In recent years, artificial intelligence has become one of the most active frontier fields of global digital technology innovation, and is a new track for the digital economy and a new hot spot for international competition. At the same time, the public has doubts about artificial intelligence, how to control the speed and direction while stepping on the accelerator, so as to build a safe and credible artificial intelligence industry ecology, and explore AIGC (generative artificial intelligence) based on trusted data, which has become an important direction for future breakthroughs in artificial intelligence.
On July 3, Qichacha released the world's first commercial inspection large model - Qichacha "Know the Alpha". This large model is based on the results of large-scale pre-training after nearly ten years of accumulation of trusted data in the field of business inquiry, and the innovative products that will be launched in the future are jointly built through the large model + enterprise credit database, using AI technology to empower enterprise credit big data innovation, create safe and credible artificial intelligence products, and provide users with more convenient and accurate business information query services.
(Photo: Qicha Shangcha large model "Zhipi Alpha")
Why is the industry's first commercial inspection model Qichacha?
Chen Deqiang, founder of Qichacha said, Qichacha has an enterprise credit database covering nearly 500 million enterprises around the world. The core resources of business information inquiry services are related data resources, such as industrial and commercial information, litigation information, etc. Moreover, the data coverage should be broad enough, not only with the full data of China, but also with the data of other countries around the world. In addition, in highly specialized fields such as enterprise standard and enterprise due diligence, it is also necessary to establish a special database. To introduce AIGC technology into the field of business information query services, the most critical action is to use the enterprise information dataset to train the relevant model. The larger the amount of data and the higher the data quality, the more accurate the final trained model will be. Compared with other large language models, which mainly use public Internet datasets, it is more difficult to obtain professional data such as industry and commerce and justice. It can be said that the 500 million enterprise credit data resources covered by Qichacha are the core barriers of Qicha and business inspection services, bringing together 80 industrial chains, 8,000 industries, and massive market real-time industrial and commercial information, risk disclosure, intellectual property rights, credit reports, equity relations and other 300+ dimensions of enterprise credit data in the current domestic market. These data not only build high competitive barriers for Qichacha but also lay a solid data foundation for its training of the Zhibi Alpha Business Investigation model.
It is understood that the AI algorithm model of Qichacha is leading in China, and won the "China's highest award for intelligent science and technology" Wu Wenjun Artificial Intelligence Science and Technology Award in May this year. Over the years, through the use of deep learning, natural language processing (NLP) and other AI technologies, Qichacha has realized automated and intelligent data analysis and text mining in massive global multilingual texts, and can further realize in-depth semantic analysis to provide users with more accurate semantic retrieval services. In the field of pre-training models, with rich data resources, Qichacha has a strong technical accumulation, and the Zhibi Alpha commercial inspection large model released by Qichacha has achieved completely independent intellectual property rights.
Rich and diverse products, services and application scenarios. Users are confronted directly with the product, not the technology. Therefore, in order to apply the large-model technology underlying AIGC to business search services, it is necessary to package the technology into a product that is convenient for users to use. Moreover, according to different user needs, it is necessary to build targeted products, and then form a relatively complete product matrix. In the form of product matrix, to provide one-stop service for business inspection users. At the product level, Qichacha has built a perfect product matrix for different user groups, specifically: for enterprise users, Qichacha provides solutions for scenarios such as accurate customer expansion, enterprise rating, due diligence, risk control, judicial investigation, public opinion monitoring, and supply chain management through customized services, assisting enterprise users to improve corporate portraits, cross-check information, and find partners; For individual users, Qichacha provides solutions for scenarios such as investment and financing, job recruitment, and risk determination through the cloud platform to integrate multi-dimensional data, so as to provide individual users with a perspective on the company's equity structure and avoid credit risks in the process of enterprise identification. For public sector users, enterprise investigation data is not only an important supplement to the central bank's official credit investigation channels, but also an important reference for local government policy formulation, social credit system construction, investment promotion, screening policy support targets, and enterprise credit supervision.
Why build a business search version of ChatGPT?
With the accumulation of data volume and the increase of product functions, many problems and demand pain points have occurred in the field of business, mainly in two aspects:
Users are still stuck in the keyword search stage, and the platform cannot well understand the complex business needs of users. At present, most business inspection platforms are essentially search engines in the field of enterprise credit information, and it is difficult for users to express the complex and structured needs of users by searching for the corresponding enterprise or risk information in the enterprise credit database through keywords. For example: a school canteen bidding process wants to understand the supplier's food safety risk, with the previous business inspection product is generally to first enter the name of the company participating in the bidding, and then jump to the main page of the corresponding company, and then by looking for the company's "business risk" in the administrative penalty or "food safety" content in the "business information", in order to determine whether the company has food safety risks, there are many steps, and it is not friendly to new users. If the user also wants to know whether other catering companies operated by the enterprise legal person have had major security incidents? Are there any affiliated companies accompanying the bid? Similar to these more in-depth requirements, it is difficult to quickly satisfy the user by simply retrieving information. Even if the requirements can be realized, the user itself needs to have relevant professional knowledge, and be proficient in the search functions of the business platform, and finally get more in-depth and perfect information about the enterprise after multiple steps of relatively cumbersome operations. However, for the vast majority of new users, the threshold for use is too high, which is not conducive to opening up information channels in a wider range.
In addition, the search engine model, the business platform feedback to the user is a large amount of basic data, not a direct answer. If it is a large group company, there may be thousands of basic enterprise credit data, and such a huge number of views is a big burden for users. Such a business information query service is still in the tool stage and cannot be called a powerful business assistant. At the same time, Qichacha has hundreds of products and services such as checking enterprises, checking bosses, checking risks, checking bidding, credit big data, risk big data, etc., it is difficult for users to learn and master systematically, based on the Zhibi Alpha large model, Qichacha will be launched in the follow-up dialogue products, you can skip the cumbersome retrieval steps, and fully release the capabilities of various products in the way of dialogue, and provide users with "holistic, easy-to-understand, high-value" business information services.
When users use large language models such as ChatGPT to search for high-value enterprise credit data, they will find obvious problems: due to the lack of professional database support, the enterprise business and credit data searched through ChatGPT are derived from public Internet data, and the accuracy of the data cannot be guaranteed. Lacking the support of professional databases, the analysis of large language models such as ChatGPT in the field of business investigation is "rice cooking", and even "making something out of nothing". The Zhipi Alpha Business Check Model is an in-depth training based on the full amount of credible data of Qichai, which can provide users with professional enterprise credit data and diversified analysis results.
AIGC+ enterprise credit database, business inspection service mode further evolution
As artificial intelligence enters the "AIGC era", Chen believes that this will completely change the query and usage patterns of data. "After the scale of industry data reaches a certain magnitude, the data query method has changed, and the AIGC+ enterprise credit database can fully utilize the data." According to reports, the Zhibi Alpha Business Check Model currently released by Qichacha is the world's first commercial inquiry large model, which is based on the global enterprise credit data covered by Qichacha for training, compared with the traditional business inspection platform, the subsequent products built based on the Zhibi Alpha Business Check Large Model have achieved the following three aspects of change:
In terms of human-computer interaction, natural language dialogue is realized, that is, complex query steps can be completed. Users want to query a certain business data, no longer limited to keyword search, but can use a natural language description to put forward their own needs and lower the user threshold. For example, when a user wants to do a shallow due diligence on a company, they can ask, "What is the position of a company in the industry?" What are the competitors? The Alpha model understands the user's needs from the user's description and "deconstructs" the requirements into corresponding instructions. In this way, when users understand a company, they are faced with the Alpha model like a professional business investigation assistant, rather than a tool without wisdom.
In terms of technology, it has achieved a second-level response to the needs of users. The Alpha model can retrieve the enterprise information data according to the user semantics, and then present the results of "sorting and summarizing" to the user. In this case, the user is no longer getting a bunch of scattered information, but a complete answer. In order to improve the response speed, Zhibi Alpha large model has been fully connected to the Qichacha supercomputing platform, and tedious steps such as query, browsing, summarizing, and structured output can be completed in seconds.
The new "multi-round dialogue" function allows the business search platform to have logical thinking capabilities. The "multi-round dialogue" function of the Zhibi Alpha large model is a highlight of the model's superiority over previous Qichacha conversational AI. With this ability, the Alpha model can guide users through multiple rounds of dialogue to conduct deeper analysis step by step. In multiple rounds of dialogue, users can ask deeper questions by asking new instructions based on the results already obtained. In this way, the Zhibi Alpha model not only becomes the user's assistant, but also a "guide", gradually guiding the user to seek the answer by himself.
![Jiang Qing touted the model play at the dinner table, and Wang Jianan was displeased, bluntly saying that the model play was like braised pork[fig]](data:image/gif;base64,R0lGODlhAQABAIAAAP///wAAACwAAAAAAQABAAACAkQBADs=)