Competing for domestic financial GPT: How the big model redefines financial technology

Source: Business Show

Image source: Generated by Unbounded AI‌

The large-scale model battlefield in China, after several months of hurricane and gathering to show off its "muscles", is ushering in a new battle in the cooling down.

Since March this year, as the generative AI represented by ChatGPT has triggered a new wave of technology, more than 20 domestic Internet companies have launched large-scale models.

By the 2023 World Artificial Intelligence Conference on July 6, it can be described as a "hundred-model battle", and even formed a "thousand-model war", comparable to the "thousand-regiment war" during the Internet development period.

No one wants to be left behind in this wave of AI megamodels. But right now, the development of large models has entered the "vertical" stage from the "general" stage.

**More and more companies have rationally realized that only a few top giants in general-purpose large models can use "computing power, algorithms, data" or even manpower and financial resources to do things all in, and focus on scene applications, customizing and adapting It is more worthy of small and medium-sized enterprises to invest in a large vertical model. **

Many companies simply train a vertical model that adapts to the scene directly based on the "base" of large domestic and foreign models because they have accumulated data and other advantages in their fields that have been cultivated for many years.

For example, in the financial field, since May this year, a group of financial technology companies such as Qifu Technology, Du Xiaoman, Lufax Holdings, Ant Group, and Mashou Consumption have deployed large AI models based on their own scenarios and data advantages.

An insider from a leading financial technology company told "Business Show" that in the past two months, all financial technology companies and leading financial institutions that have large-scale model building capabilities in the financial industry are starting from the exploratory stage Enter the stage of landing application.

**The person further stated that financial technology companies or financial institutions with their own business scenarios will give priority to internal use, and improve the capabilities of large models through the polishing of internal products. Technology companies that do not have their own business applications are more inclined to the general problem-solving capabilities of the financial industry. Some will cooperate with financial institutions to jointly create large models of the financial industry and scenarios. **

A wrestling about the big AI financial model kicked off.

What impact will this new revolution in the financial technology field triggered by the AI model bring to the industry? Due to its high degree of dataization and high professional complexity, what opportunities and challenges will the financial industry face after accessing the large model? How will the financial model evolve in the future?

Compete for domestic financial GPT

There is no doubt that in 2023, the development of AI will usher in a new era.

In March, ChatGPT, launched by the artificial intelligence laboratory OpenAI, was born, detonating a new wave of global AI large-scale models, opening a new era of AIGC, and related industries have also ushered in a revaluation.

It didn't take long for the hot air of the AI model to blow to the financial circle. On the 30th of the same month, Bloomberg launched a large-scale language model for the financial industry - BloombergGPT. This is seen as an event that could have a significant, if not disruptive, impact on the financial sector.

After two months, the domestic financial field also ushered in its AI model moment. In mid-May, Qifu Technology first announced the launch of a self-developed general model for the financial industry - Qifu GPT, which is known in the industry as "the first general model for the financial industry in China".

According to Qifu Technology, the product-level applications it supports are expected to be launched within this year and open for use by financial institutions.

An insider of Qifu Technology told "Business Show" that as early as last year, Qifu Technology began to lay out and try to apply generative large models in some internal scenarios. And in March of this year, after the large-scale model became popular, Qifu Technology also quickly established a large-scale model research department to accelerate research and development and promote the application of scenarios.

On February 9 this year, Zhou Hongyi, the founder of 360 Group, and Zhang Chaoyang, the founder of Sohu, put forward a point of view during the dialogue on "Dialogue Under the Stars": **If a company cannot catch the ChatGPT bus, it is likely to be eliminated. **

Earlier, Qifu Technology CEO Wu Haisheng also said that he is currently at the crossroads of the technological revolution, from cloud computing to ChatGPT, which is now popular all over the world, and will be committed to applying these technologies to the financial field to provide partners and users of financial institutions More efficient technology services and solutions.

It is not only Qifu Technology that is taking the lead in the layout. In late May, Du Xiaoman also announced the launch of "Xuanyuan", the first open-source large-scale model for the vertical financial industry in China, and then Lufax Holdings, Xinye Technology, etc. also announced the layout and exploration of generative large-scale model applications. On June 21, Ant Group responded that it was developing a language and multi-modal model called "Zhenyi"; on the 28th of the same month, LightGPT, a large model of Hang Seng's electronic financial industry, was also unveiled.

At the 2023 World Artificial Intelligence Conference on July 6, as many as 30 large-scale models from home and abroad were unveiled, and how large-scale model technology is applied to vertical fields such as finance has also become a hot topic. Jiang Ning, CTO of Immediate Consumers, pointed out in an interview with the media that the large artificial intelligence model has brought "a booster" to the financial industry. At the same time, he also revealed that Immediate Consumption will also launch a large financial model.

In just four months, various financial institutions and financial technology companies are gearing up and competing for deployment, and GPT in the domestic financial field is about to emerge.

Large Model Consensus: From General to Vertical

While various companies are racing against time to launch large-scale financial models, the industry has gradually reached a consensus: large-scale models must enter the vertical stage from the general stage.

At the 2023 Global Digital Economy Conference on July 2, Xu Dongliang, CTO of Du Xiaoman, also put forward a similar point of view - "Compared with the capabilities of general-purpose large-scale models, the financial industry is in great need of vertical industry large-scale models."

Xu Dongliang further analyzed that due to the high requirements of the financial industry in terms of data security and privacy, risk control, precision, and real-time performance, the general-purpose large model lacks necessary training data in terms of financial capabilities. Neither accuracy nor accuracy can meet the minimum requirements of this industry, so a large industry model customized for financial institutions is needed to be effective.

The relevant person in charge of Qifu Technology also said that the biggest difference between the large-scale model of the financial industry and other industries lies in the business complexity of the financial industry, the requirements of industry operating regulations, and security and privacy protection, which makes the financial industry more special than other industries , the business is more complex, the requirements for industry operating norms are higher, and the requirements for security and privacy protection are higher.

Immediate Consumer CTO Jiang Ning pointed out at the 2023 World Artificial Intelligence Conference that due to the characteristics of the financial industry, such as "data-intensive and technology-intensive", this industry has always hoped to capitalize data, but at the same time it is also facing challenges such as offline bank outlets. Issues such as value delivery efficiency and user experience require organizations to continue to innovate.

**That is to say, from the perspective of large model logic, existing large models cannot cover all industries. On the basis of general large models, enterprises need to fine-tune training and customize large models for vertical fields. **

The relevant person in charge of Qifu Technology said that especially for the highly data-oriented and professionally complex financial field, it needs to rely on more professional background and industry insights to optimize and adapt to specific application requirements.

** From the perspective of industry development needs, the financial industry has also entered the stage of stock competition from an incremental market, and the entire industry is facing difficulties such as difficulty in customer retention and intensified competition. At this time, it is even more necessary to use technology to improve operational efficiency and user experience. **

Considering the actual implementation of technology-enabled user experience, traditional financial services still generally face the problem of "difficult discovery, difficult experience, and difficult service" in the process of improving user experience. The emergence of AI large models can help the financial industry solve these problems to a large extent, so as to better serve users.

But now, there is still a huge gap between the general-purpose large model and the application of financial scenarios. Therefore, only by continuously optimizing the existing general-purpose large-scale model and forming a vertically professional large-scale model in the financial field can the large-scale language model serve enterprises and users better.

However, compared with other fields, finance has higher requirements for data expertise, risk control, compliance, and security, which also brings many challenges to financial institutions and enterprises in exploring large models in vertical fields.

Redefining Fintech

Looking back at the three waves of artificial intelligence development, the development of artificial intelligence technology is driven by three major elements: algorithms, computing power, and data—algorithms determine whether the designed "brain" is smart enough, and only high-performance computing power can train A large network must also have the support of big data.

In just half a year, with the rise of AIGC represented by ChatGPT, the era of artificial intelligence big models is coming. When the artificial intelligence model meets finance, technological changes and business space will be further opened, and the value of all industries will usher in a revaluation. According to iResearch, the core market size of AI+finance will reach 29.6 billion yuan in 2021, driving related industries to 67.7 billion yuan.

It can be said that the emergence of AI large models has redefined financial technology to a large extent. For example, AI large-scale models help companies reduce costs and increase efficiency, build virtual customer service online interactions, and provide users with more humane services. Financial GPT can realize the automatic generation of texts of financial information and product introduction content, and improve the efficiency of content operation of financial institutions.

Take Qifu GPT, a large model of Qifu Technology, as an example. It has been applied to business links such as customer acquisition, operation, risk control, and post-loan service. At the marketing level, build a dialogue financial business scenario through a large model, train the existing telemarketing dialogue system, help telemarketing robots accurately understand real user needs, and improve the fidelity of the response and the professionalism of the service.

The relevant person in charge of Qifu Technology said, "With the help of the large-scale model sparring robot, the call time of the telemarketing system has increased by 15.1%. In terms of risk control in the core business link in the credit field, the interpretation of intelligent credit information derived from the large-scale model as the core , can help financial institutions understand and judge users more comprehensively and efficiently.”

It is understood that the current Qifu Technology team is combining the financial industry and internal proprietary data to do incremental pre-training and tuning of large models, and relying on internal business to carry out practical application in some small and medium scenarios.

**However, insiders of the above-mentioned leading financial technology companies told "Business Show" that the current domestic financial large models are mainly used in small and medium-sized areas in some independent business scenarios, and then observe the impact of large models on business growth and risk control. The optimization capabilities in terms of human efficiency improvement and human efficiency improvement have not yet ushered in the stage of large-scale commercialization. **

At present, the domestic financial large-scale model is still facing many challenges, and it will take time to realize large-scale application.

Jiang Ning, CTO of Immediate Consumers, believes that there are still four major challenges in the current large-scale model of the financial industry:

First, in the face of key tasks and unpredictable external changes in the financial industry, large models cannot guarantee the stability and accuracy of every decision; second, the financial industry hopes to use artificial intelligence to achieve personalized user experience, but it requires personal The integration of private data and large models still has compliance and security issues; third, the financial industry has always had the problem of "data islands". Large models require the construction of a networked platform for enhanced learning and continuous contribution of data and feedback. However, the current market Fourth, the application of large-scale models in the financial industry puts forward higher requirements for hardware and software facilities such as underlying equipment and infrastructure.

The relevant person in charge of Qifu Technology also said that one of the main challenges facing the development of general financial models is the complexity of data processing. In addition, the protection of data privacy and information security must also be considered. The person in charge also pointed out that the difficulty of the general-purpose financial model mainly lies in the accuracy of the model and the flexibility of practical application. Extensive, interfaces need to be reserved for free connection in practice to adapt to ever-expanding application scenarios.”

Looking at the history of fintech development over the past decade, it is a large and growing industry. In the field of artificial intelligence, the financial industry has been continuously exploring over the years. What we can see is that artificial intelligence has appeared in fields such as payment, investment, loan, personal financial management, anti-fraud banking and insurance.

**However, it cannot be ignored that the essence of finance is still risk management, and risk control is the core of all financial businesses. Entering the era of AI big models, the role played by artificial intelligence big models, in addition to making financial business services and user experience better, its core is still to help minimize risks. **

Of course, in addition to considering risk control and the integration of technology and scenarios, human participation cannot be ignored. In the process of machine learning, human participation in training is required in generative artificial intelligence. In the field of large financial models, human participation in all aspects is equally important.

In the wave of technology triggered by this AI model, a new financial technology revolution has quietly started. Every enterprise and even everyone should not miss it.