In the afternoon of October 17, the Baidu World Financial Big Model Frontier Development Forum was held in Beijing. Wu Hequan, an academician of the CAE Member, and more than 10 financial industry executives from Du Xiaoman, SPDB, Ping An Bank, Everbright Bank, Taikang Insurance, etc. jointly discussed how to implement the big model in the financial industry.
General models are difficult to handle financial domain tasks, and financial big models are the only way for big models to be implemented in the financial industry, "said Xu Dongliang, CTO of Du Xiaoman, at the meeting. He believes that the competition pattern of the universal big model has been determined, and that the competition in the next stage of the big model is the competition of the industry big model.
Since the development of big models, we have seen that everyone has different approaches. For example, from the hierarchical structure of the model, there are universal big models, industry big models, enterprise big models, task big models, etc.; from the technical route, there are open source big models and closed source big models. "He suggested that financial institutions should first establish their own enterprise big models and task big models on the basis of industry big models, It is neither economical nor necessary to train a large industry model from scratch.
Xu Dongliang stated that "the Du Xiaoman Financial Big Model has been trained with massive financial data accumulated in the company's business scenarios, and its logical ability has been enhanced through the knowledge graph of millions of nodes, balancing the ratio of general data and financial data to solve the problem of forgetting the general big model. It has a more professional and accurate understanding of financial knowledge, higher controllability and security.
In May of this year, Du Xiaoman opened up the first large-scale financial industry model in China, and over a hundred financial institutions have applied for trial use. In September, C-Eval and CMMLU announced the results of the Big Language Model Evaluation Benchmark, and the Du Xiaoman Financial Big Model ranked first among all open source models on the two authoritative lists. It is also the first financial big model in China to rank first on both authoritative lists.
The development of large model technology is rapid, and model-based applications are also rapidly iterating. Xu Dongliang believes that applications based on the big model are more and more closely integrated with enterprise business scenarios. From the earliest pure native applications such as ChatGPT and ERNIE Bot to "Application+AIGC", such as Office 365 Copilot, the future application trend will be AI native reengineering: deep integration of business scenarios and restructuring of business processes.
He believes that in the To B scenario, large models create business value and must be deeply embedded in business scenarios, refactoring service processes and user experiences. Only by injecting scenario data can the task capabilities of financial big models be stimulated. Taking the intelligent marketing assistant in the credit industry as an example, when an intelligent marketing assistant who has not undergone scenario data training communicates with a customer about "why there is a quota not used", if the customer expresses that the reason is "the quota is too low", the marketing assistant cannot provide further solutions; After training with millions of hours of financial dialogue data, intelligent marketing assistants can further inquire if customers have other materials to prove their asset situation.
A marketing assistant based on a large model can only provide valuable solutions to customers, help them solve problems, and help enterprises improve business conversion rates, in order to form a 'model business' data flywheel and continuously iterate to improve model effectiveness, "Xu Dongliang concluded.
The big model is bringing a window period of technological 'curve overtaking' to financial institutions, and the financial industry will also usher in a strategic opportunity period for the industrial application of the big model, "Xu Dongliang predicted." However, it is difficult for any financial institution or technology company alone to achieve the industrial application of the big model, and the value creation of the big model requires the joint participation of the entire industry.
At present, Du Xiaoman is working with Baidu Cloud to build a financial industry solution based on ERNIE Bot. "We hope that on the basis of ERNIE Bot, together with our financial data and scenario data, we can adapt the model to the business scenario, so that the big model can grow from a fresh graduate to an excellent employee of the enterprise".