Recently, at the "Focusing on Future Insight AIGC New Paradigm" AI Forum held by KPMG China, Fei Haojun, Chief Algorithm Scientist of Qifu Technology, shared his views on the current status and challenges of financial big model applications at the parallel forum Fintech Salon.
Fei Haojun stated that with the rapid development of AI technology, the application of financial big models has gradually moved from theoretical exploration to practical implementation. In the past year, the industry's focus has shifted from simply pursuing the improvement of basic model capabilities to how to more effectively integrate big model technology into actual business scenarios, which has brought many technological innovations and breakthroughs. He mentioned that in order to reduce application costs and improve model stability, the financial industry has begun to explore the application of small parameter models and has made significant progress in knowledge enhancement capabilities. At the same time, the development of multimodal technology has brought new vitality to financial big models, especially in areas such as table parsing and employee work assistance, which have shown great potential.
One is the exploration in the field of customer service. Qifu Technology uses big model technology to deeply analyze millions of historical call records, extract efficient communication strategies and speech templates, and significantly improve the accuracy of customer service personnel in capturing user intentions. This system not only improves the efficiency and accuracy of customer service communication, but also achieves 100% coverage of calls through automated quality inspection. Since its implementation, while improving human efficiency, management efficiency has achieved a significant increase of 50%.
Another is the intelligent generation of marketing materials. In Qifu Technology's advertising materials, approximately 70% of the image materials and 20% of the video materials are generated by large models and optimized through a multi-dimensional rating system. This intelligent material generation method not only improves the output efficiency of advertising materials, but also enhances the attractiveness and conversion rate of advertisements by accurately analyzing user preferences. After using the large model for marketing material production, the customer reach of Qifu Technology increased by 21.4%, and the overall advertising quality was greatly improved.
Fei Haojun stated that traditional artificial intelligence technology often focuses on clear business results as learning objectives, and constructs countless small models to form specific tools for executing various tasks. These tools are highly consistent with business goals, ensuring that they can continuously evolve with the improvement and iteration of the business. For large models, in order to play a more critical role in the core business modules, they also need to have predictable iterative capabilities like small models, that is, the ability to guide the model through targeted optimization through data feedback. In the near future, financial big models will be deeply applied and integrated in a wider range of business scenarios, injecting stronger impetus into the intelligent transformation of the financial industry.