At the end of 2022, ChatGPT became popular all over the world with a "hurricane", and the technology community's awareness of artificial intelligence has been raised to a new height. When this round of AI boom swept across China, all kinds of general-purpose large models sprung up like mushrooms after a rain, and it was a challenge to coexist with opportunities under the fierce battle, and how to commercialize it became a practical problem faced by technology companies.
In the second half of this year, an obvious change is that the domestic large-scale model track switched from the "volume" model to the "volume" application. The boom of general large models has made technology companies see the significance of technological innovation, and landing applications have become the key to industrial upgrading, and data has become the core of casting "intelligent brains".
For the financial industry, which is embracing the wave of digitalization, it has naturally become the best scenario for the landing of large models. Towards the end of the year, how much application value does the large model generate in the financial industry?What challenges remain?
Exploration of landing paths: focus on customer service, marketing and other links.
From perceptual intelligence to cognitive intelligence, artificial intelligence technology innovation has brought about qualitative changes in the development mode of many industries. Among them, the financial industry has a high acceptance of artificial intelligence. Thanks to the advantages of data-intensive industries, the financial industry's "chase" of large models can be said to be a matter of course.
At the end of July this year, Tencent Research Institute released a survey data: the number of large models with parameters above 1 billion in China reached 116, of which about 18 were large models in the financial industry. So far, how effective is the large model in the financial field?In this regard, the reporter of Huanqiu.com learned through an interview that the main application scenarios involve customer service, marketing and operation.
In the first quarter of this year, through the pre-training of financial exclusive data and the fine-tuning of business data, Lexin's self-developed large model Lexinpt was officially unveiled. The application of LexinPT has enabled Lexin to improve the efficiency of traditional human agents in customer service, telemarketing, and private domain operations. At the same time, Lexin reconstructed the workflow, knowledge flow and communication flow of operations, R&D, testing, data analysis, design and back-office functions with Lexinpt as the core.
Coincidentally, the "Tianjing" model released by Immediate Consumption has been applied in retail financial scenarios such as marketing customer acquisition, customer service, and operational asset management, and is used to provide personalized and humanized services to C-end users. Not only that, the model combines the value chain efficiency and decision-making science of the immediate consumption business to achieve the whole process transformation.
In the process of wealth management, the client used to face the situation that the information overload of the financial profession did not match the information he really needed. CICC Wealth Management said that it is uncertain how much the information provided by most non-professional customers can digest and absorb and how much it will help their personal finance. In addition, in the case of mismatched information, due to the complex design of financial products, it is easy to cause irrational redemption by customers when the market fluctuates.
The rapid development of artificial intelligence makes it possible for "mismatches" to be cracked. The relevant person in charge of CICC Wealth emphasized that the learning of the knowledge base through artificial intelligence can form the most professional digital employees at the platform level. Theoretically, one person can't integrate the intelligence of a team of 300 people, but artificial intelligence can. "Many factors restrict enterprises from connecting high-quality services to everyone, and digital employees can not only solve the problem of information overload, but also achieve thousands of people in information distribution, and form a customized asset allocation plan according to the customer's own situation. The future is very optimistic, there will be a company to do this, and we really hope to be this company. ”
As stated in the Caitong ** research report, the financial model is expected to reshape the ecology of the wealth management industry, including reshaping the customer service process and improving the customer experienceImprove content productivity and creativity;Improve the level of risk control. Among them, R&D mentioned that in terms of customer service, the access of large models changes the traditional human-computer interaction mode and greatly optimizes the customer experience under artificial intelligence services. A traditional chatbot is an automated program based on rules or predefined scripts that can only perform specific tasks. Relying on a massive professional knowledge base, the large model provides customers with 24-hour uninterrupted real-time services, and can independently generate creative content, while continuously iterating and improving the quality of generated content based on customer feedback.
In addition, the Yangtze River mentioned in the research report that with the further widespread application of large models in the financial field, artificial intelligence will have a huge impact on multiple scenarios, such as customer service, risk control, transaction investment, etc., to further promote the intelligence of the financial industry, reduce information asymmetry, reduce transaction costs, improve the level of risk control, fully meet the personalized needs of customers, and accelerate the digitalization process of the financial industry.
However, the person in charge of an enterprise in the industry pointed out that the large model is still in a relatively early stage, and there is still a huge space for technology and application scenarios to be explored, which falls in finance-related fields, and the application scenarios are mainly concentrated in the fields of telemarketing and customer service, and the depth and breadth need to be further strengthened.
In addition to exploring the application of large models in scenarios, such as the self-developed AIGC Zhongtai Lingxi launched by Zhongan Insurance in July this year, it can achieve "one MaaS (Model as a Service) platform and two application scenario strategies by adapting to mainstream large models at home and abroad" In order to help insurance institutions better adapt to AIGC capabilities, Lingxi can allow institutional users to embed industry professional knowledge bases in large models to achieve rapid adaptation of AIGC applications in the insurance vertical field. In addition, Lingxi also supports packaging enterprise internal application tools into large model plug-ins, so that the large model is closer to the business application scenarios, and the improvement of enterprise business capabilities is promoted through AIGC.
From general to financial, what are the opportunities and challenges?
Regardless of the type of application, the goal is to provide new impetus for business growth. At the same time, the application of large models faces some practical obstacles. Wang Peng, an associate researcher at the Beijing Academy of Social Sciences, pointed out that these include problems such as low data quality, high cost of data labeling, and poor model interpretability. In addition, financial institutions also need to consider how to ensure the stability and security of large models and respond to regulatory requirements.
In the view of Zhongan Insurance, basic large models such as GPT and LLAMA in the technical field are more used to solve general problems, and there will be problems of lack of professionalism in specific vertical fields. As a more professional and rigorous data-intensive industry than other industries, although the financial industry has accumulated massive data on transactions, customers, markets, risk control, etc., these data and data applications have not yet been effectively used on a large scale on a general model. At the same time, the rigorous professionalism of finance also requires that the application of large models be more accurate and effective when empowering specific business scenarios.
At the same time, security compliance is the foundation for the sustainable development of large models in the financial field. "Financial institutions should strengthen data security management, including data collection, storage, transmission and processing, to ensure data security and privacy. Zhang Xinyuan, head of research at the domestic consulting agency Co-Found think tank, said that financial institutions need to establish a sound network security system and take a variety of measures to prevent security risks, such as firewalls, encryption technology, access control, etc. Institutions should also strengthen employee security awareness education to ensure employees' safe behavior and prevent internal security risks.
According to the data, Lexin, as a fintech platform, has established a stricter data management and protection mechanism at the data management level, and ensured the security of the data level through effective data cleaning and verificationAt the level of model security protection, establish and improve model security protection rules, and adopt more rigorous algorithm design and review to ensure the security protection of modelsAt the level of monitoring mechanism and channel construction, we will build a complete artificial intelligence monitoring and control mechanism to systematically prevent compliance and security problems. At the same time, Lexin collaborated with the industry to call for the establishment of an effective set of industry rules and strategies to ensure the controllability of the final output content of the large model from the underlying norms of the industry.
Moreover, the low fault tolerance rate of financial services puts forward higher requirements for the accuracy of large models. However, there are great challenges in the robustness of large models, and there are still great challenges in the reliability of large models. The person in charge of immediate consumption believes that the large model is still facing huge challenges in the fields of transaction safety and life safety for a long period of time, especially in the fields of automatic driving, medical and health care, etc., which are related to social security and life safety, and the large model cannot give 100% correct advice, and the suggestions of the large model should be effectively used and effectively managed.
The person in charge pointed out that the large model technology has not yet formed a mechanism of continuous learning and reinforcement learning, and it is necessary to give full play to the wisdom of the crowd on the basis of federated learning to achieve mutual benefit and win-win results. In the financial field, financial institutions should make use of their own data advantages to establish an ecological mechanism of co-research, co-creation and sharing.
However, it is undeniable that the application prospect of large models in the financial field is very broad. Wang Peng pointed out that with the continuous development of technology and the continuous expansion of application scenarios, large models will play an increasingly important role in the financial industry. In the future, the large model will be applied to more financial business fields, such as investment decision-making, risk management, asset allocation, etc. At the same time, the large model will also be combined with other technologies, such as blockchain and cloud computing, to form a more complete fintech ecosystem.
The relevant person in charge of Transwarp Technology said that the general scenario is based on the basic large model, which has a shorter landing time and faster application speed. The proprietary large model needs to go through the process of development, accumulation and training of proprietary domain corpus, and once it is implemented, it will have a more significant effect on the business effect. In the future, with the improvement of China's intelligent computing power, general and proprietary domain corpus data, large model technology and application construction capabilities, it will bring comprehensive artificial intelligence technological innovation to the financial field. It is reported that at the end of May 2023, Transwarp launched Transwarp Infinity, a large financial model of Transwarp, mainly focusing on the field of capital market analysis. 