Exploring the Path of AIGC Applications

Artificial intelligence is a new track in the digital economy and a new hotspot in international competition. At the same time, public concerns about artificial intelligence remain, and how to control speed and direction while stepping on the gas to build a safe and trustworthy artificial intelligence industry ecosystem and explore AIGC based on trustworthy data has become an important direction for breakthroughs in the future of artificial intelligence.

Since 2015, China has gradually introduced policy documents on artificial intelligence, which has gradually risen to a national strategy and achieved rapid development, with the core industry scale exceeding 500 billion yuan. In recent years, with multiple rounds of technological innovation in the artificial intelligence track, the industry entry threshold has continuously lowered, and the domestic market has significantly expanded. According to Qichacha data, as of May 2023, there are over 700,000 existing AI-related companies in China in the fields of information transmission, software and information technology services, scientific research, and technical services.

Since 2013, China has cumulatively registered 845,000 AI-related companies, with over 70% of the companies registered between 2020 and 2022, as shown in Figure 1.

In terms of year-on-year data, from 2013 to 2019, the market capacity of China’s artificial intelligence track steadily expanded, with the growth rate of new registrations basically maintaining at 33% to 59%. By 2019, the annual new registration volume of AI-related companies in China increased by 48.0% year-on-year, reaching 41,000.

In 2020, the number of companies entering the AI track in China saw explosive growth, with 114,000 new registrations that year, a staggering increase of 181.0% year-on-year. On this basis, in 2021, the number of new registrations further increased by 98.9% year-on-year to 226,000, and in 2022, the annual new registration volume exceeded 250,000 (the values here and in Figure 1 are rounded, and growth rates and proportions are calculated using original values).

From January to May 2023, there were 125,000 new registrations of AI-related companies in China, a year-on-year increase of 21.2%.

On February 10 this year, Wang Huiwen, former co-founder and senior vice president of Meituan, announced a capital entry declaration in his social media circle. Within a few days, he released a hero list, stating that he is learning AI technology, with personal shares not holding stock, 25% of funds for equity, and 75% of shares to invite top R&D talents, establishing Beijing Beyond Light Technology Co., Ltd., revealing that he has obtained $230 million in subscriptions from top VCs, visually placing the excitement of the tech circle in front of the public.

Regrettably, just four months after Wang Huiwen immersed himself in the large-scale model startup, he left his post due to illness, and Meituan announced the acquisition of 100% equity in the large model startup Beyond Light for 2.065 billion yuan.

On the evening of February 26, former head of AI technology at JD.com, Zhou Bowen, recruited AI talents such as R&D partners and product partners in his social media circle, frankly stating that he has been a veteran in the field of natural language generation, dialogue, and interactive artificial intelligence for over twenty years. Without technical vision and creative concepts, the path of China’s OpenAI cannot go far, and empty talk about technology without commercial landing scenarios is also unsustainable; he also stated that many top investment institutions have provided ample funding for research, development, and product validation. According to Qichacha data, Zhou Bowen’s startup, Xianyuan Technology, founded at the end of 2021, completed hundreds of millions in angel round financing led by Qiming Venture Partners and followed by Matrix Partners in early March this year.

March saw a gathering of giants, with media reports on the 7th suggesting that former Amazon chief scientist Li Mu had seemingly joined a large model startup. The next day, multiple media outlets reported that Li Yan, former head of Kuaishou’s MMU, founded Yuan Stone Technology, mainly focusing on the R&D of multimodal large models; on the 19th, AI guru Li Kaifu officially announced in his social media circle that he is personally organizing Project 2.0, a global company dedicated to building a new platform for AI 2.0 and AI-first productivity applications; at the end of the month, former ByteDance AILab director Wang Changhu left Longhu, with a startup direction focusing on generative AI visual multimodal algorithm platforms.

Domestic AI experts are rising, and the AI layout battle of tech giants is heating up, with the inaugural moment beckoning.

Generative large models, due to their technical characteristics, require more data for continuous training of the models. Therefore, data-driven tech giants inherently possess the ability to develop large models.

On March 16, the large language generative model “Wenxin Yiyan” was launched, marking Baidu’s first shot in the domestic large model product competition, with tech giants joining the multidimensional exploration of the large model field.

On April 11, the “Tongyi Qianwen” large model was unveiled at the Alibaba Cloud Summit, followed closely by Zhihu, which collaborated with Mianbi Technology to release the Chinese large model Zhihai Map AI, officially starting internal testing on April 13, 2023.

In the same month, Huawei and Inspur again showcased their previously released large models from 2021, with Inspur’s “Yuan 1.0” surpassing OpenAI’s 175 billion parameter English language model GPT-3 at launch, becoming the largest-scale artificial intelligence model globally. Huawei’s “Pangu” has continuously released Pangu drug analysis large models, Pangu mining large models, Pangu meteorological large models, Pangu wave large models, and Pangu financial OCR large models since 2021, continuously empowering industry development.

On May 20, Qichacha publicly released the large model “Zhibi Alpha” and conducted a live roadshow. As the world’s first commercial inquiry large model, “Zhibi Alpha” is deeply trained based on Qichacha’s full volume of trustworthy data, achieving conversational querying, instant complete responses, and multi-turn dialogue transformations, providing users with professional, precise, and convenient commercial information inquiry services.

In addition to the tech giants with existing large models, Tencent announced that its “Hunyuan” will be launched within the year, JD.com is about to introduce a new generation of large models, and ByteDance and Ant Group responded to the development of language and image large models, as well as language and multimodal large models, with Ant Group internally naming its large model “Zhenyi”. Kuaishou stated that it has formed a large model R&D team and will proceed with the development and training of large language models as planned…

Before the troops move, logistics must be prepared first. For enterprises, expanding the domestic market requires prior trademark registration. According to Qichacha data, as of mid-June this year, there are currently 3,222 applications for GPT and large model-related trademarks in China, of which over 86% were registered in 2023. Among these 3,222 trademarks, 2,688 are in the status of “registration application”, accounting for 83.4%, while 235 are in the status of “registered”, accounting for 7.3%.

From the perspective of applicants for large model-related trademarks, iFlytek has applied for the most trademarks, with 24, followed by Baidu with 12, and Huawei with 10. Among them, Baidu acted the fastest, applying for 4 large model trademarks in 2021, 2 in 2022, and Huawei applied for 7 in the same year.

On May 20, Qichacha released the commercial inquiry large model “Zhibi Alpha”, as shown in Figure 2.

This large model is the result of large-scale pre-training based on nearly a decade of trustworthy data accumulation by Qichacha in the field of commercial inquiry. The innovative products to be launched later will be constructed jointly through the large model and the corporate credit database, empowering corporate credit big data innovation with AI technology, to create safe and trustworthy artificial intelligence products, providing users with more convenient and precise commercial information inquiry services.

Why create a commercial inquiry version of ChatGPT? Qichacha founder Chen Deqiang stated that with the accumulation of data and the increase in product functions, many problems and pain points have emerged in the commercial inquiry field, mainly reflected in two aspects.

On one hand, users remain at the keyword search stage, and the platform still cannot well understand users’ complex business needs.

Currently, most commercial inquiry platforms are essentially search engines in the field of corporate credit information, where users search for corresponding companies or risk information in the corporate credit database through keywords, making it difficult to express complex, structured needs.

For example, during the bidding process of a school cafeteria, if one wants to know the food safety risks of suppliers, the previous commercial inquiry products typically require entering the names of participating bidding companies, then jumping to the corresponding company homepage, and checking the company’s “operating risks” for administrative penalties or “operating information” for “food safety” content to ascertain whether the company has food safety risks. This process is cumbersome and not user-friendly for new users.

If users also want to know whether the legal representative of that enterprise has had major safety incidents in other catering companies or whether there are related companies accompanying the bid, similar deeper needs are difficult to quickly satisfy with simple retrieval information.

Even if the needs can be met, users themselves must possess relevant professional knowledge and be proficient in various search functions of the commercial inquiry platform, going through multiple relatively cumbersome operations to ultimately obtain more in-depth information about the enterprise. This presents a high barrier to entry for most new users, hindering the broader flow of information.

On the other hand, based on the search engine model, commercial inquiry platforms provide users with a large amount of basic data, rather than direct answers.

If it is a large group company, there may be thousands of basic corporate credit data, and such a massive volume of browsing is a significant burden for users. Such commercial information inquiry services still remain at the tool stage and cannot be called powerful business assistants.

At the same time, Qichacha has hundreds of product services such as checking companies, checking company leaders, checking risks, checking bidding, credit big data, and risk big data, making it difficult for users to systematically learn and master.

Based on the “Zhibi Alpha” large model, the conversational products that Qichacha will launch later can skip cumbersome retrieval steps and fully release various product capabilities through dialogue, providing users with “integrated, easily understandable, and high-value” commercial information services.

When users search for high-value corporate credit data using large language models like ChatGPT, they will encounter obvious problems: due to the lack of professional database support, the corporate business credit data searched through ChatGPT comes from publicly available internet data, which cannot guarantee data accuracy, while some high-quality, commercialized corporate credit databases are not open to it.

Lacking professional database support, large language models like ChatGPT face the situation of “cooking without rice” in the field of commercial inquiry, even resulting in “fabricating” information. In contrast, the “Zhibi Alpha” commercial inquiry large model is deeply trained based on Qichacha’s full volume of trustworthy data, providing users with professional corporate credit data and diverse analytical results.

Why is Qichacha the first in the industry to launch a commercial inquiry large model? Chen Deqiang stated that Qichacha possesses a corporate credit database covering nearly 500 million enterprises globally.

The core resource for commercial information inquiry services is related data resources, such as business registration information, litigation information, etc. Additionally, the data coverage must be broad enough. Furthermore, in highly specialized fields such as corporate announcements and due diligence, dedicated databases need to be established. To introduce AIGC technology into the commercial information inquiry service field, the most critical action is to train related models using corporate credit datasets. The larger the data volume and the higher the data quality, the more accurate the resulting trained model will be. In contrast, other large language models mainly use publicly available internet datasets, making it more challenging to obtain specialized data such as business registration and judicial data.

It can be said that the corporate credit data resources covering 500 million enterprises by Qichacha are its core barrier to commercial inquiry services, aggregating 80 industrial chains, 8,000 industries, and a vast amount of real-time market business registration information, risk disclosures, intellectual property, credit reports, equity relationships, and over 300 dimensions of corporate credit data. This data lays a solid foundation for training the “Zhibi Alpha” commercial inquiry large model.

On the algorithm level, Qichacha also has a deep accumulation of AI technology. Over the years, Qichacha has utilized deep learning, natural language processing (NLP), and other AI technologies to achieve automated and intelligent data analysis and text mining across massive global multilingual texts, further enabling deep semantic analysis to provide users with more accurate semantic search services.

In the field of pre-trained models, based on rich data resources, Qichacha has strong technical accumulation. The released “Zhibi Alpha” commercial inquiry large model has achieved complete independent intellectual property rights. Qichacha’s AI algorithm model is leading domestically and has won the “Wu Wenjun Artificial Intelligence Science and Technology Award”, the highest award for intelligent science and technology in China.

To apply the underlying large model technology of AIGC to commercial inquiry services, it is essential to package the technology into user-friendly products. Moreover, for different user needs, targeted products must be developed, thus forming a relatively complete product matrix to provide one-stop services. At the product level, Qichacha has built a comprehensive product matrix targeting different user groups.

For corporate users, Qichacha provides customized services for accurate customer acquisition, corporate rating, due diligence, risk control, judicial investigations, public opinion monitoring, and supply chain management scenarios, assisting corporate users in refining corporate profiles, cross-verifying information, and finding partners; for individual users, Qichacha provides investment and financing, job hunting and recruitment, risk assessment, and other solutions through a cloud platform that integrates multidimensional data, allowing individual users to gain insights into corporate equity structures and avoid credit risks in the corporate identification process; for public sector users, Qichacha data is an important supplement to the central bank’s official credit investigation channel and serves as an important reference for local government policy formulation, social credit system construction, investment promotion, selecting policy support targets, and conducting corporate credit supervision.

What transformations will AIGC + corporate credit database bring to commercial inquiry services? Chen Deqiang believes that the data query and usage model will be fundamentally changed. “Once the industry data scale reaches a certain level, the data query method will change; AIGC + corporate credit database can fully utilize the data.” The “Zhibi Alpha” commercial inquiry large model has achieved three transformations compared to traditional commercial inquiry platforms.

First, in terms of human-computer interaction, it employs natural language dialogue to complete complex query steps. Users no longer need to be limited to keyword searches but can make requests using a natural language description, lowering the user threshold.

For example, when a user wants to conduct a shallow due diligence on a company, they can ask, “What is the industry position of a certain company? Who are its competitors?” The “Zhibi Alpha” large model will understand the user’s request from the description and deconstruct it into corresponding instructions. When users learn about a company, the “Zhibi Alpha” large model acts as a professional business investigation assistant rather than a tool devoid of intelligence.

Secondly, on the technical side, it achieves second-level responses to user requests. The “Zhibi Alpha” large model can retrieve corporate credit data from Qichacha based on user semantics and present the “organized and summarized” results to users. In this case, users receive a complete answer instead of a pile of fragmented information. To improve response speed, the “Zhibi Alpha” large model has fully integrated into Qichacha’s supercomputing platform, enabling it to complete queries, browsing, summarization, and structured output in second-level time.

Finally, the newly added “multi-turn dialogue” feature gives the commercial inquiry platform logical thinking capabilities. The “multi-turn dialogue” feature is a significant highlight of the “Zhibi Alpha” large model. In multi-turn dialogues, users can ask new instructions based on already known results, thereby posing deeper questions. This approach makes the “Zhibi Alpha” large model not just an assistant for users but also a “guide” that progressively leads users to seek answers.