Knowledge management is one of the central drivers of sustained enterprise development and innovation, and has long faced the challenge of inefficiencies. The advent of the age of large models has opened up new opportunities for enterprise knowledge management, and how can corporate knowledge management processes be reshaped through large models to improve efficiency? How can the boundaries of knowledge be widened and the intrinsic value of knowledge be tapped? What is the value of knowledge management re-engineered by large models for enterprises? With these issues in mind, in the section on dialogue ai: cloud imam laboratories, an in-depth discussion was conducted with the editor of infoq, general manager of the 100-degree smart cloud knowledge management product line, a knowledge management specialist at the south network institute of science and research。
Below is the essence of the current column
Infoq: what is enterprise knowledge management? What's it worth to a business
Song hong-hyun: knowledge management has emerged abroad since the 1960s, and since 1998 domestic enterprises have been exploring its application。
Knowledge management development is divided into three main phases:
The first phase is basic digitization. The knowledge management platform at this stage is more about the aggregation and aggregation of the scattered data of the enterprise, but the knowledge is not being used in depth, and the initial knowledge management is mainly the documentation system。
The second stage is the informationization phase of knowledge. At this stage, firms process and organize the collection of personal and systemic knowledge, which leads to knowledge portals or business searches。
The third phase is knowledge management in the age of large models, which has been accompanied by a significant improvement in the overall process efficiency of knowledge management. The strong understanding capacity of the large models has made knowledge-processing and application issues that the knowledge management platform had previously been difficult to address。
In the third phase, knowledge management can become more business-oriented, making it more visible within the enterprise as a teacher who can forever transmit knowledge, providing a continuous supply of knowledge to the enterprise's staff and better measure of value to the enterprise。
Infoq: how does knowledge differ from traditional knowledge management platforms
Song hong-hyun: knowledge is the first large model in the country to be fully re-engineered as a one-stop knowledge management platform to fully reshape the knowledge management process。
Indeed, four issues in the area of knowledge management have been very difficult to resolve and are bottlenecks to the development of knowledge management:
With a large model, knowledge management has actually entered a new phase. The four previous problems, if addressed one by one in the age of small models, would be very costly, and each model would need to be heavily labelled and fine-tuned. And in the current age of big models, we can solve these problems without any engineering, at very low cost。
In addition, i think the biggest change that a large model can bring to knowledge management is actually agent technology. An intelligent person is going to perform complex tasks, not only to understand the mission itself, but more to develop a knowledge-driven solution paradigm, so i also think that knowledge management is the first stop in a mature application of a large model。
Infoq: what are the challenges faced by the south network institute of science in knowledge management
Lin chingping: while standard knowledge management at the south network institute of scientific research has in the past focused primarily on file levels, the fact that former line producers can retrieve information only through paper-based standards or through standard names on standard platforms, and how to drive all people to complete the work with uniform standards, is an urgent issue for us to address throughout the business. This is the biggest problem facing standard knowledge management。
Infoq: know how to solve the knowledge management challenges of the south net institute of science and research
Song hong-hyun: we've been working with the south net institute for several years, and we've been working on the question of the visibility of knowledge, and many companies have similar needs - - there's a lot of experience in the business, and it's in the brain of an old expert. Can you give me a platform or a system to “extract” the knowledge of an old expert
This vision responds to the need to “be able to introduce rules and hidden data into knowledge management systems”. In response to this original demand, the knowledge management platform provides multiple forms of data access. For example, api access, manual uploading and pull-and-pull methods, such as a series of knowledge synchronization mechanisms, enable enterprises to access the system quickly with structured, or abundant, unstructured data, or even data-type knowledge。
There is also a more important product upgrade in 2024: the platform connects a very large number of enterprise-level sources of knowledge, such as confluence, wiki, a variety of enterprise webs, including even flying books, nail files, enterprise micro-intelligence files, which are widely available within the enterprise, allowing rapid and seamless integration and access to the knowledge platform and greatly increasing the efficiency of the entire knowledge access。
More importantly, knowledge of the platform provides ai-driven, collaborative editing and intelligent writing tool components that enable many enterprises to write and write on their own in their day-to-day work, making it possible to create, i. E., sink, improve efficiency and have lower management costs。
Infoq: know how to help enterprises build knowledge systems
Sang-hoon: we need to organize raw knowledge, raw data into forms that big models can understand. This is also one of the most labour-intensive systems in the knowledge management tradition。
The knowledge management platform provides three more core technical capabilities, first, in multi-modular resolution, second, in the structure of knowledge, and third, in the processing of knowledge elements。
The capability of multi-modular resolution refers to a growing number of enterprises with extensive data in unstructured documents containing complex tables, diagrams and large sections of text. For example, tens of thousands of national, line and business tenders exist in PDF files. Knowledge of the visual understanding of the ocr through large models can be very easily, efficiently and quickly deciphered into chapters, bars, subparagraphs, paragraphs, or even graphs of the enterprise's inventory, which is critical to the success of subsequent knowledge processing。
In addition, for enterprise-level searches and rags, knowledge provides the capability to configure the cut-off of paragraphs and to configure fixed lengths. The most important thing is to know the semanticization units that can be very precise by intelligent means, so that our rag and search can reach a very precise level。
With the help of large models, this series of processes, as just described, can achieve large-scale production in a very efficient manner. Enterprises need to be equipped with the necessary knowledge operators to verify, validate and store the results of the production of large models, at least two to three times more efficiently than in traditional ways。
Infoq: how does the south net institute of science build knowledge systems
Lin chingping: the south net institute of science is actually very concerned with the field of science and technology innovation. We have huge amounts of knowledge data, such as papers, standards, patents, project results, etc. It's a huge challenge how to divide our project data into the technology system tree. In the past, we have classified these resources more according to some rules, that is, the rules that experts had to combe. The overall efficiency is relatively low and its accuracy rate is not high enough to require extensive manual review。
With large models now in place, and with associated smart technologies, we will organize a knowledge system framework for the classification of knowledge based on business needs, using models to automatically label and fast-track the classification of big knowledge, allowing for analysis and discovery of that knowledge, which is still relatively efficient。
And it's also about the discovery of knowledge linkages, where our knowledge used to be alone, like our paper, which i can only see in some of its own content, and where knowledge links are rarely found. Like the standard, it has a reference relationship, and in the past i could only see one standard, but i do not know what some of the operational instructions that this standard relates to and what are the technical specifications that relate to it? The use of knowledge is not very deep, and the linkages between knowledge as a whole are very difficult to achieve. With the addition of smart technology, the platform can better tap into the links between knowledge, ultimately building knowledge maps, supporting the reasoning of standard texts, dissertation-related knowledge and question-and-answer applications, which are of great help to us。
Infoq: what are the technological breakthrough points for reconfiguring knowledge systems through large models
Song hong-hyun: for knowledge organizations, our knowledge is derived from isolated business systems that are not in themselves systematic and organized. Essentially, the creation of knowledge organizations and systems is the establishment of a very complex multi-dimensional classification system。
The traditional knowledge-management age is largely people-based, based on a very powerful semantic understanding of the big model, and we can use simple p-Rompt, which allows for classifications such as zero-shot and few-shot, allows for very little involvement of operators in the process and enables the construction of complex classification systems。
The technology of the knowledge mapping of the past, which ranged from relationships to dissension to codification, depended on a very large number of small models. Large models are now available, and in fact in some applications they have proved that there is little need for specialized data labelling, and that an initial accuracy rate of over 80 per cent can be achieved based on the understanding capabilities of the large models themselves, combined with the mechanisms built by the knowledge mapping platform。
With large models, knowledge demonstrates a strong ability to organize and link knowledge. At the application level, we have successfully achieved a range of functions, such as knowledge-based portal construction and individualized referral, and we have truly achieved a thousand people. Indeed, the realization of these capabilities has benefited from the introduction of large models that have enabled us to systematize the organization and management of business knowledge at a lower cost。
Infoq: know how to improve the efficiency of knowledge production at the south net institute of science
Rin jung-ping: before the big models, we worked together on a number of studies related to standard questions and answers, which we used to do in the past in the form of knowledge mapping. I recall that two to three years ago, we organized a large number of experts to comb into our standard indicator systems, which in total were to combe a dozen levels of systems and eventually build a standard knowledge map of our pilot equipment through a series of work related to knowledge extraction, dissimilaration, etc。
With the big model, the workload that we actually need to involve experts has been significantly reduced, and now basically we just need to give a few simple questions and answers, training through a big model, to complete the construction of the question and answer model and the development of question and answer applications. This has been a great efficiency improvement for our knowledge managers, which has greatly liberated our productivity. We have now built standard questions and answers, as well as semantic search-related service tools, and access to standard knowledge has improved by more than 50 per cent。
At present, large models can also assist staff in preparing standard-relevant outlines, textual summaries of scientific and technical reports of scientific institutions, background, etc. The efficiency of our work has also been more than tripled, which is still evident in terms of the efficiency of knowledge production in enterprises。
Infoq: know how to help businesses use good knowledge
Sung-hoon: to be able to drive innovation, knowledge has to be integrated with scenery. We just said that knowledge management is an old expert in the southern grid. But this old expert, who should be a role-player, who should be oriented towards the senior management of the enterprise, who may be more interested in the macro-business data of the enterprise, who may be more interested in the mid-level management of the enterprise, who may need to know the progress of the project, and who may be more interested in technical programmes. Every key player in a business has different demands for knowledge. We therefore believe that if innovation in knowledge is to be broadly realized within an enterprise, knowledge must be diffused, that is, it should be applied to the business system to form a true business and landscape knowledge assistant。
Knowledge actually provides enterprises with three capacities to help them better use knowledge。
The first is the centralized rag capacity, which has created an enterprise-level rag platform to ensure that every employee in an enterprise can truly obtain a valuable answer to questions in the km system。
Second, the knowledge referral system enables the personalization of thousands of individualized recommendations, as well as immediate recommendations for business scenes, through the analysis of the image of the employees of the multi-isomers within the enterprise. The knowledge management platform is no longer just a knowledge portal or a knowledge retrieval system, but has been upgraded to a platform built by agent, a knowledge-based enterprise. On this platform, based on built knowledge systems, we are able to create a range of knowledge-driven agent assistants, such as knowledge travel assistants, reimbursement assistants, sales assistants and even hr assistants. I believe that only when the three levels of search, referral and agent service are well integrated on the same platform will our in-house applications become very diverse and the sources of innovation continue to flourish。
Finally, the knowledge production, processing and application processes of enterprises are covered by this knowledge, which will eventually lead to a closed loop and feedback mechanism for enterprise employees, knowledge applications and knowledge production. As a result, in the course of its use, we continuously collect feedback from enterprises on the effectiveness of the application of knowledge management products, which enables us to continuously improve the efficiency of knowledge production and processing。
Infoq: how does the south net institute of science use its knowledge to empower every employee
Lin jungping: the south net institute of science has been promoting the upgrading of knowledge from human to human. The whole of the southern power grid has a large number of jobs and demands for efficiency and security, and our concept of knowledge management is based on a three-dimensional perspective of users, scenes and knowledge, for example, where our scientists focus on quick access to high-quality reports, core papers, recent research results and historical project experiences in relevant fields. Our first-line producers, on the other hand, are paying more attention to the methods of handling deficiencies and to operational norms, and there is some difference in the need for knowledge in each job。
In the past, we gained knowledge mainly by searching. However, in the age of artificial intelligence, we need more than ever to recommend targeted, high-quality knowledge for users from the point of view of their image and use, in order to increase the efficiency of access to knowledge. Through knowledge management, we want to summarize and settle best practices while building on the legacy of experience. The mission of our knowledge management platform is to support the pooling, storage and use of knowledge, and even to embed knowledge directly into the business landscape, without manual intervention, to better support innovation and high-quality enterprise development。
