Under the wave of artificial intelligence, the financial industry is undergoing in-depth transformation. Currently, the application of large models in the financial field has fully shifted from proof-of-concept to large-scale implementation in business processes, customer services, and organizational structures, propelling the industry's digital transformation into a new stage. At the 2025 "China Merchants Bank Pujiang Digital Financial Ecology Conference", industry experts focused on the opportunities and challenges of large models and reached a significant consensus: scene cognition will be a key direction in the AI First era, and tracks such as embodied intelligence and emotional value will further determine whether large models can penetrate the core businesses of the financial industry.
Large Model Implementation: Coexistence of Achievements and Difficulties
Over the past year, the rapid development of domestic large-model technology has injected impetus into innovation in the financial industry. In August this year, although GPT-5 released by OpenAI did not fully meet expectations, the improvement in the capabilities of its foundation model and the mitigation of the hallucination phenomenon have laid the foundation for the in-depth application of large models in the financial sector. Tan Dai, President of Volcano Engine, stated that AI in the Agent era can operate automatically. It can not only revolutionize the efficiency and scale of financial services, but its enhanced multimodal data processing capability can also automatically generate business summaries, analyze data, activate unstructured information, and provide institutions with accurate decision-making support.
Professor Zhu Jun from Tsinghua University summarized the three major trends of large models in 2025: First, the upgrading of multimodal deep reasoning capabilities, which can improve the efficiency of financial research; second, breakthroughs in video generation technology, with better performance in aspects such as camera switching and subject retention; third, the enhancement of Agentic capabilities for complex multi-step tasks, which can integrate tools to control terminals and achieve automated processing.
However, the implementation of large models still faces challenges. Zhou Tianhong, Chief Information Officer of China Merchants Bank, pointed out that three major issues need to be addressed: how to adapt the foundation model to the business logic of the banking industry, how to suppress model hallucinations, and how to resolve the disconnection between technological R&D and business needs. He proposed a "three-pronged approach" solution — with context engineering as the foundation, enterprise-level knowledge management as the core, and post-training to strengthen professionalism — to help improve the ability of large models to solve professional problems.
Zhang Yu from Qingzhi Capital focused on technical bottlenecks, believing that the current large models have insufficient multimodal and scene cognition capabilities. Most of them rely on text semantic integration for information processing, which is quite different from the way the human brain works and has low efficiency. In the future, breakthroughs need to be achieved through algorithm optimization.
Future Direction: Scene Cognition Leads New Explorations
When talking about the future path of large models in the financial industry, experts made predictions from multiple dimensions. Technically, Zhu Jun believes that generative models will break through the digital scope and extend to the physical world: "At present, industrial robots are suitable for fixed scenarios. In the future, the demand for open scenarios such as offices and homes will be more urgent, and models need to have stronger adaptability and generalization capabilities."
Security and compliance are prerequisites for the application of AI in the financial industry. Tan Dai emphasized that only when technologies solve issues related to security, credibility, and compliance can large models fully penetrate into various links of enterprises: "Financial institutions rely on data-driven operations. If AI analysis assistants are provided to employees, the digital capabilities and decision-making efficiency of all staff will be significantly improved."
From an investment perspective, Zhang Yu is optimistic about fields such as embodied intelligence, life sciences, and industry-specific models. "These fields have already generated returns, and some have achieved profitability. In the AI ecosystem, traffic is the core driver, and any application may become a super entry point." He particularly pointed out that scene cognition is the core direction. Currently, AI is mostly passive and requires a large amount of input from users. However, truly intelligent AI should actively perceive the environment, such as automatically generating suitable PPTs during meetings; emotional value will also become a new track. "In the future, more than half of robots may be used for display and entertainment, and the supply of emotions will become a rigid demand."
Wei Jiangbo from China Merchants Bank expounded on the reshaping of banking models by AI: In terms of services, the focus will shift from key customers to "personalized services for each individual"; in terms of interaction, the limitation of graphical interfaces will be broken, allowing functions such as fund transfers and approvals to be completed with voice input; in terms of organization, AI will act as "digital employees" to improve efficiency and realize intelligent collaboration across departments.
In fact, financial institutions have already actively deployed relevant initiatives. WeBank has upgraded to an AI-native bank; MYbank has launched a matrix of AI products for small and micro enterprises; Industrial and Commercial Bank of China replaced the workload of over 42,000 people/year with AI in 2024; China Merchants Bank has invested in accordance with the "AI First" concept — in the first half of this year, the number of R&D personnel reached 10,800, and the investment in information technology amounted to 4.444 billion yuan.
With the continuous breakthroughs in technology and the deepening of applications, the financial industry is taking AI as the core driving force to embark on a new chapter of intelligent development. Explorations in key areas such as scene cognition, security and compliance, and personalized services will drive the financial industry towards a more efficient, accurate, and human-centered direction.
