The general large model ＂reduces fever＂ and the production AI begins to ＂squeeze bubbles＂

Industrial model becomes a new favorite, vertical field or main battlefield

Yangcheng Evening News All-Media Reporter Shen Zhao Intern reporter Xu Xiaowei

On July 18th, Ctrip Group released the first vertical model of tourism industry "Ctrip asked". On July 13th, JD.COM released the large-scale model rhinoceros and demonstrated its application in health, finance, e-commerce and other fields, all of which poured a handful of hot oil on the large-scale model industry, which was slightly less popular in recent days. Since the end of June, the number of ChatGPT visits has declined, and the A-share ChatGPT index has dropped by more than 200 points. All kinds of signs indicate that the popularity of the general big model is declining. On the other hand, at the artificial intelligence conference held not long ago, more than 30 kinds of large models were exhibited together, among which the industrial large model was more eye-catching. Many people in the industry said that compared with the general large model (AGI), the industrial large model is more optimistic at this stage.

The correlation index fell by nearly 300 points, and the heat of the big model dropped.

Not long ago, the statistics of Similarweb, a network analysis company, showed that in the first five months of 2023, the global traffic volume of ChatGPT increased by 131.6%, 62.5%, 55.8%, 12.6% and 2.8% respectively, and the growth rate dropped significantly. In June, the number of visits to ChatGPT decreased by 9.7% month-on-month, which was the first time since its launch. This set of data has aroused people’s concerns and discussions about the bubble risk of the AI industry.

In the A-share market, the concept of ChatGPT (886031) continued to decline after falling more than 5% for two consecutive days on June 21st and 22nd. By July 12th, the ChatGPT concept index had dropped to 1396.33, which was nearly 300 points lower than the high of 1688.64 on June 20th.

Many industry experts believe that the overall maturity of domestic AGI technology is not enough, and the AGI industry has already appeared the clue of overheating. Once the technology fails to land effectively and the market expectation cannot be realized, the bubble may burst. In this regard, Dr. He Xiaodong, Dean of JD.COM Exploration Research Institute and President of JD.COM Science and Technology Intelligence Service and Products Department, said in an interview with Yangcheng Evening News reporter during the Shanghai Artificial Intelligence Conference (WAIC): "There have been many bubbles in the AI field. When the big model is really put into the industry, many problems that seemed unremarkable at first will emerge one by one and need to be solved before the big model can be really used. "

Is it the direction for enterprises to push the "matrix" big model to land in the industry?

He Xiaodong believes that the landing industry may be the most important direction for the big model to finally generate economic value and social value. "Now we can see that Microsoft has landed ChatGPT in Office, and in the marketing industry, many materials can be generated by AIGC, which can also improve the efficiency of marketing."

As evidenced by He Xiaodong’s cognition, WAIC exhibited more than 30 kinds of large models, and many enterprises even launched "matrix" large models, hoping to empower vertical industries.

The Spark model exhibited by Iflytek at the conference site shows its application achievements in education, medical care, office, digital employees, automobiles, finance and industry. For example, in the field of education, iFLYTEK AI learning machine under the blessing of Spark model can make AI correct compositions like a teacher; In the medical scene, based on the iterative optimization of the big model technology, iFLYTEK Medical comprehensively upgraded the post-diagnosis rehabilitation management platform, and extended the professional post-diagnosis management and rehabilitation guidance to the outside of the hospital; In the automobile scene, the interactive experience of the intelligent cockpit based on the Spark cognitive model is demonstrated.

Baidu, on the other hand, exhibited ERNIE Bot’s application in office, conference, coding and other scenes. Wang Haifeng, chief technology officer of Baidu and director of the National Engineering Research Center for Deep Learning Technology and Application, said that any application scenario that has to deal with language or program code may have a place for ERNIE Bot. "There are already many scenarios that are actively applying ERNIE Bot, such as energy, finance, education, office, media and so on."

How to get out? The vertical field will be the main battlefield

Just in early July, Tencent released "Man-Machine Symbiosis-Top Ten Trends Report of AI in the Age of Big Models", which pointed out that vertical domain application will be the main battlefield of big models. With the rapid development of generative artificial intelligence technology, it has shown brand-new commercial value in many fields. The report pointed out that the financial industry, cultural and entertainment industries and other head institutions are expected to try to introduce large models and generative AI capabilities in relatively mature scenarios within one year.

In fact, it has become the consensus of the industry to be optimistic about the big model of vertical field. Many people in the industry believe that the China big model startups that can run out are likely to be vertically integrated. That is, while making a large model at the bottom, you can find a main application scenario, collect user data and do rapid iteration, and walk on two legs, which is indispensable.

On July 13th, JD.COM’s large-scale model was launched at the 2023 JD.COM Global Science and Technology Explorers Conference and jingdong cloud Summit. As a well-known domestic enterprise, the big model launched by JD.COM has attracted the attention of the industry. It is reported that compared with the general big model, Yanxi big model integrates 70% general data and 30% digital intelligence supply chain raw data, which has the advantages of "higher industrial attributes, stronger generalization ability and more security" and is committed to solving real industrial problems in knowledge-intensive and task-oriented industries such as retail, logistics, finance, health and government affairs.

It is noteworthy that JD.COM’s release of the big model coincides with the "squeezing bubble" stage of the domestic big model. While emphasizing its industrial application ability of the big model, it also released its own "three-step" plan for the practical application of the big model: At present, jingdong cloud has built a general big model based on internal practice; By the end of this year, JD.COM will undergo large-scale tempering through highly complex scenes and iterate out solid industrial services; It is expected that in early 2024, large-scale model capabilities will be opened to serious external business scenarios. At present, JD.COM has reached the second step, and has achieved rich practical results internally.

"In a sense, our big model is equivalent to both general education and four years of professional undergraduate education." He Xiaodong said that the difficulty of the big model is not technological catch-up, but industrial breakthrough.

Extended reading

What should AI do?

Relevant management measures will be implemented next month.

On July 13th, the National Network Information Office, together with the National Development and Reform Commission, the Ministry of Education, the Ministry of Science and Technology, the Ministry of Industry and Information Technology, the Ministry of Public Security and the State Administration of Radio, Film and Television, promulgated the Interim Measures for the Administration of Generative Artificial Intelligence Services (hereinafter referred to as the Measures), which will take effect on August 15th, 2023.

The relevant person in charge of the National Internet Information Office said that the "Measures" were issued to promote the healthy development and standardized application of generative artificial intelligence. In recent years, the rapid development of generative artificial intelligence technology has brought new opportunities for economic and social development, but also brought about problems such as spreading false information, infringing on personal information rights and interests, data security and prejudice discrimination. How to coordinate the development and security of generative artificial intelligence has attracted the attention of all parties. The promulgation of the "Measures" is not only an important requirement to promote the healthy development of generative artificial intelligence, but also a realistic need to prevent the service risks of generative artificial intelligence.

It is worth mentioning that the release of the Measures is regarded as a major positive in the field of artificial intelligence by people in the industry. On July 14, the day after the publication of the Measures, the ChatGPT concept index ushered in the biggest single-day increase since June 21, and 79 of the 87 constituent stocks ushered in an increase that day.

The reporter found out that the "Administrative Measures for Generative Artificial Intelligence Services (Draft for Comment)" released in April this year laid more emphasis on restrictions on information security and data processing, and optimized the safety supervision of generative artificial intelligence technology. The latest "Measures" are to further propose various measures for technological development and encourage applications, innovations and collaboration among multiple industries.

Specifically, the Measures encourage the innovative application of generative artificial intelligence technology in various industries and fields. The Measures also propose to promote the construction of a generative artificial intelligence infrastructure and a public training data resource platform. Promote the collaborative sharing of computing power resources and improve the utilization efficiency of computing power resources. Promote the orderly and open classification of public data and expand high-quality public training data resources. Encourage the use of safe and reliable chips, software, tools, computing power and data resources.

The relevant person in charge of the National Internet Information Office pointed out that the development and governance of generative artificial intelligence services need the participation of the government, enterprises, society, netizens and other parties to jointly promote the healthy development of generative artificial intelligence and make generative artificial intelligence technology better benefit the people.