In an astonishingly rapid turnaround, the securities industry is embracing a new wave of innovation brought forth by DeepSeek, a leading player in advanced AI technologiesThis phenomenon is not merely a trend but a significant shift that many brokerages are actively adopting to enhance their operational capabilities and service offerings.
Preliminary estimates indicate that more than ten brokerage firms have announced their successful deployment of the local version of the DeepSeek R1 modelNotable firms include prominent names such as Guotai Junan, Guojin Securities, Zhongtai Securities, and many othersThese institutions are leveraging the potential of AI to streamline complex processes and improve efficiency.
According to sources within the sector, the practical applications of these AI models are currently focused on internal processesThis approach enables firms to mitigate risks associated with AI, such as the infamous "hallucination" problem, where AI systems generate false or misleading informationWhile the primary use remains employee empowerment, some brokerages are also venturing into client-facing applications.
Industry experts view DeepSeek's evolution as offering unparalleled opportunities for innovation in the financial sector, creating an environment ripe for fresh business models and a new driving force for growth that could reshape the landscape of finance.
Over Ten Brokerages Complete Local Deployment of DeepSeek R1 Model
On January 20, 2025, DeepSeek captured the global spotlight by launching its first reasoning model, DeepSeek-R1, trained through reinforcement learning (RL). In its wake, brokerages quickly began reconfiguring their operations to delve deeper into AI's capabilities across core business functions.
A representative from Guotai Junan reported that the firm completed the local deployment of the DeepSeek R1 model before the Lunar New Year, and it is currently engaged in application testing
Advertisements
This move is expected to enhance the capabilities of its "Junhong Lingxi" AI model in investment research and intelligent services.
The introduction of the DeepSeek R1 model empowers Guotai Junan's services in various areas such as intelligent Q&A, investment services, regulatory compliance, market analysis, and document processing, promising a more targeted and efficient AI-driven experience for both employees and clients alike.
Similarly, representatives from Zhongtai Securities confirmed that upon the launch of DeepSeek R1, they promptly executed the localized deployment of several modelsThey anticipate using DeepSeek V3 in mid-2024, combining expert rules and cognitive chain technologies to achieve breakthroughs in financial news text mining and analysis.
GuoXin Securities also reported that on the first day after the Lunar New Year, they've successfully completed the local deployment of the DeepSeek-R1-Distill-32B model and have integrated cloud versions of V3 and R1.
In addition, Guojin Securities recently announced that they have completed local deployment testing of DeepSeekThey plan to explore applications across various scenarios, including intelligent office automation, public sentiment monitoring, and analysis of market trends.
Meanwhile, Guoyuan Securities has carried out local deployments and adaptability tests of the DeepSeek-R1 model
Advertisements
They plan to integrate this model into their own intelligent service platform, targeting enhanced semantic understanding and multi-turn conversational capabilities, particularly in compliance Q&A and business guidance.
Everbright Securities has also reported the successful deployment of DeepSeek AI models and extensive application testing, utilizing Huawei's NPU computing platforms for domestic adaptation.
Xingye Securities claimed to have established a digital intelligence platform supporting integration of various models, including DeepSeek’s V3 and R1, into their broader operational framework, thereby enhancing diverse business functions.
Other firms like Huazhong Securities, Huafu Securities, and Dongxing Securities have also completed local deployments of the DeepSeek-R1 model, marking widespread acceptance of this innovative technology across the sector.
Addressing the Hallucination Problem in Large Models
DeepSeek-R1 is praised for its innovative algorithms and exceptional performance, boasting a dual advantage of "performance multiplication and cost reduction," which are accelerating the application of artificial intelligence across specialized fields
Advertisements
However, how effective is its implementation among brokerages, and how do they plan to mitigate the hallucination issue associated with large models?
Representatives from Zhongtai Securities divulged that they have established 215 knowledge bases using DeepSeek, deploying it in operational assistants and Q&A services, achieving accuracy rates exceeding 95%.
“Preliminary validations across various business scenarios indicate that DeepSeek outshines previous open-source models, particularly in areas like intelligent Q&A and stock analysis, revealing vast potential for integrated business applicationsWe plan to extend its use across essential securities operations, including our financial management software and investment banking,” said a representative from GuoXin Securities.
Zhongjin Wealth echoed similar sentiments, lauding DeepSeek for its significant efficiency advantagesIts natural language processing and event reasoning capabilites enable the automation of analyzing unstructured texts, assisting financial advisors to swiftly tackle market insights, thereby boosting decision accuracy and processing efficiency with daily documentation surpassing ten thousand entries—a staggering 90% more efficient than traditional methods.
As brokerages harness AI to boost operational efficiency, avoiding the "hallucination" dilemma—where AI generates inaccurate or misleading content—remains crucial
An industry expert from East China pointed out that the hallucination issue is an inherent flaw in current technology, posing significant challenges without perfect solutions, necessitating selective application of these large models.
“The existing large model technology is better suited for internal empowerment scenarios, serving employee functions while playing a supplementary role in client services,” they noted.
Moreover, another representative from Zhongtai Securities emphasized that their application of the DeepSeek model is primarily for internal use. "To counter the hallucination problem, we rely on our own judgments, whether it pertains to the application of output results or adjustments in the model itself, which involve substantial human oversight," they asserted.
“For now, we are using it internally as our employees effectively carry out manual reviews for AI outputs,” stated another representative from a major brokerage.
Beyond primarily empowering employees, the aforementioned expert believes the basic principles for deploying high-capacity models should encompass two aspects: prioritizing private deployment while supplementing with public services for data security; and focusing on technology acquisition while balancing in-house development for personalized functionality.
What New Changes Await the Financial Industry?
With its robust entry into the realm of high-capacity models, DeepSeek is poised to redefine innovation within the brokerage community
But what specific new business scenarios could arise, and what changes could unfold across the broader financial industry?
Zhang Guowei, Chief Information Officer at Guoyuan Securities, asserts that DeepSeek will foster many inventive business scenariosThe model’s high-performance APIs can enhance the capabilities of intelligent customer services, facilitate responsive engagement with clients, and contribute to the development of advanced apps like stock-trading bots by integrating complex reasoning abilities into consumer technology.
In addition, Zhang believes DeepSeek's powerful code generation abilities could streamline developers' workflows and facilitate the emergence of "conversational programming," allowing for the swift production of high-quality, customized content.
In terms of how emerging financial practices could evolve, Zhang highlighted the possibility of constructing integrated intelligent customer service systems that provide around-the-clock support and personalized financial product recommendations using DeepSeek’s large modelsAdditionally, they can enhance risk management capabilities, improve predictive abilities, and optimize credit evaluations and risk pricing.
“Moreover, DeepSeek represents a stepping stone for financial innovation, enabling the creation of novel products, reducing service costs, and promoting inclusive financial development,” Zhang elaborated.
Wang Hongtao, Chief Information Officer at Guojin Securities, supports this sentiment, noting that DeepSeek affords unprecedented opportunities for innovation and application of AI in finance
Advertisements
Advertisements