The world of technology is undergoing rapid transformation, particularly with the rise of artificial intelligence (AI) and its various applicationsIn this current landscape, a remarkable advancement has emerged in the form of the DeepSeek model, a new player in the market that promises to change the foundational principles of technological evolutionIts growth has been phenomenal, quickly amassing a staggering 30 million daily active users in a short spanThe question arises—how can we critically evaluate the ongoing AI phenomena, and what is the true potential of DeepSeek?

Recently, an engaging seminar titled “How Can DeepSeek Change the Rules of the AI Game? Is the High Barrier to AGI Disappearing?” was held, co-hosted by Xinxin Bank and the School of Management Science and Engineering at Southwestern University of Finance and EconomicsKey speakers included Li Xiusheng, vice president of Xinxin Bank, Wang Jun, professor and director of the Computational Finance Department at the university, and Wei Hao, who heads the Risk Control Science Department at Xinxin BankThey shed light on the technological secrets behind DeepSeek and its prospective applications in the banking sector.

One major theme discussed was the comparison between open-source and closed-source models and their implications for competition within the industryAs the collaboration between OpenAI and Microsoft has sparked debates over possible monopolistic practices, and NVIDIA faces stringent export controls for AI chips, DeepSeek’s open-source approach presents a refreshing alternativeUnlike conventional closed AI models, DeepSeek’s accessibility allows companies to utilize advanced models at a substantially lower cost, enhancing the functionalities of intelligent assistants across various scenarios.

The software industry has long seen both open-source and closed-source models coexisting, with each demonstrating unique success storiesLi Xiusheng pointed to Linux and Android as prime examples of open-source software fostering significant advancements in the operating system domain

Advertisements

Meanwhile, he acknowledged Apple as the epitome of a closed-source model, successfully maintaining its dominance in the premium smartphone marketAlthough these entities adopt divergent paths, both have achieved remarkable successes.

"From the perspective of absorbing global contributors, I am personally more optimistic about the open-source model," stated Li Xiusheng. "It allows for the aggregation of diverse insights and capacities, collectively propelling technological progress and innovationIn the future, it is likely that open-source and closed-source will continue to develop in parallel, but the potential of open-source is certainly worth anticipating."

Wang Jun complemented this view, suggesting that the relationship between open-source and closed-source reflects a blend of competition and collaborationWhile open-source technologies attract a broad range of developers, fostering rapid iterations, they face uncertainties regarding profit generation and commercial viabilityConversely, closed-source models emphasize building proprietary advantages, often at the cost of flexibility and adaptabilityEach avenue presents respective strengths and weaknesses, thereby creating a scenario where both models learn from one another and occasionally enter competitive terrains.

From a market perspective, the introduction of DeepSeek as a low-cost yet highly effective open-source model has significantly impacted leading technology firms. "For closed-source models like OpenAI, DeepSeek's pricing strategy compels them to reassess their business models and technical optimization directionsMeanwhile, for chip manufacturers like NVIDIA, DeepSeek’s launch indicates that high-end GPUs are not the only way to achieve superior inference, prompting a reevaluation of investment logic and developmental strategies in AI infrastructure," Wang Jun clarified.

However, there are notable challenges that emerge when addressing general artificial intelligence models in the realm of digital risk management

Advertisements

Wei Hao explained, "Despite the broad capabilities of large models—such as problem comprehension, mathematical operations, and code generation—performance in specialized fields like risk control can be unsatisfactory." This stems from the fact that these large models usually rely on publicly available data and codes that lack the specialized datasets necessary for training within the risk control domain, thus not fully aligning with the specific needs of risk management.

As smaller banks look to leverage DeepSeek for a competitive edge, the quest to cultivate smart technology application capabilities also springs to the forefrontResearch revealed by Zhejiang Merchants Securities indicates that the training process for DeepSeek-V3 required less than 2.8 million GPU hours—starkly contrasting with the staggering 30.8 million GPU hours required for Meta's Llama3-405BWith training costs pegged at around $557,600, DeepSeek demonstrates a stark contrast to the hundreds of millions spent on other models such as OpenAI's ChatGPT-GPT-4.

In comparison to the traditional models that demanded investments in the millions, costs associated with DeepSeek’s localized deployment are comparably less than one millionRecently released updates from the Ministry of Industry and Information Technology indicate that three major telecom companies have fully integrated the DeepSeek open-source modelIn the financial sector, numerous institutions—including banks, funds, and securities firms—are actively adopting DeepSeekFrom May 2024, Xinxin Bank began utilizing DeepSeek in its system development activities, creating both a knowledge inquiry assistant and a code completion assistant to significantly streamline the research and development process for engineers.

Li Xiusheng contends that DeepSeek’s emergence has heralded two key ideological shifts in the world of artificial intelligenceFirst, the notion of achieving miracles through heavy computational power has been dislodged; it is no longer necessary to chase after extreme computing capacity for breakthroughs

Advertisements

In the past, it was a widely held belief that only through sheer computational force could monumental advancements be madeHowever, DeepSeek has demonstrated that optimizing algorithms and models can yield highly efficient performance even with lower computational resourcesAdditionally, it has intensified the ongoing rivalry between open-source and closed-source paradigmsWhile OpenAI’s ChatGPT has popularized the large model concept, its closed-source strategy has limited widespread technological adoptionThe advent of open-source models like DeepSeek has lowered barriers to entry, enabling more institutions to leverage large models—a shift that will potentially transform sectors like banking.

"Looking ahead, as technologies continue to advance and costs decline, large models will no longer be exclusive to major banksInstead, they will become widely accessible across smaller financial institutions as well, leading to significant technological transformations within commercial banking," Li Xiusheng asserted.

In terms of digital risk management within the banking sector, technologies like DeepSeek are poised for extensive applicationsWei Hao shared his enthusiasm, stating that DeepSeek can rival the leading inference models offered by OpenAI, being open-source, permissively licensed, and capable of local controlHe elaborated on practical experiences with DeepSeek in processing unstructured data, highlighting its enhanced semantic comprehension and text processing abilities. "We can extract information from a broader database thanks to these capabilitiesAdditionally, techniques from general intelligent models can be leveraged to improve client assessment accuracy, facilitating better decision-making," he noted.

Furthermore, Wei emphasized DeepSeek R1's depth of thought capabilities, which can enhance intent and semantic comprehension through a chain of thought training modelThis capability excels not only in Mandarin but also shows robustness in navigating complex intents and lengthy context.

The banking industry, recognized for its high level of informatization, has undergone numerous significant transformational phases, from replacing manual operations with computerized systems to adapting to mobile internet advancements

Now, as AI continues its rapid advancement, banks are facing what some refer to as the fourth wave of information system evolutionThe pertinent question is how banks can establish a suitable smart technology application capability within this large model arena.

Li Xiusheng proposed that the arrival of the large model era urges banks to rethink their operational management and processes through the lens of artificial intelligence applicationThe priority should be on developing applications, followed by organizing data, enhancing data quality, tagging data accurately, and employing external data sourcesIn this regard, it is crucial for commercial banks to engage in strategic considerations while simultaneously addressing computational power, data, algorithms, and application variables comprehensively.

Since its inception, Xinxin Bank has been fully employing AI technologies to combat fraud and manage credit risks, enabling highly efficient large-scale loan processingHowever, the emergence of large models has stimulated banks to explore and test applications across various fieldsCurrently, Xinxin Bank has successfully integrated large models in customer service, replacing certain human roles and is now examining applications primarily in marketing and post-loan management.

Apart from the banking sector, Wang Jun predicts that significant advancements in intelligent applications linked to large models will occur across industries such as manufacturing, climate risk assessment, computing, education, and media entertainmentFor instance, within manufacturing, large models could be applied to monitor the reliability of components or batteries and predict their lifespanIn climate risk forecasting, AI can interpret future weather patterns to provide alerts and optimize routes, particularly for highwaysIn computing, large models enhance code completion and understanding, while in educational contexts, personalized large models can assist students based on their learning habits and behaviors

In media entertainment, large models can aid in content production and scene building, such as animation and gaming design, and even synthesize digital personas for e-commerce recommendations.

As AI technology broadens its reach, a critical question emerges: will it take away jobs or create new opportunities? The intersection of finance and artificial intelligence is inherently synergisticAI large model technologies can exploit the vast troves of data within the banking sector, where a variety of applicable scenarios existCurrently, AI models are catalyzing a revolution across multiple banking domains, thus expediting the arrival of what many refer to as the 'future bank.'

Alongside this widespread integration of large models, the demands on tech staff have become more rigorousLi Xiusheng articulated a need for a shift in talent requirementsIn the internet domain, proactive mindsets contributed significantly to the success of tech companiesWith the dawn of the AI era, there is an increasing demand for hybrid talents combining finance and AI thinking.

Xinxin Bank has underscored the importance of internet thinking and will place a strong emphasis on AI thinking going forwardThis mindset is being woven into product design, client marketing, daily operations, and overall management strategiesThus, the bank will assess employees against this capability to nurture talent sufficient for the evolution of the bank in the future.

"The continuous evolution of AI technology poses challenges for banking professionals but also brings forth novel prospectsIn the face of transformation, it is vital for practitioners to maintain a calm demeanor, constantly learn, and adapt to changing times to find their place within society and enterprise," Li Xiusheng encouraged. "Technicians must align themselves with AI technologies to boost their capabilities, while frontline staff need not fear displacement, as technology’s lower entry barriers allow those without computer backgrounds to effectively use AI tools in constructing processes and applications, thereby demonstrating their value

Hence, as long as they embrace learning and keep pace with technical shifts, banking professionals can avoid obsolescence, adapting effectively to the evolving technological landscape."

From the standpoint of risk control operations, Wei noted that hands-on practice plays a pivotal role in mastering artificial intelligence softwareIn risk management, applying AI technologies demands not only a profound understanding of technical principles but also a robust grasp of models’ advantages, limitations, and risksTherefore, risk managers should develop deep technical competencies alongside a broad knowledge base.

Wang Jun echoed this sentiment, indicating that academia is actively working to foster professionals skilled in both AI and their specific fields. "We are optimizing our curriculum to include courses on data analysis, data mining, machine learning, deep learning, and multi-modal data to expose students to AI knowledge early in their academic pursuitsIncreased emphasis is placed on training projects and laboratory courses, encouraging students to engage in finance technology competitions to transition knowledge into practical skillsMoreover, we aspire to strengthen industry collaboration, such as through joint labs and expert lectures, to enhance students' understanding of industry demands and motivations while cultivating talent that aligns with these needs," he conveyed.

As for future trends in AI application within the banking sphere, Li Xiusheng anticipates that the advancement of AI and large model technologies heralds a substantial reshaping in commercial banksThis transformation will involve not just system upgrades but will significantly alter banking processes, product forms, decision-making mechanisms, personnel structures, and role definitions. "While the essence of financial risk management remains constant, the means of service delivery, product compositions, and operational frameworks will undergo a sea changeThis evolution may be incremental, with commercial banks undergoing a complete facelift within the next three to five years," he concluded.

Advertisements

Advertisements