.5-32B model [3] - Qwen3 has outperformed global top models like DeepSeek-R1 and OpenAI-o1 in various benchmarks, securing its position as the strongest open-source model globally [1][2] Group 2: Cost Efficiency and Deployment - The deployment cost for Qwen3 has significantly decreased, requiring only 4 H20 units for full deployment, with memory usage being one-third of that of DeepSeek-R1 [1][3] - All Qwen3 models are hybrid reasoning models, allowing users to set a "thinking budget" for performance and cost optimization in AI applications [3][4] Group 3: Future Developments and Goals - Future enhancements for Qwen3 will focus on expanding data scale, increasing model size, extending context length, and broadening modality range, while leveraging environmental feedback for long-term reasoning [4] - The Qwen3 team views this launch as a significant milestone towards achieving general artificial intelligence (AGI) and superintelligent AI (ASI) [4]

点击 上方 硬AI 关注我们 Qwen3系列包括两个专家混合 (MoE) 模型和另外六个模型。阿里巴巴表示，最新发型的旗舰模型Qwen3- 235B-A22B在代码、数学、通用能力等基准测试中，与DeepSeek-R1、o1、o3-mini、Grok-3和Gemini- 2.5-Pro等顶级模型相比，表现出极具竞争力。 此外，被称为"专家混合"（MoE，Mixture-of-Experts）模型的Qwen3-30B-A3B的激活参数数量是QwQ- 32B的10%，表现更胜一筹，甚至像Qwen3-4B这样的小模型也能匹敌Qwen2.5-72B-Instruct的性能。这 类系统模拟人类解决问题的思维方式，将任务划分为更小的数据集，类似于让一组各有所长的专家分别负 责不同部分，从而提升整体效率。 | | Qwen3-235B-A22B | Qwen3-32B | OpenAl-o1 | Deepseek-R1 | Grok 3 Beta | Gemini2.5-Pro | OpenAl-o3-mini | | --- | --- | --- | --- | --- | --- | --- | --- | | | ...

Group 1 - The core viewpoint of the article is that AGI (Artificial General Intelligence) may be achieved within the next decade, with significant advancements in AI research and development being made by companies like Google DeepMind [1][3][4] - Demis Hassabis, CEO of Google DeepMind, emphasizes the transformative potential of AGI in addressing major global challenges such as diseases and energy crises, while also warning of the risks associated with its misuse [1][4][5] - The article discusses the importance of defining AGI accurately, with Hassabis stating that the timeline for its realization depends on how AGI is defined, highlighting the need for a consistent definition that encompasses human cognitive abilities [3][12] Group 2 - Hassabis identifies two main risks associated with AI: the potential for malicious use of AI technologies and the challenges of maintaining human control over increasingly autonomous systems [5][7] - He calls for the establishment of a global governance framework for AI, emphasizing the need for international cooperation to create safety standards and regulatory measures [7][10] - The article highlights the necessity of a multi-dimensional risk assessment system to proactively address potential dangers posed by AI technologies [9][10] Group 3 - The discussion includes the philosophical implications of AGI, particularly regarding wealth distribution and the potential need for political reform in a future of extreme abundance enabled by AI [21][22][24] - Hassabis suggests that achieving extreme abundance through technological advancements could fundamentally change the nature of resource scarcity and inequality, necessitating new political philosophies [22][23][24] - The article concludes with a call for new philosophical frameworks to address the societal changes brought about by AGI and its impact on human life and governance [20][24][25]

整理｜冬梅、核子可乐 去年成功斩获诺贝尔奖之后，Demis Hassabis 决定与一位国际象棋世界冠军打场扑克以示庆 祝。Hassabis 一直痴迷于游戏，这股热情也成为他 AI 先驱之路上的契机与驱力。 近日，做客一档名为《60 分钟》的访谈栏目，讲述了他如何带领众多研究者追逐新的技术"圣 杯"——通用人工智能（AGI），一种兼具人类灵活性与超人般速度与知识储备的硅基智能形态。 除此之外，他也在访谈中透露了 DeepMind 未来的研究方向以及有可能亮相的产品和技术。 "天才少年"Hassabis AI 之旅始于国际象棋 Hassabis 于 2010 年与他人共同创立了了 AI 公司 DeepMind，2014 年该公司被谷歌以 5 亿多 美元收购。2017 年，他发明了 AI 算法 AlphaZero，它只需要国际象棋规则和四个小时的自对 弈，就能成为有史以来最强的国际象棋选手，击败人类国际象棋大师。 2024 年，Hassabis 与同为诺贝尔化学奖得主的 DeepMind 总监约翰·江珀 (John Jumper) 共同 获得了诺贝尔化学奖，获奖原因是他创建了一个 AI 模型 AlphaFold2 ...

Core Insights - After nearly two years of adjustment, iFlytek's performance has returned to double-digit growth, with a reported revenue of 23.343 billion yuan for 2024, marking an 18.79% year-on-year increase, and a net profit of 560 million yuan, with a nearly 60% growth in net profit excluding non-recurring gains [1][3]. Revenue Growth - iFlytek achieved total revenue of 23.343 billion yuan, up from 19.650 billion yuan in 2023, reflecting an 18.79% increase, with gross profit rising by 1.568 billion yuan [3]. - The growth in revenue is primarily attributed to strong performance in smart education, open platforms, and consumer business segments, with the open platform and consumer business generating 7.886 billion yuan, a 27.58% increase, maintaining its position as the largest business segment [4]. Business Segment Performance - The smart education segment reported revenue of 7.23 billion yuan, nearly a 30% increase, with its share of total revenue rising from 28.31% to 30.97% [4]. - The automotive, medical, and enterprise AI sectors also showed rapid growth, with revenues of 989 million yuan, 692 million yuan, and 643 million yuan, respectively, reflecting year-on-year increases of 42.16%, 28.18%, and 122.56% [4]. Cash Flow Improvement - iFlytek's cash flow situation significantly improved in 2024, with net cash flow from operating activities reaching 2.495 billion yuan, a 613.40% increase from 350 million yuan in the previous year [4][5]. Strategic Focus on AI - iFlytek continues to focus on its strategic layout in the AI field, emphasizing the "1+N" strategy centered around the "Xunfei Spark" cognitive model to capture the benefits of the general artificial intelligence (AGI) era [2][6]. - The company is committed to self-research of foundational large models, aiming to maintain international leadership in core technologies while ensuring large-scale industrial application of technological achievements [6][8]. AI Model Development - iFlytek's "Xunfei Spark X1" model, with 70 billion parameters, has achieved advanced industry-level deep reasoning capabilities, comparable to larger parameter models [8]. - The model was primarily trained and optimized on the domestic Huawei Ascend 910B computing platform, demonstrating the feasibility and potential of training top-tier large models on domestic computing platforms [8][9]. Market Demand and Custom Solutions - There is a growing demand from state-owned enterprises and key industry clients for self-researched models due to issues encountered with open-source models, such as hallucinations and security vulnerabilities [9]. - iFlytek's self-researched models can achieve better performance and reliability, with an average improvement of 10% over general large models, and further enhancements of 10%-20% through scenario-specific customization [9]. Industry Responsibility - iFlytek's commitment to self-research and development of foundational large models reflects its responsibility and exploration in promoting technological self-innovation and ensuring the security of the industrial supply chain [10].

Core Insights - In 2024, iFlytek achieved a revenue of 23.343 billion yuan, marking an 18.79% year-on-year growth, returning to double-digit growth after two years, with a net profit of 560 million yuan [1] - The fourth quarter showed strong performance with a revenue of 8.494 billion yuan, accounting for 36.4% of the annual revenue, and a net profit of 904 million yuan [1] - In Q1 2025, iFlytek continued its growth trend with a revenue of 4.658 billion yuan, a 27.74% increase year-on-year, and net profits growing by 35.68% [1] Business Segments - The core business segment of iFlytek maintained steady growth, with the open platform and consumer business generating 7.886 billion yuan, a 27.58% increase [3] - Revenue from the open platform and smart hardware reached 5.172 billion yuan and 2.023 billion yuan, growing by 31.33% and 25.07% respectively [4] - The smart education business generated 7.229 billion yuan, reflecting a 29.94% year-on-year growth [5] Strategic Initiatives - iFlytek aims to leverage the advantages of general artificial intelligence (AGI) and focus on developing a leading general model on a fully controllable platform [2] - The company plans to enhance its multilingual model offerings, targeting international markets, particularly along the Belt and Road Initiative [4] - iFlytek's medical segment, iFlytek Medical, successfully listed on the Hong Kong Stock Exchange, with significant revenue growth across its three business lines [8] R&D and Innovation - In 2024, iFlytek's R&D investment reached 4.58 billion yuan, accounting for 19.62% of its revenue, focusing on model development and core technology [9] - The upgraded deep reasoning model, iFlytek Spark X1, shows significant improvements in various tasks while maintaining a smaller parameter size compared to industry peers [9] - iFlytek emphasizes the importance of developing industry-specific models based on its foundational models, ensuring higher quality and adaptability [10] Market Position and Challenges - Despite challenges from the US-China trade tensions, iFlytek's domestic revenue remains strong, and the company continues to pursue international opportunities [11] - The company believes that the current geopolitical climate presents more opportunities than risks, particularly in the context of global interdependence [12]

Core Viewpoint - The company achieved significant sales collection and improved cash flow in 2024, indicating a positive business outlook despite a decline in net profit [2][3]. Financial Performance - In 2024, the company reported total revenue of 23.343 billion yuan, a year-on-year increase of 18.79%, marking a return to double-digit growth after two years [3]. - The net profit attributable to shareholders was 560 million yuan, a decrease of 14.78% compared to the previous year [3]. - In Q1 2025, the company generated revenue of 4.658 billion yuan, up 27.74% year-on-year, but reported a net loss of 193 million yuan [3]. - The operating cash flow for the year was 2.495 billion yuan, a 613% increase from the previous year [2][3]. Cash Flow and Collection Mechanism - The company optimized its collection mechanism, leading to a historical high in cash flow, with a dedicated department for receivables and a supportive GBC (Government, Business, Consumer) structure [5]. - The operating cash flow turned positive in Q4 2024, reaching 3.316 billion yuan [4]. Business Structure and Revenue Growth - The proportion of sustainable revenue increased from approximately 65% in 2023 to 70% in 2024, following a reduction in product lines from 60 to 46 [5]. - The largest business segment, open platform and consumer business, generated revenue of 7.886 billion yuan, a 27.58% increase [5]. - The smart education segment achieved revenue of 7.229 billion yuan, growing by 29.94% [5]. - Sales of AI learning machines increased by over 100% in the first three quarters of 2024 [5]. Sector-Specific Revenue - Revenue from smart automotive, smart healthcare, and enterprise AI solutions reached 989 million yuan, 692 million yuan, and 643 million yuan, respectively, with growth rates of 42.16%, 28.18%, and 122.56% [6]. R&D Strategy - The company continues to invest in foundational model research despite market trends, citing the superior performance of its proprietary models compared to those trained on open-source models [7]. - The chairman emphasized the strategic importance of developing foundational models on domestic computing platforms, which are crucial for national security and trust among key clients [7].