【Qingxin Observation Room】ChatGPT, Sora's Big Explosion Is Generative AI Credible?

【Qingdao News Network Exclusive】

(Reporter Han Tongtong, Ning Guanyu)

ChatGPT and Sora remain popular, and generative AI makes people feel the convenience of technology more intuitively.

How can you ensure that generative AI-generated content is credible?

In this interview, Zhang Weishan, chairman of the Qingdao Artificial Intelligence Society, said that due to the characteristics of the current artificial intelligence technology itself, there are still a series of problems such as unprofessional and untrustworthy content generated by generative artificial intelligence. Therefore, ensuring that content is credible needs to start with the data source, training process, and inference process. ”

Some people say that artificial intelligence + big data will become the two major engines to promote future development, and the development trend of artificial intelligence in the future will show the characteristics of higher, faster and stronger. It is true that technology is evolving rapidly, but while we appreciate the convenience of technology, we should still retain the ability to think independently.

Reporter: Could you briefly introduce your research direction, what is the "trusted artificial intelligence ecosystem", and what are its advantages?

Weishan Zhang: The "Trusted AI Ecosystem" is an alliance ecosystem based on trusted AI, which aims to enable AI to be used trustedly and safely, and to be trusted at all stages of design, development, deployment, and use. To put it simply, this ecosystem will improve people's trust in AI technology and promote the efficient aggregation and sharing of data elements.

It has many advantages, privacy, security, and explainability, which can better promote the deep integration of AI technology with all walks of life, and is more acceptable.

Reporter: Artificial intelligence has become an important technical means for the digital transformation of traditional fields, what changes will generative AI such as "ChatGPT and Sora" bring?

Zhang Weishan: Recently, generative AI has developed very rapidly, which can be regarded as a revolutionary progress, which has greatly changed the current working mode in many fields, greatly improved the efficiency of video image generation, replaced some repetitive tasks such as translation, polishing, image generation, (partial) code writing, etc., greatly improved work efficiency, and will also have a huge impact on intelligent manufacturing, smart cities and other related industries, changing some key technical support in the past, such as its prediction, analysis, decision-making technology, etc.

Of course, when dealing with professional field problems, these generative large models also need a lot of domain adaptation work to adapt to domain business knowledge, and also produce many professional large models, such as traditional Chinese medicine large models, which are also one of the key tasks of the industry at present.

Due to the characteristics of the current AI technology itself, there are a series of problems with the content generated by generative AI:

1. Unprofessional. For example, in the writing of scientific papers, the generated ones cannot be used and need to be vigorously corrected.

2. Untrustworthy (hallucinations of large models). For example: generating non-existent documents, etc.

For example, researchers at Stanford University published WikiChat, which greatly improved the credibility of model generation results through external knowledge embedding and fact checking. In addition, similar problems can be alleviated by designing security fences through blockchain.

Reporter: In the process of promoting the "artificial intelligence +" action, how do you think we can better gather the strength of the government, enterprises and schools?

Zhang Weishan: Improve the level of data openness. Data is the basic element of artificial intelligence, and in order to improve the level of data openness, it is necessary to strengthen the collection, sorting, analysis and application of data resources to improve the value and benefits of data resources. At the same time, a data resource sharing mechanism should be established to break down data silos, promote the circulation and exchange of data resources of all parties, and provide sufficient data support for trustworthy artificial intelligence technology.

Strengthen government guidance, improve innovation mechanisms, and jointly build a new development pattern of artificial intelligence technology. Formulate relevant special research plans and science and technology projects, establish more efficient incentive policies, and guide and encourage venture capital institutions to participate in the combination of artificial intelligence and all walks of life by providing financial support and risk sharing. Improve the innovation mechanism, take the major key technologies of artificial intelligence as the goal and problem orientation, optimize the allocation of scientific and technological resources, and stimulate the vitality of innovation subjects.

Strengthen industry-university-research collaboration and build a whole-chain industrial ecology. With artificial intelligence technology as the core, integrate upstream and downstream industrial resources to form a complete ecology covering technology research and development, product production, market application, service support and other links. Give full play to the role of universities, scientific research institutes, enterprises and other innovation subjects, promote the sharing of scientific and technological resources, and promote joint research and collaborative innovation of major scientific and technological projects and platforms in the field of artificial intelligence. At the same time, we will increase support for industry-university-research collaborative innovation platforms and alliances, and build an innovation network with multi-party participation and collaboration.

Reporter: Qingdao is a big tourist city, but also a city with a strong industrial base, what do you think can be used to build Qingdao's "artificial intelligence +" action path?

Zhang Weishan: We can build Qingdao's "artificial intelligence+" action path through the following four perspectives to promote the intelligent development of the city and enhance the city's competitiveness and sustainable development capabilities.

Smart tourism: Use artificial intelligence technology, especially the current hot large model technology, to build a smart tourism Q&A platform to provide tourists with personalized and accurate tourism information and services. Through data analysis, recommendation algorithms and other technologies, personalized travel routes, restaurant recommendations, and scenic spot introductions can be customized for tourists to enhance their travel experience.

Intelligent manufacturing: Combined with Qingdao's industrial base, machine learning algorithms can be used to optimize production scheduling, real-time monitoring of transportation conditions, predict sales directions, equipment operation failures, realize the intelligence and automation of production, transportation, sales, and after-sales processes, improve production efficiency and product quality, and save manpower and material resources.

Smart Mobility: Optimize traffic signal control, intelligent street lamp management, and traffic flow management, improve traffic efficiency, and reduce congestion through traffic big data analysis. At the same time, autonomous driving technology and intelligent route planning can be introduced to promote the application of autonomous vehicles on urban roads and improve transportation efficiency.

Intelligent medical care: In the field of medical healthcare, the accuracy and efficiency of medical diagnosis can be improved through joint learning of multiple hospitals, medical big data analysis, and intelligent image diagnosis technology. At the same time, it can promote the application of telemedicine, intelligent health monitoring and other technologies to realize the optimal allocation of medical resources and the intelligence of medical services.

Reporter: Now that the development of artificial intelligence is changing with each passing day, what do you think is the future development trend?

Zhang Weishan: With the rapid development of artificial intelligence, the future development trend will be larger-scale, faster-speed, more modal and more personalized.

Convergence of larger and smaller models: Future AI models will be further scaled up, with more parameters and richer representation capabilities. These large-scale models are better able to handle complex tasks and more diverse data, improving the performance and applicability of AI systems. At the same time, small models are usually deployed on the edge device side, and personalized low-power models, such as the combination of the server side and the edge side, require large-scale deep integration with small models.

Faster training and inference speeds: With the advancement of hardware technology and the optimization of algorithms, future AI models will be able to achieve faster training and inference speeds.

More modal, highly integrated: Future AI models may involve more multimodal learning, that is, processing multiple types of data at the same time, such as text, images, sounds, videos, etc. These multimodal models enable a more comprehensive understanding and processing of information.

More personalized: Future AI models may be more personalized and customized, able to be trained and optimized according to specific industries, specific user needs, and specific scenarios. This will enable the model to better adapt to the needs of different fields and different user groups, and improve the applicability and effectiveness of the model.

To sum up, the development trend of artificial intelligence in the future will show the characteristics of higher, faster and stronger. This will provide a strong impetus for the rapid development of AI technology and promote social progress and economic development.