Du Xiaoman and Tsinghua Economic Management released the ＂2024 Financial Industry Generative AI Application Report＂, and experts discussed the financial generative AI innovation ecology.
January 24, 2024,Tsinghua University School of Economics and Management, Du Xiaoman, MIT Science and Technology Review China, Tsinghua University School of Economics and Management Research Center for Dynamic Competition and Innovation Strategy.Co-sponsored "The release and seminar of 2024 Financial Industry Generative AI Application Report was successfully held in Tsinghua University Institute of Economics and Management.
The event invitedAcademician of the European Academy of Sciences, Professor of Changjiang Scholars, Guo Song, Professor of Computing Science and Engineering Department of Hong Kong University of Science and Technology, Li Jizhen, Professor and Vice President of Tsinghua University School of Economics and Management, Xu Dongliang, Chief Technology Officer of Du Xiaoman, Cao Gang, Vice President of Beijing Zhiyuan Artificial Intelligence Research Institute, Chloe Wang, Researcher of China Academy of Sciences and Chairman of Zhongke Wenge, and Sun Kewei, Technical Director of ICBC.Experts from other industries attended, and dozens of audiences from financial institutions, research institutes and Internet companies attended the report release and seminar.
(picture：Group photo of guests at the event site)
The conference was held byLi Jizhen, Professor and Vice President of Tsinghua University Institute of Economics and Management.Host. Professor Li Jizhen first introduced the relevant background of the conference and delivered a speech on behalf of Tsinghua University Institute of Economics and Management, welcoming the guests and all the audience.
(picture：Li Jizhen, Professor and Vice President of Tsinghua University Institute of Economics and Management.)
Subsequently, Professor Li Jizhen invited Xiaoman. Xu Dongliang, CTO, and Zhang Lan, deputy publisher of MIT Science and Technology Review in China, delivered speeches respectively.
In this paper, the large model and generative formula are summarized After the development status of AI, Xu Dongliang said that the application and breakthrough of the big model in the financial industry needs the concerted efforts of academic circles, industries and other relevant parties to make suggestions and put them into practice together.Du Xiaoman has actively explored the financial big model, and Xuanyuan, the first domestic open source financial big model, has been updated many times, showing outstanding financial problem solving ability, and successfully put it into production. AI is applied to the core value chain of enterprises such as risk control, marketing, customer service, office, research and development.
(picture：Du xiaoman CTO Xu Dongliang)
Zhang Lan, who gave a speech later, emphasized that the MIT Science and Technology Review has been Introduced to China by DeepTech in 2016, it has been focusing on the China market. The generative AI discussed in this activity is one of the top ten breakthrough technologies in the world in 2024 selected by MIT Science and Technology Review. Therefore, the exchange of AI applications in China’s financial industry and the release of "Report on the Application of Generative Artificial Intelligence in Financial Industry in 2024" have contributed inspiration and inspiration to the technology application and development in China market.
(picture：Zhang Lan, Deputy Publisher of MIT Science and Technology Review in China)
After the speech, the guests gave a keynote speech and shared their insights.
Guo Song: Generative formula Challenges and prospects of AI
Professor Guo Song from the Department of Computing Science and Engineering, Hong Kong University of Science and Technology,The topic of "Financial Industry Joins Hands with Generative AI, Challenges and Prospects" was shared.
Guo Song first discussed the generative formula. The development process and present situation of AI. Generally speaking, the complexity of the model ranges from "shallow" to "deep", the size from "small" to "large", the mass from "low" to "high" and the diversity from "single mode" to "multi-mode". Application types have gradually developed from content analysis to content creation, including text, images, audio and video.
In the financial field, generative AI’s ability to "let machines create" is leading the industry to a new chapter of digital and intelligent transformation. With excellent cost reduction, efficiency improvement, productivity improvement and innovative ability of products and services, it has given strong driving force to internal and external applications, thus bringing immeasurable application value to the whole industry.
Next, Guo Song discussed the generative formula The challenges that AI will face in the future, as well as his current work and prospects for the future with the whole industry. For example, by creating a brand-new Agent-as-a-service service model, the big model has the ability to solve financial practical problems; Aiming at the problem of isolated island of financial data, he led the team to study a federal fine-tuning algorithm for large-scale models, which improved the scalability of federal training.
"We also see some security challenges." Guo Song said, "How can we make it safe, especially in line with some regulations, how to turn rules and ethics into system tips, and implant system tips into large models to produce compliant content? These are also key issues."
(picture：Guo Song, academician of European Academy of Sciences, Professor of Changjiang Scholars and Professor of Department of Computing Science and Engineering, Hong Kong University of Science and Technology.)
Cao Gang: Reducing Cost and Improving Quality of Large Model
Cao Gangyi, Vice President of Beijing Zhiyuan Institute of Artificial IntelligenceThe topic of "Application Practice of Low-cost and High-reliability Large Model" shared Zhiyuan’s progress in large model and generative AI.
Cao Gang pointed out that a big wave of science and technology will always have some flaws, that is,"Small spots in the rough jade of the big model": cost and reliability. Among them, reliability includes hallucination and prejudice. He quoted Sam Altman, CEO of OpenAI, as saying that illusion is a feature rather than a defect, which can bring creative work.
Cao Gang shared Zhiyuan’s exploration in solving the illusion problem and cost, especially the method of training the model through growth technology. One way is to train the small parameter model first, and then gradually increase the parameters, which can effectively save computational power. He also added that in the process of training the modelThe method of "reducing cost and increasing efficiency" includes adopting growth technology to accelerate the learning process of the model.
"(We) train a model, let the model know what is right and what is wrong during the training process, speed up the absorption, and improve the model ability as much as possible in the small parameter model." He explained.
Finally, Cao Gang also mentioned Zhiyuan’s semantic vector model.The work of BGE aims to meet the requirements of the financial industry for low error tolerance rate. By making natural language text segments into vectors and training models to process the text, the research team improved the accuracy and production efficiency of the model, which was widely concerned and adopted by open source communities such as HuggingFace.
(picture：Cao Gang, Vice President of Beijing Zhiyuan Institute of Artificial Intelligence)
Chloe Wang: Exploration on the Application of Yayi Big Model
Followed by sharing is Chloe Wang, chairman of Zhongke Wenge, who discussed it from three aspects."Exploration on the application of big model technology in the financial field".
First of all, Chloe Wang deeply discussed the differences between the general big model and the domain big model. Large-scale model technology is fluent, general, unobstructed and universal.The "four links" feature, the most exciting of which is the general ability. For example, ChatGPT can perform well in many tasks and fields.
For the domain model, Chloe Wang thinks"There is no big difference between the two", but in terms of technical form, the general big model pays more attention to the learning of general knowledge of events, while the domain big model pays more attention to domain tasks.
From a business perspective, the big model has three business scenarios in the financial field: forecasting and decision support, risk management and monitoring, and compliance supervision automation. From the technical point of view, the related research mainly focuses on text processing, research and analysis, time series prediction, asset pricing, portfolio optimization and risk management, etc., and the application prospect is very broad.
Next, Chloe Wang focused on sharing the Yayi Big Model independently developed by Zhongke Wenge team, as well as its extensive application cases in the financial field. The Yayi big model has been iterated to now. Version 2.0 is the first batch of multilingual large models in China, with 30 billion parameters, fully autonomous model architecture and fully independent intellectual property rights.
According to him, in the cooperation with customers, the Yayi big model has been realized."Simplified and traditional Chinese-English mixed question and answer, cross-document long document analysis and reasoning, performance trend and complex data reasoning".
(picture：Chloe Wang, researcher of China Academy of Sciences and chairman of Zhongke Wenge.)
Yang Qing: Technical Practice of Du Xiaoman
Next, Yang Qing, general manager of Xiaoman Data Intelligence Department, shared."Technical practice of financial big model in Xuanyuan".
Yang Qing pointed out that for the financial industry, the big model will create value increment, because it needs to process a lot of data to support decision-making and forecasting, and the whole process needs a lot of manual participation. This coincides with the memory, understanding, planning and other abilities of the big model.
But he also believes that the landing of the big model faces many challenges, because the financial field has three remarkable characteristics: rich professional knowledge, data-driven and long chain. The challenges faced by the big model include the lack of professional financial knowledge, the inability to meet financial tasks, and the high cost of training and operation.
In order to unite the community to meet these challenges together, the degree is small and full In May 2023, Xuanyuan, the first large-scale financial model in China, was opened. In September, a large-scale model with parameters of Xuanyuan 70B was released. In January this year, Xuanyuan 13B model with lower cost was opened.
Yang Qing said that the degree is small and full"I hope to be positioned as the best big model of the financial industry, and all the tasks in the financial field outperform the advanced general model." To this end, they have done a lot of optimization work in financial understanding, application, dialogue ability and implementation.
Finally, Yang Qing said that Du Xiaoman will release more open-source financial models with different parameters, constantly iterate and upgrade existing models, and constantly reduce the user’s use cost. He hopes to use open source to promote innovation, build an ecology with industry partners, make progress together and create a win-win situation.
(picture：Yang Qing, General Manager of Xiaoman Data Intelligence Department)
Sun Kewei: A New Paradigm of Man-Machine Collaboration
Subsequently, Sun Kewei, technical director of ICBC, tookThe theme of "Application of Big Model in the Field of Financial Technology" was shared.
Sun Kewei pointed out that with the arrival of the big model, the application of artificial intelligence can be divided into three stages. The first is the initial stage of artificial intelligence, which is mainly built by purchasing. Secondly, the multi-dimensional attempt layout stage, using the mode of combining self-research and development, diversified attempts. Finally, in the stage of deep mature application, large financial institutions prefer self-research, supplemented by procurement.
In Sun Kewei’s view, the construction of artificial intelligence technology system in the financial field involves six aspects. The first isComputing power, algorithms and modeling,Computational power corresponds to the technical platform, and the algorithm corresponds to the crystallization of the highest wisdom. Then there iscompetenceBuild on, rely ondataAccumulation and applicationframeworkConsideration. The capabilities, data and framework are all around the business.
From the business point of view, artificial intelligence big model can be divided into four categories: basic big model, industry big model, enterprise big model and task big model. Another change brought by the big model is to create a new paradigm of human-computer collaboration. Sun Kewei believes that after the introduction of the big model, it can replace some people’s work, which brings a challenge of role and paradigm change.
"(We) need to solve the problems of technical stability and technical logic, comprehensively consider the technical advantages of large model technology in content generation, multi-mode, small sample and generate uncontrollable risks, and adhere to problem-oriented and demand-oriented." He explained.
(picture：Sun Kewei, Technical Director of ICBC.)
The financial industry, the future has come.
After the guest sharing, Professor Li Jizhen, the host of the conference, officially released.2024 Financial Industry Generative AI Application Report (hereinafter referred to as "Report"), and made a key interpretation of this report.
Professor Li Jizhen introduced that the Report was jointly compiled by Tsinghua University School of Economics and Management, Xiaoman, MIT Science and Technology Review, China and Tsinghua University School of Economics and Management, which fully integrated the advantages of Industry-University-Research. The report revolves aroundThe main thread of "How does the financial industry apply generative AI with the maximum benefit and minimum risk" not only pays attention to the potential risks of the application of generative AI in the financial industry, but also focuses on the successful solutions brought about by the iterative innovation of technology such as "Xuanyuan" model.
Professor Li Jizhen pointed out that the Report discussed the generative formula. The innovation mechanism of large-scale application of AI is analyzed, and the innovation password, industry value and policy expectation of generative AI application in financial industry are analyzed. On this basis, new viewpoints such as "Generative AI is a new quality productivity", "Generative AI is expected to bring 3 trillion scale incremental commercial value to financial industry" and "Generative AI is expected to usher in large-scale application in financial industry in three years" are extracted, which provides eight action guides for financial institutions to integrate generative AI.
Professor Li Jizhen finally pointed out that the Report shows that the future of the financial industry has come. Generative formula AI technology brings faster, more real-time and more personalized industrial changes, and its expansion in various key areas shows its far-reaching influence and great potential. It has given some financial institutions such as banks, insurance and funds a new dimension of creativity and efficiency, and is expected to bring incremental commercial value of 3 trillion scale to the financial industry. However, while enjoying the efficiency and convenience brought by the new quality productivity, we still need to face up to the practical challenges we face, especially in data security, risk prevention and control, ethics and supervision.
(picture：Li Jizhen, Professor and Vice President of Tsinghua University Institute of Economics and Management.)
Roundtable discussion
After sharing and reporting, we entered a round-table discussion session to discuss how to create a generative financial industry. The innovative ecology of AI invited all the guests present to discuss.
As a topic to attract jade, the guests made a big model.How long will it take for the killer application and the big model to really change human social life?
Guo Song believes that the so-calledThe application of "killer" can mainly produce huge benefits and positive impacts, but "scientific leap is not predictable", so it is necessary to "reshape traditional industries with generative AI" and create new values, which is "killer". On this issue, Sun Kewei said that the emergence of "killer" applications will be a process of cost and effectiveness game.
"Killer application takes time and will not be achieved overnight," Xu Dongliang, CTO of Xiaoman, added. "The flywheel of cost keeps falling, and when it reaches a certain critical point, it will achieve a qualitative leap. For Xuanyuan, the goal in 2024 is to upgrade the full-parameter series model system. "
In view of the pain points and sticking points in the application of large models, Ding Zhiyong believes that the most difficult thing is to determine."Whether the big model application is business-oriented or technology-oriented" needs to ensure that there is no problem on the business side, and this is the premise. As for the great opportunity of big model in the next two or three years, Chloe Wang thinks that AI will be a big industrial track in the future, and that "the software is completely AI-centric and uses AI to predict and support decision-making".
finally
In the heated discussion and unfinished communication, this time "The 2024 Financial Industry Generative AI Application Report was released and the seminar was successfully concluded. It can be seen that even though there are many difficulties and uncertainties in the landing of generative AI, all participants are full of optimism and expectation for its future. Obviously, whether it is finance or other industries, the changes brought by generative AI have quietly begun. Looking forward to 2024 and the future, it is increasingly important for the financial industry to actively embrace generative artificial intelligence, a revolutionary technology, and start a responsible artificial intelligence governance strategy, which will enable society to make full use of the transformative power of generative artificial intelligence and better enhance human welfare.
