01
Overview of the application of large-scale model technology
From the birth of ChatGPT at the end of 2022 to the large-scale model boom in 2023, in the great wave of technology, large-scale model technology is like a bright star, rapidly rising and leading an unprecedented technological revolution. Large models are like engines that drive innovation, pushing the boundaries of science and technology.
The advancement of large models will have a disruptive impact on many fields. For individuals, from text creation to daily office, large models are empowering various scenarios with more accurate and efficient services. For enterprise-level applications, large models are playing an immeasurable role in marketing, customer service, R&D and other business fields, accelerating the digital transformation and intelligent upgrading of the industry.
The real value of technology is not just in theory, but needs to be exerted through practical application scenarios. Based on the capabilities and characteristics of large models, industries and scenarios with high digital level, perfect data foundation, and complex knowledge system are often the first to implement large model technology and give full play to its application value.
After more than one year of tracking the application of large model technology in the Dune community, as of mid-January 2024, the financial industry (35%) accounts for the largest proportion of large model application cases, followed by manufacturing (13%), medical care (10%), and government and public services (8%). Among them, in the financial sector, the banking industry (17%) accounted for the highest proportion.
From the perspective of application scenarios, scenarios such as knowledge management (22%), data analysis (13%), content creation (12%), and dialogue interaction (9%) are naturally closely integrated with large model technology, and are also the main directions explored by enterprises.
In 2024, with the continuous innovation and iteration of large model technology, enterprises will focus on the full implementation of large models and deeply explore the application value of large model technology in various fields.
02
2024 Top 30 Pioneer Cases of China's Large Model
Through case studies, expert interviews, desk research, etc., 212 large model landing cases were evaluated from the four dimensions of innovation, value, practicability and demonstration, and the most pioneering TOP30 cases were selected to provide reference for the application of enterprise large models.
The results of the "2024 China Large Model Pioneer Cases TOP30" are as follows (in no particular order, in alphabetical order):
03
Introduction of selected cases
▎Case 1: Bo Xiaozhi RAG Medical Data Retrieval Enhancement
Case Parties/Vendors: Boehringer Ingelheim/Corgi Data
Applications: Medical
Case Details:
Boehringer Ingelheim offers a chatbot for medical representatives to ask medical and platform questions. On this basis, based on the capabilities of large models and knowledge graphs, the automation of medical content production can be quickly improved, the timeliness and compliance problems in the process of medical inquiries can be solved, and the efficiency of academic promotion can be improved.
Reasons for selection:
• Traditional intelligent Q&A services are manually generated by FAQ and then submitted for compliance review, which takes a long time and makes it difficult to ensure that the content is updated in a timely manner.
• Based on the large model and knowledge graph, it can better understand user questions, calculate the similarity between the problem and the corresponding knowledge base by retrieving keywords and vectors, retrieve the problem-related knowledge base, give objective scientific basis, provide users with more accurate answers, and improve the user question answering rate and satisfaction.
▎Case 2: Headhunting management and decision-making insights based on conversational AI indicator analysis
Case Party/Vendor: Octopus Network/Kyligence
Application: Talent service
Case Details:
By adopting Kyligence Zen, an intelligent one-stop indicator analysis platform, and Kyligence Copilot, an AI digital intelligence assistant, Hunttech simplifies the data analysis and management process, provides detailed headhunting efficiency indicators and funnel analysis, and enables the company's headhunters and operation personnel to improve their work efficiency by themselves, which not only optimizes the company's overall management of the headhunting group, but also improves the quality and efficiency of the entire headhunting service.
Reasons for selection:
• The innovation of Kyligence Zen's intelligent one-stop indicator platform lies in its integration of AI data analysis technology and large model capabilities. The platform leverages AI-enhanced capabilities such as an intelligent metric engine and an AI digital assistant to provide self-service analytics to meet the needs of enterprises for data-driven decision-making.
• Through timely and effective intelligent data index analysis and management, this project enables headhunters to better understand market demand, provide more personalized and accurate career advice for job seekers, improve recruitment efficiency and improve job search experience.
▎Case 3: Dongfanghong Smart Maverick
Case party/supplier: Orient Asset Management
Applications: Finance
Case Details:
Dongfanghong Smart Maverick is a large-model-based AI investment research assistant platform jointly developed by Orient Securities Asset Management and Orient Securities, which provides users with financial assistance services through questions and answers, and carries out customized application research and development in roadshow speed reading, intelligent risk control, bond inquiry trading, question system, etc., so that investment researchers and fund managers can focus more on core tasks, quickly obtain information and make accurate decisions.
Reasons for selection:
• Fund managers and researchers are faced with massive financial information every day, and it is difficult to obtain valuable information quickly and accurately.
• Dongfanghong Smart Mavericks will help improve the ability of investment managers to manage large-scale funds, expand the company's asset management boundaries, and align with the goal of empowering and increasing the efficiency of every employee with artificial intelligence technology.
▎Case 4: Intelligent verification project based on financial model
Case party/supplier: GF Securities/Wenyin Internet
Applications: Finance
Case Details:
In this case, we use financial model technology to build financial model application infrastructure services, including corpus engineering, prompt engineering, and quality control engineering. At the same time, it builds three application scenarios of the financial model, intelligent document extraction, intelligent generation, and intelligent verification, and creates an integrated solution for investment bank document extraction, generation, and verification based on the financial model.
Reasons for selection:
• Large model technology empowers investment banking business document generation and standardizes business document output process, which is not only conducive to improving business efficiency, but also improving document quality and reducing business risks.
• The precipitation and maintenance of data is an important guarantee to empower investment banking business, but a lot of data is scattered in various manuscript documents of investment banks, which cannot be effectively extracted and summarized. The entity extraction capability of the large model fills the gap in this regard, and realizes automatic data collection and structured storage.
▎Case 5: Equipment operation and inspection knowledge assistant based on power cognitive model
Case party/supplier: State Grid/Baidu
Applications: Energy
Case Details:
Based on the artificial intelligence research and development experience of massive high-quality power samples and power operation and inspection scenarios of the State Grid Smart Grid Research Institute, as well as Baidu Wenxin series large models and leading deep learning, knowledge graph enhancement and other technologies, jointly develop equipment operation and inspection knowledge assistants based on power cognitive large models, empower knowledge services and calculation engines, and realize convenient knowledge query, rapid learning and operation assistance in equipment operation and inspection links.
Reasons for selection:
• The equipment operation and inspection knowledge assistant based on the power cognitive model supports the online management of equipment data of operation and inspection caliber, accurate retrieval of technical standards and general systems, intelligent Q&A of technical standards, mobile applications of knowledge base, and various operation and maintenance scenarios, promotes the refinement, automation and intelligent development of power grid operation, improves the intelligent level of power equipment operation and inspection, and provides reference and inspiration for the digital transformation of the power industry.
▎Case 6: Intelligent software development and application practice based on aiXcoder code model
Case party/supplier: Guojin Securities/Silicon Technology
Applications: Finance
Case Details:
With the aiXcoder code model as the core engine, combined with the 30 years of software asset precipitation in the financial industry and the ecological integration of artificial intelligence, the intelligent development solution of Sinolink Securities has jointly built a powerful engineering application framework of the code model. Since its promotion, the intelligent development solution has increased development efficiency by an average of 30% and unit test coverage by 20%.
Reasons for selection:
• The aiXcoder code model combines the domain-specific datasets and domain expert knowledge of IFC in the securities industry, so that the code model can better understand the problems and contexts in the software engineering securities field, and make the generated code syntax more standardized and more in line with the actual project logic;
• The aiXcoder code model has a wide range of applications, which can be applied to coding, testing, search and other links of software development in the securities industry, and can realize code automation, design automation and system automation for enterprises.
▎案例7：「AI组卷判卷」Agent实践
Case party/supplier: Propaganda and Education Center of Hubei Provincial Market Management and Supervision Bureau/Lanma Technology
Applications: Education
Case Details:
Based on AskXBOT, an enterprise-level AI agent platform based on AskXBOT, AskXBOT, Landma cooperated with Hubei Market Supervision and Education Center to develop the "AI Roll Judgment Agent".
The "AI Marking Paper" Agent quickly assembles a complete test paper based on the input documents, question type templates, specified quantities and other information, while the "AI Marking Paper" Agent provides teachers with intelligent "Correction Test Papers" to empower training and exams.
Reasons for selection:
• The "Folder Judgment Agent" jointly developed by Landma Technology and Hubei Market Supervision and Education Center is the first in China, and can be widely replicated and applied to the education industry.
• The "Agent of Marking Papers" has the opportunity to achieve "inclusive education", so that the educational resources that can only be enjoyed by a few people can be popularized to more people, so that more students can benefit from personalized teaching, and make up for the problems of manpower shortage and student differences that are common in traditional teaching.
▎Case 8: Haimeng Holdings' digital employee application practice based on the supply chain model
Case party/supplier: Haimeng Holdings/Yifeng Technology
Applications: Supply Chain
Case Details:
Based on the supply chain model, the digital employee hyperautomation platform tailors a virtual digital employee expert team for Haimeng Holdings through natural language dialogue, including senior supply chain freight managers, logistics visual tracking managers, supply chain newcomer growers, industry case experts, industry translation masters, administrative Q&A assistants, etc., to assist white-collar employees to complete various digital tasks and allow employees to focus on high-value work such as creativity and decision-making.
Reasons for selection:
• Based on the large model of the supply chain, the hyper-automation platform of digital employees provides customers with out-of-the-box Agent digital employees and accurate industry answers, subverting the existing interaction paradigm of SaaS products and realizing intelligent changes from human-computer graphic interaction to natural language interaction;
• By providing personalized and intelligent supply chain services, the digital employee hyperautomation platform can help enterprises better understand customer needs, help enterprises achieve end-to-end global hyperautomation from marketing to fulfillment to financial settlement, empower enterprise business model reform and efficiency improvement, and promote enterprise productivity leap.
▎Case 9: Geely Xingrui AI model
Case party/supplier: Geely Automobile
Applications: Manufacturing
Case Details:
Geely Xingrui AI model is a full-stack self-developed full-scene model of the automotive industry, which deeply focuses on the vertical field of automobiles, has the most complete professional knowledge reserve in the automotive industry, opens up the whole link of automobile R&D and manufacturing, comprehensively improves the efficiency of automobile R&D and manufacturing, and enhances Geely Automobile's competitiveness and influence in the market.
Reasons for selection:
• Geely Xingrui AI model has been self-developed at multiple levels such as chips, algorithms, platforms, and applications, creating a complete AI ecosystem and realizing Geely Automobile's autonomous control and continuous optimization of AI technology.
• Geely Automobile has released a number of AI-native applications based on the Xingrui AI model, such as Wow wallpapers and AI picture books, which are installed on a variety of Geely smart electric products, providing consumers with a more intelligent, personalized and humanized travel experience.
▎Case 10: Application of AI large model for grid modernization governance scenarios
Case party/supplier: Jinan Central Public Security Bureau / Baiying Technology
Applications: Government
Case Details:
Jinan City Public Security "Weimin Small and Micro" police WeChat work platform uses WeChat client and artificial intelligence to provide the masses with simple, easy-to-use, fast and authoritative police-public interaction channels, with five major functions, such as direct connection between police and people, AI intelligent reply, information collection, flat management of elements, and WeChat personalized publicity and defense, to build an integrated application pattern of public security online government services.
Reasons for selection:
• Based on the large model capability, this case studies the diversity policy document and processes it into an in-situ knowledge base. According to the content of communication with residents/enterprises, quickly determine and accurately adapt to the latest policies, provide all-weather intelligent government services, and help the public quickly understand government affairs information, query policies, apply for certificates, etc.;
• This case extends its tentacles to the "last mile" of the grassroots through online normalization and grid implementation of public management and grassroots governance work such as convenient services, dispute reporting, and anti-fraud publicity and prevention, so as to achieve a non-inductive link and in-depth reach with residents.
▎Case 11: Danqing large model helps the art production of Thunderfire game
Case party/supplier: Thunderfire Games/NetEase Fuxi
Applications: Gaming
Case Details:
Based on the art assets accumulated by Thunderfire Games for many years, NetEase Fuxi has developed a number of game art models for different game project styles, and at the same time built an AI drawing platform that can be used online, which is integrated into the work pipeline of art asset production, bringing a new set of production methods to game art and greatly improving the overall production efficiency.
Reasons for selection:
• This case focuses on the full integration of AI graphics with traditional art-made game art assets, and pays more attention to "human-machine collaboration" rather than directly using AI-generated results, which not only ensures the high quality of Thunderfire game art assets, but also achieves the goal of reducing costs and increasing efficiency.
• Through the experience of several key service cases, Fuxi has formed a set of service standards for model training and art pipeline construction, which can quickly migrate Fuxi's AI mapping technology to various projects.
▎Case 12: Noah Fortune's intelligent knowledge base and knowledge assistant application based on domain large model
Case party/supplier: Noah Fortune / Zhongguancun Kejin
Applications: Finance
Case Details:
Noah Fortune has built an intelligent knowledge base based on artificial intelligence technologies such as large models in the financial field and intelligent customer service developed by Zhongguancun Kejin, and provides employees and users with intelligent Q&A query functions based on enterprise knowledge documents in the form of knowledge assistants, which greatly improves the accuracy of Q&A intent recognition and reply accuracy of the customer service system, and helps enterprises achieve more intelligent and lower-cost knowledge Q&A services.
Reasons for selection:
• In this case, through the model fine-tuning technology developed by Zhongguancun Kejin, single-card inference and single-week iteration are realized, and the model fine-tuning is completed quickly and cost-effectively, and the application of large models is implemented.
• Combined with the intelligent knowledge base of large model application, it has the capabilities of multi-modal document analysis, automatic extraction of QA Q&A pairs, and automatic labeling of knowledge content, realizing the management application scenarios of documents and Q&A-based QA, and promoting the application paradigm of enterprise knowledge management.
▎Case 13: The application of the DFM-2 model in the smart office scenario
Case party/supplier: Spire
Applications: Office
Case Details:
In the office field, the DFM-2 model is deeply integrated with office hardware products and office software products to bring users smarter office products and promote the improvement of work efficiency. With the support of the large model, the office software supports functions such as AI summary, AI to-do, discourse regularization, and one-click drafting, and the office hardware is upgraded from one-way noise reduction to local + remote two-way noise reduction with the support of the large model, and supports AI tracking mode.
Reasons for selection:
• Combining smart office products with the capabilities of the DFM-2 model, Spires has upgraded and evolved product functions, and launched smarter and more user-friendly functions such as AI summarization, one-click drafting, AI noise reduction, and AI tracking, which is more intelligent than traditional office software.
• The smart office products based on large models have greatly improved the work efficiency of users, from "1 hour of meeting and 3 hours of sorting out meeting minutes" to "1 hour of meeting and 5 minutes of manuscript", the user's office efficiency has been significantly improved, and the AI noise reduction and AI tracking functions ensure that users can hear and see remote calls clearly, and improve the efficiency of remote communication.
▎Case 14: Intelligent mine project based on Pangu large model
Case party/supplier: Shandong Energy Group/Huawei
Applications: Energy
Case Details:
Based on the Pangu mine model, Shandong Energy Group has built a "1+4+N" system, based on an AI development platform and four large model capabilities, developed N high-value application scenarios, and applied a set of reusable algorithm models to different business scenarios in a streamlined manner, reducing the consumption of expert intervention and human tuning, reducing the development threshold and time cycle of artificial intelligence, and achieving rapid replication and promotion.
Reasons for selection:
• Large models have the characteristics of large-scale, high-precision and high generalization, which can effectively solve the problems of poor generalization and inability of traditional artificial intelligence solutions to support large-scale replication.
• At present, Shandong Energy Group has fully applied the Pangu mine model to 9 major business systems and more than 40 scenarios, including mining, excavation, machinery, transportation, communication and washing, and has been used on a large scale in 8 mines across the country.
▎Case 15: An online Q&A platform for automobiles based on AI large models
Case party/supplier: SAIC Passenger Vehicle/Dingjie Software
Applications: Manufacturing
Case Details:
Based on Dingjie Athena ChatFile, SAIC Motor Passenger Vehicle has built an online Q&A platform based on AI large model, which is mainly aimed at unstructured documents such as PDF, WORD, PPT, TXT, EXCEL, etc., to realize the interaction between natural language and knowledge, obtain and use corporate knowledge accurately, compliantly and safely, and improve the work efficiency and learning ability of employees, which is applied in multiple scenarios such as SAIC's R&D, internal operation and maintenance, and after-sales service.
Reasons for selection:
• In this case, the built-in pre-trained small model of the knowledge middle platform combined with the GPT large model realizes knowledge self-learning through various methods such as industry dictionaries and problem self-analysis, so that the knowledge application becomes smarter and smarter, creates a knowledge engineering system for enterprises, and helps enterprises transform into digital intelligence.
• Dingjie Athena ChatFile supports multiple rounds of interaction, cross-document interaction, multi-language interaction, multi-modal interaction and other ways to help users unlock the rich knowledge accumulated by enterprises for a long time, provide accurate answers, and not over-generalize.
▎Case 16: The 360 intelligent brain model empowers the digital intelligence upgrade of the financial industry
Case party/supplier: Tianjin Jincheng Bank/360 Group
Applications: Finance
Case details: Focusing on the scenarios and office automation needs of Tianjin Jincheng Bank, 360 Group gives full play to the core capabilities of the self-developed general large model 360 intelligent brain, on the one hand, it creates an enterprise knowledge base to create a standardized dialogue model, and employees can quickly obtain information in the form of Q&A dialogues, and on the other hand, it creates digital employees and virtual analysts to provide intelligent assistance for enterprise office and improve work efficiency.
Reasons for selection:
• With the support of the 360 intelligent brain model, Tianjin Jincheng Bank has achieved an average of 1,000 Q&A sessions, 100 output documents, and 20 meetings per day, effectively improving office efficiency.
• The 360 Intelligent Brain Model builds technical capabilities covering all financial scenarios, including knowledge quizzes, multi-round dialogues, reading comprehension, logical reasoning, intelligent recommendation, intelligent search, analysis and early warning, text classification, generation and creation, and code capabilities, etc., supporting financial institutions to carry out intelligent innovative applications and development in combination with actual business, and improving productivity and competitiveness.
▎Case 17: Taikang Group's human resources shared service center scenario large model application practice
Case party/supplier: Taikang Insurance Group/Wofone Technology
Applications: Finance
Case Details:
Based on the "large model + AI knowledge platform" solution provided by Wofone Technology, Taikang Insurance Group has built a unified service platform for human resources intelligence, providing employees with convenient and efficient human resources knowledge services through intelligent knowledge search, unified management of knowledge base and employee data portraits, empowering the construction of HR shared service centers, and comprehensively improving HR work efficiency and employee problem solving rate.
Reasons for selection:
• The application of large language model technology in the whole life cycle management of knowledge production, management, application and innovation can reduce the workload of knowledge base operators and improve the efficiency of knowledge base management and maintenance.
• The AI knowledge center can be connected with a variety of products such as third-party knowledge bases, internal knowledge bases, intelligent robots, online customer service systems, and marketing systems, enabling enterprises to share services (finance, IT, HR, etc.), customer service, on-site service management, content marketing and other business scenarios.
▎Case 18: Tasly Chinese medicine model
Case party/supplier: Tasly Group/Huawei
Applications: Medical
Case Details:
With the advantages accumulated in the field of traditional Chinese medicine, Tasly and Huawei Pangu Model have cooperated to build a large model of traditional Chinese medicine, and create scenario applications such as R&D efficiency, traditional Chinese medicine diagnosis, natural medicine screening, intelligent marketing, and intelligent diagnosis and treatment, so as to solve the R&D pain points of traditional Chinese medicine such as sparse and discrete information, promote the inheritance of the knowledge base of traditional Chinese medicine, and accelerate the whole process of R&D in the innovative Chinese medicine industry.
Reasons for selection:
• Tasly TCM model is the industry's first TCM vertical model, and is an industry model for accelerating the innovation and transformation of TCM with the help of the model;
• The TCM model will greatly improve the intelligent ability of Tasly's TCM R&D efficiency and accuracy, quickly realize the intelligent R&D and upgrading of TCM, bring more choices and value to the market, and help high-quality innovation and development in the field of TCM R&D.
▎Case 19: The application of the real TARS model in Tianyi Digital
Case party/supplier: Tianyi Digital Technology/Real Intelligence
Applications: Finance
Case Details:
With the help of historical data accumulation and the semantic understanding ability of large models, the material review of pre-loan review is realized through human-computer interaction. The overall interaction mode of pre-loan review has achieved revolutionary changes, and the overall efficiency of the review system and platform has been increased by more than 300%.
Reasons for selection:
• In this case, the large model technology is applied to the field of intelligent risk control, especially in the process optimization and interactive experience improvement of pre-loan review, to reshape the existing review process, achieve more intelligent and automated risk control, and improve the work efficiency of reviewers.
• In this case, the knowledge in the field of risk control is learned into the large model, and the model is trained through instruction fine-tuning, artificial feedback reinforcement learning, etc., so that the model output is more in line with expectations, so that the knowledge accumulated by domain experts can be carried in the form of a large model.
▎Case 20: Chang'e Project
Case party/supplier: Wuhan Institute of Artificial Intelligence
Applications: Medical
Case details: Wuhan Institute of Artificial Intelligence and Jointown jointly built an intelligent identification system for orthopedic surgical medical devices based on the 100 billion parameter "Zidong Taichu" full-modal large model, which realized the rapid identification, convenient circulation and information tracking of surgical instruments and consumables, and realized the revolutionary leap of the whole process from a few hours to one minute, significantly reducing the industry threshold and fully improving the operation efficiency.
Reasons for selection:
• This case realizes the intelligent identification and full tracking of complex orthopedic implants and tools, as well as the systematic control of the whole process of preoperative, distribution, intraoperative and postoperative, leading a new paradigm of digital transformation in the orthopedic surgery industry.
• This case solves the problem of identifying complex items in and out of the warehouse, not only in the field of orthopedics, but also in cardiac surgery, general surgery, sterilization supply room, operating room and other surgical environments that require frequent use of various types of in-hospital instruments and foreign surgical instruments.
▎Case 21: Insurance personal assistant based on the zero-rhino causal model
Case party/supplier: Xinghuo Bao/Lingxi Technology
Applications: Finance
Case Details:
Based on the causal model, Lingxi Technology creates a user-oriented insurance personal assistant for Xinghuo Insurance, and reaches customers through multiple channels such as enterprise micro, telephone, and small program. The insurance personal assistant built based on Lingxi Technology has a sales efficiency of up to 5 times that of the traditional manual conversion of Xinghuo Bao, and the cost of sales per order is reduced by 70%, making the short-term insurance business of Xinghuo Bao achieve positive growth.
Reasons for selection:
• The Zero Rhino Causal Model focuses on solving problems in the insurance field, and is a model that understands insurance better, aiming to promote the innovation and development of the insurance industry.
• Compared with the traditional pure manual transformation of Xinghuo Insurance, the insurance personal assistant created by Zerorhino Technology has a higher level of logical reasoning and decision-making ability, realizes the accurate matching of user needs and insurance products, and effectively realizes the growth of insurance business.
▎Case 22: Yuanbei Beibei & Huazang large model smart crib project
Case party/supplier: Yuanbeibeibei/Xiaoi robot
Applications: Medical
Case Details:
Through the training of various maternal and infant vertical professional knowledge and databases, as well as the deeper development of comprehensive maternal and infant health monitoring, abnormal situation prediction and alarm, personalized suggestions, report interpretation, etc., the Huazang general model is used as the starting point to open up the intelligent maternal and infant application scenarios and continuously iterate maternal and infant health management services.
Reasons for selection:
• Yuanbeibei artificial intelligence crib explores a new generation of maternal and infant care mode, providing more intelligent and personalized services for newborn babies and new parents. With "active health" as the core of management, we pay attention to the health needs of the new generation of family parenting, and open a new era of digital and intelligent parenting;
• Based on the general model of Xiaoi Huazang, Yuanbeibei has developed an AI parenting consultant - Beibei Know, which can provide detailed professional maternal and infant knowledge answers, become a digital intelligent assistant that understands parents better, and fully covers the professionalism of the vertical field of parenting.
▎Case 23: Sunshine Zhengyan GPT large model open platform
Case Party/Supplier: Sunshine Insurance
Applications: Finance
Case Details:
Sunshine Insurance Group's "Zhengyan Large Model Open Platform" adopts an internal and external hybrid large model architecture to create a comprehensive large model platform. The platform supports the construction of three types of robots: sales, management and service, and empowers the intelligent development of the business side, including intelligent sales, intelligent underwriting, intelligent claims settlement, intelligent services, etc., bringing comprehensive end-to-end empowerment to the insurance business value chain.
Reasons for selection:
• In the future, the company will reconstruct its core business scenarios such as sales, service, underwriting, and claims, and Sunshine Insurance Group will take the lead in realizing the in-depth application of large models in the insurance field, ushering in opportunities for the company to overtake in corners.
• The GPT large model open platform is the foundation of a new generation of artificial intelligence technology, which can not only meet the needs of key business scenarios, but also have the ability to multi-modal processing of complex scenarios.
▎Case 24: Intelligent OTC discovery platform
Case party/supplier: Galaxy Securities/Baidu
Applications: Finance
Case Details:
Galaxy Securities has built an intelligent OTC transaction discovery platform based on Baidu Wenxin's large model, which realizes end-to-end OTC derivatives intelligent trading services by automating the whole process of transaction inquiry and quotation business, forming a closed loop from product understanding, demand understanding and conversational trading mode to new transaction transformation, realizing intelligent OTC derivatives business operations, improving customer service efficiency, and optimizing institutional customer satisfaction.
Reasons for selection:
• The intelligent OTC discovery platform has excellent generalization ability of large models, and can achieve good model results through a small number of sample training, and has supported automatic quotation service for options trading such as vanilla and snowball, so as to realize automatic quotation inquiry for institutional customers;
• The intelligent OTC trading platform can analyze the intent of trading rules inquiry, and provide intelligent document Q&A service for trading rules based on the internal knowledge documents of Galaxy Securities, opening up the closed loop of trading rule Q&A from knowledge precipitation to application.
▎Case 25: Knowledge base Q&A large model project
Case party/supplier: China Telecom Group/Datatop Technology
Applications: Communications
Case Details:
China Telecom Group uses technologies such as industry model, intelligent document parsing, intelligent retrieval, and Q&A to conduct interactive dialogues on massive knowledge bases, helping users obtain knowledge, summarize documents, analyze and attribute data, and automatically generate research reports.
Reasons for selection:
• In this case, a variety of models are used to parse multiple types of documents and the multimodal information in the documents, including OCR models, layout recognition models, paragraph segmentation models, etc., taking into account the diversity, accuracy and efficiency of document parsing, and improving the overall effect of document parsing.
• Compared with the document intelligent Q&A system based on traditional NLP technology and search technology, the document intelligent Q&A system based on large model technology can more accurately understand the user's intent, conduct logical reasoning, and summarize the answers from paragraphs, which is more flexible and accurate.
• The knowledge base Q&A model product jointly developed by China Telecom and Datatop Technology effectively solves the illusion problem and knowledge privacy problem, and meets the diverse needs of government and enterprise customers.
▎Case 26: Construction of intelligent R&D system based on large model
Case party/supplier: Industrial and Commercial Bank of China
Applications: Finance
Case Details:
ICBC leverages its all-round capabilities in the fields of large-scale model code generation, code recognition and detection, and code-to-natural language to build an intelligent R&D system based on large models, covering the entire R&D life cycle of requirements, design, coding, testing, and release.
Reasons for selection:
• With the support of large model technology, mature applications have gradually emerged in the direction of code generation, code completion, unit test generation, and code defect detection, bringing technological changes to the improvement of software R&D efficiency, and the combination of large models and R&D systems has become an important trend in future intelligent R&D.
• At present, ICBC has formed the capabilities of code deduction and prediction, automatic code generation, code retrieval and reuse, etc., and integrated them into the development center in the form of IDE plug-ins to effectively improve R&D efficiency.
▎Case 27: Industrial and Commercial Bank of China Digital Employee Solution
Case party/supplier: Industrial and Commercial Bank of China / Mobvoi
Applications: Finance
Case Details:
Mobvoi uses the digital human production technology based on the "Sequence Monkey" large model to reproduce ICBC's real wealth advisors on a 1:1 basis, customize private bank digital wealth advisors for different sub-branches, and enrich the professional knowledge of digital wealth advisors by putting relevant policies and speech data into the large model for training, creating a digital wealth advisor who "can't ask questions".
Reasons for selection:
• With the help of large models, digital wealth advisors can continuously update the knowledge base and analyze business data, open up the closed loop of services, and promote the intelligent management and operation of the banking system.
• Digital wealth advisors have a wide range of application scenarios, and in the future, they can promote financial services on social platforms through short videos, live broadcasts, etc., and can also be used as virtual IP to endorse the industry.
▎Case 28: Agricultural Bank of China ChatABC model
Case party/supplier: Agricultural Bank of China
Applications: Finance
Case Details:
Based on open source technology, Agricultural Bank of China conducts end-to-end whole-process independent research and development, launches ChatABC, a large financial AI model, and forms nine thematic scenario domains based on the idea of "large model +" scenario ecological construction, including intelligent customer service, intelligent marketing, intelligent risk control, intelligent decision-making, intelligent R&D, smart office, network security, intelligent operation and maintenance, smart agriculture, and rural areas, and 30+ scenarios are simultaneously constructed.
Reasons for selection:
• Agricultural Bank of China's ChatABC model focuses on the knowledge understanding, content generation and security Q&A capabilities of large models in the financial field, which is an important exploration of AI large models in the financial industry.
• At present, the ChatABC model of the Agricultural Bank of China has played a valuable role in the fields of intelligent research and development, network security, smart agriculture, and smart credit.
▎Case 29: Auditing digital workforce based on large models
Case Parties/Suppliers: CPIC
Applications: Finance
Case Details:
CPIC uses digital labor as a breakthrough to carry out the application of large models in the insurance field, build a large model of the foundation of digital employees, realize the systematic modeling of business personnel, and align digital employees with real employees in terms of thoughts, actions and professional abilities, so as to provide more equivalent labor force.
Reasons for selection:
• Traditional AI mainly solves problems in specific business processes, and large models have the ability to understand, think, and solve problems, and can complete unspecific tasks through actions, bringing a new paradigm change to insurance companies.
• CPIC has built a number of audit digital employees, such as audit inspection, document quality inspection, and consultation Q&A, to help improve the efficiency and accuracy of audit work and fill the shortage of audit manpower through the working method of digital employees and real employees.
▎Case 30: Customer service problem attribution analysis and service quality analysis system based on AI large model
Case Party/Supplier: Ziru Ziroom/Cycle Intelligence
Application: Internet
Case Details:
Ziru ziroom uses advanced AI large model technology to conduct in-depth mining and intelligent analysis of customer service interaction data, accurately identify the root cause of customer problems, and realize the automation of problem attribution. At the same time, by analyzing the emotional state of customers at the beginning and end of communication, the quality and efficiency of customer service response can be effectively evaluated, and data support and strategy optimization suggestions are provided for improving the overall service quality.
Reasons for selection:
• With the help of large model technology for the analysis of communication data, it can realize the rapid summary and analysis of customer problem demands, the summary of customer service solutions, the analysis of customer emotional state, the analysis of the root cause of problems, and the summary analysis of customer voices.
• This case uses a large model with 100 billion parameters of cyclic intelligence, supports an ultra-long context window length of up to 200,000 words, and achieves a lower hallucination rate through specially designed instruction engineering technology, greatly improving the usability of the large model in enterprise-level application scenarios.
04
Core evaluation dimensions
This selection qualitatively and quantitatively evaluates the cases from four dimensions: innovation, value, practicability, and demonstration:
• Innovativeness: The case has a unique solution, which demonstrates the excellence of technological innovation, leads the development of the industry, and injects new innovation momentum into the market;
• Value: The case promotes the realization of the company's business goals and provides clear business value, or the case has social value, actively solves important social problems, and has a positive impact on society;
• Practicability: The implementation of this case has brought remarkable results, performed well in practical applications, and created practical value for enterprises and users;
• Exemplary: This case has reference and reference significance for the development and capacity building of large-scale model technology application in the same industry or other industries.