Loading...
[x]
Player version
Player FPS
Video type
Video url
Video resolution
Video duration
Character sketch: Observing Zong Yunbing's daily work, you will find that he and his team not only have the basic professionalism of algorithm engineers, but also have keen insight and business thinking. Right now, Zong Yunbing is highly focused on a group discussion, and in his well-documented analysis, a new application scenario is gradually becoming clear.
On the afternoon of April 25, in Jinan Jinhyundai Company, Zong Yunbing, who had just returned to Jinan from Guangzhou, devoted himself to the discussion with the team. They plan to cooperate with a large securities company in Guangzhou to try to apply the AI model to the query of policies and regulations and the writing of securities market reports.
This is another exploration of the practical application of AI technology by Zong Yunbing and his team in addition to electric power, construction, laboratory, and enterprise management. It has only been a year since they formed a team and opened the door to a large AI model.
According to the industry, the current artificial intelligence industry is in a golden age of "fast enough, no wind is needed". Zong Yunbing, like many algorithm engineers, is stepping up his efforts to seize this opportunity.
A "magic hammer" fell from the sky
"Everyone knows that AI has been very hot in the past two years, but what we need to do is to analyze what prospects this popularity has from a business perspective and what business scenarios can be matched. In other words, it can't just thunder, it has to rain. Zong Yunbing recalled that around this time last year, the company set up an artificial intelligence research institute, and under the strategic adjustment, more than half of the algorithm engineers turned to the direction of large models. As the leader, Zong Yunbing knows that what they are going to do at this moment is of epoch-making significance.
"We all know that there is a huge difference in the efficiency of artificial and large models, which is the qualitative change of productivity. The "big models" that Zong Yunbing is deeply involved in refer to AI models with huge parameter scales, and these models can understand huge amounts of data through "learning". Its advent means that human beings can use technology to compress massive amounts of data, and extract and use them easily and quickly.
In the past, Zong Yunbing's team could only rely on intelligent document processing technology to solve intelligent problems for enterprises, and the emergence of large models is undoubtedly exciting for these engineers. "It's like there is a big row of nails in front of you, and suddenly a 'magic hammer' falls from the sky. Zong Yunbing described their encounter with new technology. Curiosity and a desire to explore also drive them to try to "domesticate" the large model, to make it closer to human logic, and to be used by humans.
The "imperfections" of large models
Full of hope brought by technological innovation, Zong Yunbing is ready to show his strength and start development, and their initial focus is the digital field of power and other industries that have served for many years.
But it is not easy to put it on the ground. With the practical application of large models, a series of problems have gradually become prominent. One of the most prominent is that the large model can only support very simple conversations initially, cannot understand complex language, and lacks logic in thinking and answering. It can be said that "serious nonsense" has become the fatal flaw of the application of large models.
Whether this defect can be solved is the key to the enterprise application of large models. Because the training of the general large model is mainly based on the public Internet knowledge, it does not know anything about the internal knowledge of the enterprise, so it cannot meet the requirements of business development.
"Enterprises often have the characteristics of private data, so we try to deploy large models privately to the enterprise intranet. Zong Yunbing introduced that the process of "knowledge feeding" behind this is not only a huge amount of engineering, but also puts forward higher requirements for algorithm engineers: "We must understand the business knowledge of these industries, otherwise it is difficult to process data." ”
This process takes time and cost, and tests confidence and patience. Zong Yunbing said that they are willing to fight for their dreams once. At the office site, we felt the attitude of these young people's hard work: among the more than a dozen algorithm engineers, some concentrated on typing code; Some people gathered in front of the display board and had a heated discussion on a technical difficulty; There are also engineers who have been stationed in the enterprise for a long time with computers, just to understand the business of the enterprise......
A "Golden Age" of Opportunity
Data collation based on business knowledge is very hard work, and it is also the most important infrastructure to determine the accuracy of large models. They sit down with customers to learn about the business, and work with customers and business people to verify accuracy...... After repeated adjustment and adaptation, the technical difficulties were overcome one by one. With the support of Zong Yunbing's team, the company has incubated a large-scale model product - "Xiaojin Zhiwen". The product is gradually able to learn and adapt to various business scenarios, and the data generated is based on real facts and meets user expectations.
In the emergency management of electric power enterprises, the staff can "ask" questions about industry norm requirements, emergencies and emergency measures at any time during the emergency repair process, and the system can give accurate answers and the source of the answers, such as "what emergency repair measures should be taken in the face of a 12-level typhoon" and so on.
In the field of automobile manufacturing and maintenance, large models can help the R&D department, legal department, production department and test department of car companies locate and understand various standard requirements at any time, whether it is material selection in the design stage, process flow in the manufacturing stage, performance evaluation in the testing stage, or even fault diagnosis in the maintenance stage, it can provide accurate and timely guidance.
The laboratory is a department with a very high data density, and after all kinds of data such as order information, quality control data, raw data, analysis reports, and performance rankings are connected to the large model, they can retrieve data at any time, answer all kinds of questions, and support display in the form of charts.
In the past year, Zong Yunbing led the team to seize the technical outlet of "large model" and successfully realized the application in many fields, helping customers greatly improve work efficiency, reduce operating costs, and achieve data-driven decision-making. Today, they don't stop there, as the success stories give them more inspiration and inspiration......
"Artificial intelligence technology represented by large models, in the view of our algorithm engineers, is a new quality of productivity, which will play a powerful role in promoting social development. Zong Yunbing said frankly that this era is full of infinite possibilities, and what they have to do is to seize the technological outlet and turn "possibility" into "reality".