本文摘自《云栖战略参考》，这本刊物由阿里云与钛媒体联合策划。目的是为了把各个行业 先行者的技术探索、业务实践呈现出来，与思考同样问题的"数智先行者"共同探讨、碰撞， 希望这些内容能让你有所启发。 AGI 并非终点，而是通往 ASI 的起点。在迈向超级人工智能（ASI）的征程中，阿里巴巴集团CEO吴泳 铭将其清晰划分为智能涌现、自主行动、自我迭代三个演进阶段，当前正处于承上启下的自主行动阶 段。 回溯 AI 发展进程，智能涌现阶段为迈向 ASI 奠定了基础。大模型的问世标志着一个里程碑式的突破， 它使 AI 摆脱传统任务局限，具备认知理解、内容生成与逻辑推理的通用智能基础，为 ASI 征程搭建了 认知底座，也为进入自主行动阶段做好了铺垫。 如今我们正处在 AI 自主行动的阶段：在智能涌现的基础上，AI 从感知与生成加速迈向决策与行动。智 能体技术 体系推动 AI 能力升级，实现了从被动响应指令到主动感知环境、规划任务、调用工具的本质 性转变，也重构了人机协作模式。与此同时，AI 正突破虚拟边界，以机器人、智能汽车、智能硬件等 形态为载体，深度融入生产制造、 公共服务、日常生活等真实物理场景。AI 的自主行动使得 ...

Core Insights - The rapid advancement of AI technologies, particularly with breakthroughs like DeepSeek, is leading to a significant acceleration in the iteration of AI applications, especially in the financial sector, which is poised to become a leading example of deep integration of large model technologies [1] - There are notable differences in the development of large models between domestic and international financial institutions, with international players often opting for commercial models while domestic institutions focus on open-source or self-built models [2] - Industrial banks, such as ICBC, are developing a "1+X" application paradigm for large models, which aims to enhance business capabilities through a dual integration of technology and business [3] Domestic and International Trends - International financial institutions tend to purchase external commercial large models and utilize public cloud deployment, while domestic institutions prefer self-built or collaboratively developed models with private cloud deployment [2] - The application scenarios in international finance are more diverse, focusing on core business areas like sentiment analysis and risk assessment, whereas domestic institutions are initially targeting efficiency improvements for frontline employees [2] New Application Models - ICBC has established a "1+X" model that includes a financial intelligence core and various capabilities such as knowledge retrieval and data analysis, enabling over 200 application scenarios [3] - The model allows for significant innovation in business processes, transitioning from single-scene empowerment to comprehensive business restructuring [3] Future Trends - Large models are expected to evolve into foundational infrastructure for financial services, with advancements in computing power supporting a "cloud-edge-end" AI deployment model [4] - The development of a model matrix layout is anticipated, featuring one versatile base model complemented by multiple specialized models for specific financial scenarios [5] - Regulatory bodies are expected to introduce clearer standards and guidelines for the ethical and compliant use of AI technologies in finance [6] Challenges in Implementation - Financial institutions face challenges in balancing the costs and value of AI model applications, as the demand for computational resources continues to rise [7] - The slow accumulation of high-quality data poses a significant barrier to achieving optimal AI performance, as the effectiveness of AI applications is increasingly dependent on data quality [7] - There is a notable shortage of interdisciplinary talent capable of bridging the gap between finance and AI technology, necessitating the establishment of robust talent development systems [7] Strategies for Smaller Institutions - Smaller financial institutions are encouraged to adopt a mixed model of "external collaboration + lightweight adaptation" to effectively leverage large models [9] - Focusing on core business areas and creating benchmark application scenarios can help smaller institutions maximize their resources [9] - Building a lightweight data ecosystem through distributed collaboration can address data limitations faced by smaller institutions [9] Future Development Pathways - Financial institutions should aim to enhance their intelligent infrastructure and develop a layered technical architecture to address the complexities of model development and computational infrastructure [10] - Accelerating the iteration of specialized models in vertical fields will enhance competitive advantages in core financial areas [10] - The integration of large model technologies is seen as a key driver for advancing financial services from process optimization to cognitive transformation [10]