ChatGPT is a general-purpose chatbot program released on November 30, 2022 by OpenAI, an American company that studies artificial intelligence. After its launch, ChatGPT reached 100 million active users in 2 months, making it the fastest growing consumer app in history, attracting widespread attention around the world and setting off a huge wave of technology in the field of artificial intelligence. Analyzing the relevant technical characteristics of ChatGPT and studying the advantages and existing problems of application in the field of market supervision are of great significance for doing a good job in smart supervision in the new era.
Overview of ChatGPT
As a natural language processing tool driven by artificial intelligence technology, ChatGPT can conduct conversations by understanding and learning human language, generate corresponding intelligent answers according to the user's text input and the context content of the chat, chat and communicate like humans, and can also complete various tasks such as writing code, designing copywriting, writing papers, machine translation, and replying to emails.
ChatGPT has amazing language understanding, generation and knowledge reasoning capabilities, can understand user intentions well, truly achieve multiple rounds of communication, and answer content is complete, focused, generalized, logical and organized. ChatGPT is a new generation of knowledge representation and invocation after databases and search engines.
After its launch, ChatGPT quickly formed a phenomenal application, showing a high level of human-computer interaction, has the characteristics of general artificial intelligence, and has a wide application potential in many industries.
Second, ChatGPT related technology and development
(1) ChatGPT related technologies.
Model architecture for Chat GPT. ChatGPT is implemented using the Transformer model architecture, a large-language model that uses attention mechanisms to improve the speed of model training. ChatGPT includes a multi-layer self-attention mechanism and a forward neural network, which can capture long-distance dependencies and has strong language understanding ability.
The learning mechanism of ChatGPT. ChatGPT is trained through supervised learning of a large-scale corpus, using massive text data on the Internet for pre-training, so as to learn rich language knowledge and rules. The advantage of large-scale corpus training is that it can utilize a large amount of text data, which improves the generalization ability, accuracy, and coverage of the model. In addition, large-scale corpus training can also improve the robustness of the model, so that the model can better cope with various complex natural language processing tasks.
Training methods for ChatGPT. ChatGPT uses a training method of "human feedback reinforcement learning" in the training process, which uses human feedback during AI training to minimize meaningless, distorted, or biased output. Thanks to this training method, ChatGPT is able to memorize the conversation message of the previous user during the conversation, i.e. contextual understanding, to answer certain hypothetical questions. ChatGPT enables continuous conversation, which greatly improves the user experience in conversational interaction mode.
(2) ChatGPT technology development.
ChatGPT uses large language models with billions of parameters to learn complex patterns in linguistic data. The first generation GPT-1 model launched by OpenAI in 2018 has 117 million parameters, the second generation GPT-2 model launched in 2019 has 1.5 billion parameters, and the GPT-3 model launched in 2020 has 175 billion parameters. In March 2022, OpenAI launched the InstructGPT model, which is a fine-tuned version of GPT-3. On November 30, 2022, OpenAI launched a new conversational general artificial intelligence tool, ChatGPT. ChatGPT shows very amazing language understanding, generation, knowledge reasoning ability, can well understand user intentions, achieve effective multi-round communication, and answer content is complete, focused, generalized, logical and organized.
On March 15, 2023, OpenAI released its latest artificial intelligence language model, GPT-4, a landmark technological advance. GPT-4 can not only generate text similar to human language, but also accept images and text as input, and output text. GPT-4 can process many types of information and communicate with humans in a more natural and fluid way. Compared to GPT-3.5, GPT-4 has a larger model size and more training data, which improves the quality and diversity of its generated text. GPT4 has strong abilities in language understanding, logical reasoning, mathematical calculations, etc. In addition to text input, the GPT-4 has the ability to process image input, allowing users to specify any visual or linguistic task. According to OpenAI, GPT-4 has demonstrated "human-level" performance in benchmarks in a variety of professional and academic fields. On March 17, Microsoft released Microsoft 365 Copilot, using the latest release of GPT-4 and introducing it into Microsoft 365 applications, including Word, Excel, PowerPoint, Outlook, etc., marking a new stage in the way GPT has brought people to interact with computer office software.
(3) Domestic ChatGPT-like large model products.
With the increasing attention of OpenAI's GPT series products in the world, domestic institutions committed to the research and development of artificial intelligence technology have also launched their own ChatGPT-like large-model technology products. On March 16, 2023, Baidu officially launched China's first generative AI product "Wen Xin Yiyan", which can support literary creation, copywriting, mathematical calculation, multimodal generation and other functions. On April 10, SenseTime released the "SenseNova" big model system, launching a variety of large models and capabilities such as natural language processing, content generation, automatic data annotation, and custom model training. At the Alibaba Cloud Summit on April 11, Alibaba launched the big language model "Tongyi Qianwen", which has functions such as multi-round dialogue, copywriting, logical reasoning, multimodal understanding, and multi-language support. On April 25, HUAWEI CLOUD released a series of ultra-large-scale pre-training models for Pangu, including the world's largest vision (CV) pre-training model with 3 billion parameters, and the world's largest Chinese language (NLP) pre-training model with 100 billion parameters and 40 TB of training data jointly developed with Loop Intelligence and Pengcheng Lab. On May 6, 2023, iFLYTEK officially released the iFLYTEK Xinghuo cognitive intelligence model, which has cross-domain knowledge and language understanding capabilities, and can understand and perform tasks based on natural dialogue. Continuous evolution from massive data and large-scale knowledge to achieve a closed loop from proposing, planning to solving problems. iFLYTEK has put the big model into application and launched solutions for education, office, automotive and other industries. In the future, more ChatGPT-like large-model products will be released in China, pushing the development and application of domestic artificial intelligence technology to a new stage.
Third, the application of ChatGPT technology in the field of market supervision
In view of ChatGPT's excellent ability in language understanding, generation and knowledge reasoning, if the large models, learning mechanisms and training methods supporting ChatGPT can be applied to the field of market supervision, it will greatly promote the service experience of market supervision departments and improve the efficiency of supervision.
(1) Market supervision of intelligent customer service applications to optimize service experience.
At present, some business hotlines of market supervision departments have a relatively large number of daily telephone consultation and complaint visits, and the introduction of ChatGPT related technologies to build an intelligent customer service system can better alleviate the service pressure of the complaint consultation service hotline, improve the efficiency of user complaint consultation and consultation, and improve the level of supervision services. Using the technical advantages of ChatGPT in natural language understanding and processing and knowledge reasoning, it automatically identifies the content of user complaints and consultations, intelligently understands the intentions and behaviors of user complaints and consultations, builds a business knowledge base related to market supervision, intelligently interacts and responds to the information of user complaints and consultations through voice or text, improves the efficiency of market supervision departments in handling user complaints and consultations, improves user satisfaction with complaint handling, and improves the level of supervision services. At the same time, data analysis is carried out on the records of user complaints and consultations, and hot spots of user complaints, difficulties in rights protection and areas where business consultation is concentrated, so that the market supervision department can timely grasp the "barometer" and "wind vane" of complaint and rights protection, and promote the continuous improvement of the business optimization and service capabilities of the market supervision department.
(2) Build a knowledge base for market supervision and strengthen wisdom empowerment.
Based on ChatGPT technology, build a knowledge base of market supervision laws and regulations, regulatory rules, business standards, data standards, technical standards, law enforcement cases, etc., to provide intelligent query and special knowledge services for regulatory law enforcement personnel on law enforcement rules, law enforcement basis, law enforcement processes, and law enforcement cases, standardize law enforcement by regulatory personnel, and improve their law enforcement capabilities. At the same time, for market supervision related business consultation and business handling of market entities and the public, we will continuously enrich the interaction methods and channels between market supervision departments and users, optimize user business handling experience, promote the intelligent display of information, improve the efficiency of user information inquiry, business consultation and business handling, provide more targeted, more accurate and more intelligent personalized services according to user preferences, and continuously improve user convenience and satisfaction with the wisdom of knowledge.
(3) Early warning of market supervision risks to improve regulatory efficiency.
In carrying out market supervision work, market supervision departments have generated and aggregated big data on market supervision including market entities' access, supervision, law enforcement, rights protection, business behavior, as well as the development and change of market environment status and related objects. Using ChatGPT's large-language model construction and pre-training technology, the market supervision big data is processed, analyzed and trained, which can more clearly and accurately present the panoramic portrait of market entities, continuously optimize the credit risk model and security risk model of market entities, and provide insight and early warning for the risk diffusion situation of macro-regional industries, meso regulatory matters and micro-subject behavior, from individual change to group explosiveness to systemic contagion, so as to achieve early detection, early reminder and early disposal of discovery. Improve the timeliness, accuracy and effectiveness of supervision.
(4) Innovate regulatory methods and promote regulatory modernization.
The use of ChatGPT-related technologies in market supervision work is bound to promote the innovation of regulatory tools and supervision methods, further tap the value of market supervision big data, promote the empowerment of market supervision technology, promote the optimization and upgrading of the functions of market supervision information application systems and platforms, promote the deep integration of market supervision business and information technology, promote the reshaping of market supervision business, rational reengineering of supervision processes, and support the whole process of market supervision work such as access, production, circulation, and consumption. Innovation in the whole chain enhances regulatory efficiency, optimizes regulatory costs, and promotes the modernization of market supervision mechanisms.
Fourth, the problems faced in the application of ChatGPT technology
ChatGPT is a new breakthrough in the field of artificial intelligence technology, as a new technology in the field of market supervision to be applied, but also faces some problems such as data security, technology maturity and resource investment.
(1) Data security issues.
Applications using ChatGPT technology require a large amount of data for training, and have the function of collecting, storing and using massive data. After the user asks a question on the input port, the data of the user's conversation with it is stored on the cloud computing platform where the application runs. In the human-computer interaction Q&A, the data shared by the questioner and ChatGPT is used for iterative training of future models, which may involve private information and sensitive data, and there is a security risk of data leakage. ChatGPT runs on OpenAI's cloud platform, and if OpenAI's ChatGPT is used directly, there is a security risk for cross-border data flow. Therefore, it is not recommended to directly use or call ChatGPT in market supervision applications, and ChatGPT-related technologies can be applied under the premise of ensuring data security.
(2) Man-machine coordination.
Before the emergence of ChatGPT, artificial intelligence applications were mainly specialized, and the emergence of ChatGPT was an important milestone for general-purpose artificial intelligence applications. However, ChatGPT can not completely replace people, can not be too superstitious to use ChatGPT technology by the machine to give the results, ChatGPT will also appear misjudgment, affecting the accuracy and effect of market supervision work. The application of market supervision must fully consider the judgment of market supervision staff based on their own knowledge and experience, and properly handle the problem of human-machine coordination.
(3) Technical barriers and maturity issues.
OpenAI does not open ChatGPT applications to Chinese mainland users, and domestic users can only have a limited experience through ChatGPT's mirror website. After ChatGPT, relevant domestic institutions have successively launched applications similar to ChatGPT, such as Baidu's "Wen Xin Yiyan", Ali's "Tongyi Qianqian", iFLYTEK's "iFLYTEK Spark Cognitive Big Model", SenseTime's "Daily New SenseNova" large model system, etc., the maturity of ChatGPT technology needs to be further tested. The current ChatGPT application mainly focuses on question-and-answer applications, where the user asks questions and the computer gives the answer results of the questions. Since the launch of ChatGPT, no successful cases related to government departments have been seen, and the application field of ChatGPT has yet to be further observed and paid attention to by readers.
(4) Resources and costs.
ChatGPT requires a very large amount of computing power to support its training and deployment. At present, ChatGPT applications require server support with large computing power, and the cost of these servers is unaffordable for ordinary users, even the deployment of billions of parameters in search engines requires a staggering amount of computing resources to run and train. As a result, most businesses cannot afford this high cost. For the general public, it is still necessary to wait for a lighter model or a more cost-effective computing platform.
Image
5. Suggestions on the application of ChatGPT technology in the field of market supervision
Although the application of ChatGPT-related technologies in the field of market supervision has yet to be tested by the development of technology and time, timely attention to the development of ChatGPT-related technologies and preparation and response to possible problems in advance are of great value for meeting the challenges brought by new technologies, promoting market supervision application innovation and in-depth smart supervision.
(1) Strengthen data security management.
To use ChatGPT-related technologies in the field of market supervision, it is necessary to strengthen data security management, establish strict data rights management and classified application systems to protect data security involving sensitive information and personal privacy. According to the sensitivity of market supervision data, the hierarchical division and authority management are carried out, and the generative artificial intelligence model system represented by ChatGPT does not have the right to obtain data information with high security level, and can open the data that can be disclosed to the artificial intelligence model system. The market supervision department is responsible for supervising the collection, storage, transmission and sharing of data to ensure the application security of data. In addition, a security assessment system should be established, and the application system using Chat GPT technology should be regularly reviewed for security, and vulnerabilities should be patched against the evaluation indicators to ensure the security of service data.
(2) Strengthen risk prevention and control.
In the face of the wave of rapid development of ChatGPT technology, the market supervision department must be clear that the application of ChatGPT technology is a booster to improve work efficiency and regulatory efficiency, and cannot completely replace labor, market supervision workers must continue to learn relevant technical knowledge to improve their own ability to better play the core role, not to be trapped by technology. Market supervision departments should fully understand the role of ChatGPT-related technologies in empowering market supervision, conduct a comprehensive review of the risks of ChatGPT-related technologies, strengthen risk prevention and control awareness, objectively and rationally handle and evaluate the role and impact of ChatGPT-related technology applications on market supervision, give full play to the technical advantages of human-machine integration, and need to establish emergency response plans for emergencies, and then deal with technical failures, contradictions and disputes in a timely manner. Ensure that the public interest is fully safeguarded.
(3) Strengthen innovation in the application of new technologies.
ChatGPT-related technologies still have a lot of room for improvement in complexity and diversity, and there are still certain defects in solving complex problems or understanding multiple hidden emotions, and the application of ChatGPT-related technologies in the field of government affairs still has a considerable distance. However, the success of ChatGPT also enlightens us that we must strengthen the application and innovation of new technologies in market supervision work, based on the technological accumulation and information resource advantages formed by market supervision departments in the construction of market supervision informatization, based on the institutional mechanisms, methods, means, tools, cases and experience advantages formed by market supervision business in regulatory law enforcement work, strengthen the application innovation of a new generation of information technology such as artificial intelligence and big data, promote application system optimization and technology upgrading, and promote supervision methods, regulatory tools, Innovate regulatory methods, further promote smart supervision, and continuously improve the efficiency of market supervision.
(4) Strengthen the construction of a contingent of qualified personnel.
The application of ChatGPT technology requires users to have certain technology and professional knowledge, establish a management system and operation mechanism that is compatible with market supervision work and technology application innovation in the new era, and can reflect the characteristics of work and professional characteristics, continuously strengthen the training and introduction of talents familiar with the development of a new generation of information technology and proficient in application, strengthen the training of ChatGPT technology among business personnel and information workers in market supervision departments, and promote the exchange of talents in business and technology. Cultivate and build professionals who are familiar with business and proficient in technology.
Author | Zheng Gang
Edit | Liu Qi
Audit | Zhou Shan