Dialogue with Liu Cong of HKUST Xunfei: How to win the "endurance race" in the "straight charge" of the large model?

Image source: Generated by Unbounded AI

Recently, hundreds of domestic AI applications have been removed from the App Store, which has aroused widespread concern in the industry. With more and more large-scale model companies and AI applications emerging, how can the domestic large-scale model industry develop healthily and sustainably under the “flurry of models”? It has become a problem that more and more people pay attention to and think about.

Recently, Liu Cong, Vice President of HKUST Xunfei and Dean of the Research Institute, pointed out in an interview with Sina Technology that at present, the development of my country’s artificial intelligence industry has made breakthroughs, but it still faces the problem of lacking a unified and controllable generation framework and content. Self-inspection standards, “With the emergence of ethical and security issues brought about by the implementation of large-scale models, there is an urgent need to strengthen the standardization of large-scale models in China, and to ensure that the quality and performance of large-scale models reach the expected level by formulating unified national standards to protect users. Data security and privacy protection to support the healthy and sustainable development of the large-scale model industry.”

In Liu Cong’s view, the domestic large-scale model competition is a protracted battle. After making staged results, industrial development and scientific research innovation must go hand in hand, and the paradigms of each other will continue to iterate and promote each other, so as to provide a steady stream for the development of my country’s independent large-scale models. Continuously inject long-distance running power.

The healthy development of the large model industry urgently needs standardization

A few days ago, the list of leaders of the large-scale model special group of the National Artificial Intelligence Standardization Overall Group was announced. Liu Cong, vice president of HKUST Xunfei and director of the research institute, served as the joint leader of the special group on behalf of the company. According to Liu Cong, since its establishment, HKUST Xunfei has attached great importance to standardization work. From the group level, a standardization management committee with the chairman as the director has been established. Members of the company’s leadership team and heads of various departments serve as members of the standardization committee. Layout, coordinate and manage various standardization work of the group.

“As early as 2003, HKUST Xunfei served as the leader unit of the Chinese Voice Interaction Standard Working Group of the former Ministry of Information Industry; since 2013, Chairman Liu Qingfeng has served as the deputy director of the User Interface Sub-Technical Committee of the National Information Technology Standardization Committee. The company Served as the team leader unit of the Voice Interaction Standard Working Group; from 2018 to the present, Senior Vice President Hu Guoping has served as the deputy leader of the National Artificial Intelligence Standardization Overall Group; from 2022 to the present, I have served as the Knowledge Graph of the Artificial Intelligence Subcommittee of the National Information Technology Standardization Committee Deputy head of the working group.” Liu Cong said.

According to him, up to now, iFLYTEK has led and participated in the formulation of more than 70 domestic and foreign standards in the fields of intelligent voice and artificial intelligence, among which 2 international standards, 29 national standards, and 8 industry standards have been released and implemented. There are 9 group standards, covering speech recognition, speech synthesis and smart home, smart customer service, mobile terminals, vehicle terminals, smart TVs, machine translation and other fields.

Today, with the re-election as the co-chairman of the large model project group of the National Artificial Intelligence Standardization General Group, HKUST Xunfei has added a new identity to participate in the formulation of national artificial intelligence standards.

In Liu Cong’s view, standards are the technical support for economic activities and social development, and are an important aspect of the country’s basic system. Standardization plays a fundamental and leading role in the national governance system and modernization governance capabilities. At present, the development of my country’s artificial intelligence industry has made breakthroughs, but it still faces the lack of a unified and controllable generation framework and self-testing standards for generated content. Strengthen the standardization of large models, ensure that the quality and performance of large models reach the expected level by formulating unified national standards, ensure user data security and privacy protection, and support the healthy and sustainable development of the large model industry.”

Under the “group model flurry”, the large-scale model competition is a protracted battle

According to incomplete statistics from Sina Technology, as of now, over 100 large-scale model products have been launched in China, among which there are as many as 26 companies that only target general-purpose large-scale models and “general + industry”. In addition, there are 7 “basic + industry large-scale models”; 67 large-scale models in other industries such as finance, medical care, biology, transportation, and data analysis.

With more and more large-scale model companies and AI applications emerging, how can the domestic large-scale model industry develop in a standardized manner under the “cluster model dance”? Liu Cong saw that in this large-scale model competition triggered by ChatGPT, domestic large-scale model companies are destined to encounter various difficulties and problems in the process of “straight-line charge” aiming at the advanced level of foreign countries. Obstacles, such as challenges in chips, high-quality data, system iteration, regulation, etc. “Straight charge” will also be a “protracted battle”.

Liu Congjian saw that on this road, the industry needs to go first, and innovation and application should drive the overall development of the ecosystem. Of course, if you want to match the target and keep the gap narrowing, “stamina” is also crucial. After the phased results are made, industrial development and scientific research innovation must go hand in hand, and the paradigms of each other will continue to iterate and promote each other, so as to continuously inject long-distance running power into the development of my country’s independent large-scale model.

Be confident to do a good job in the “national standard development” of large models, and will do three things well in the future

Since its establishment, iFLYTEK has been engaged in research on core technologies such as speech and language, natural language understanding, machine learning reasoning, and autonomous learning, and has repeatedly worked in machine translation, natural language understanding, image recognition, image understanding, knowledge graphs, and knowledge discovery. , Machine reasoning and other international evaluations have achieved good results.

“Based on our company’s strong technical foundation and strength, we have the ability and confidence to do a good job in the research and development of large-scale model national standards. We will give full play to our own technology and R&D advantages, actively communicate and cooperate with all parties, and jointly promote large-scale national standards. The formulation and improvement of standards. At the same time, we will continue to explore and innovate to provide technical support for the high-quality development of the artificial intelligence industry.” Liu Cong said.

According to him, under the guidance of the National Artificial Intelligence Standardization General Group and the Large Model Special Group, iFLYTEK will do the following work to participate in the formulation of national standards for large models in the future:

First, provide technical support and experience sharing. HKUST Xunfei has deep technical accumulation and rich application experience in the fields of intelligent voice, image, NLP, etc., and will provide technical support and experience sharing for the formulation of national standards, and actively promote the application of national large-scale model standards in finance, medical care, education, transportation, etc. In order to better understand and solve the problems in practical application, it can be applied and promoted in various industries and fields.

Second, promote the establishment and improvement of the scientific system evaluation system. HKUST Xunfei has long been committed to the research and innovation of artificial intelligence technology, and will actively strengthen the research and innovation of large-scale model evaluation technology and methods, improve the scientificity and objectivity of evaluation, establish a professional evaluation team and a comprehensive evaluation system, and serve the country The establishment of standards provides scientific evaluation results and suggestions, and promotes the continuous innovation and development of large-scale model technology.

Third, strengthen cooperation and exchanges. HKUST Xunfei will strengthen cooperation and exchanges with other joint leader units and related companies, research institutes, universities, etc., to jointly promote the construction and monitoring of large-scale model ecology, collect and analyze data and feedback information of large-scale models in different scenarios, Guarantee the rationality and operability of the formulation of national standards.