Recently, the first batch of large model products filed through the "Interim Measures for the Management of Generative Artificial Intelligence Services" in China have been launched, officially open to the whole public service, large model technology refers to the use of large-scale neural network models, language-based interaction to complete various artificial intelligence tasks of technology, at present, large model technology in the world rapid iteration, has ChatGPT and other applications. So, what is the current status of the development of the domestic large model market?
"Hundred Model Wars" races Domestic artificial intelligence large models are launched centrallyIn Beijing, a start-up enterprise that independently develops large models, the large model application launched has been open to the public for nearly 10 days, and the staff is still optimizing and improving the content of large model generation, and a large number of questions and answers are currently focused on text creation, education and other fields.
Zhang Peng, CEO of an artificial intelligence large model company: It is a large language model itself, so the ability in language is stronger, we estimate that it is equivalent to the level of a bachelor's graduate, and the ability of mathematics and science is close to the middle school level. We will do an upgrade to the model once in a few months.
According to statistics, 11 large models in China have successively passed the "Interim Measures for the Management of Generative Artificial Intelligence Services" for the record, of which 5 in Beijing, 3 in Shanghai are the first to go online, 2 in Guangdong and 1 in Anhui are also opening one after another.
The reporter learned that the latest domestic online general large model is more like an infrastructure, on which more vertical models in professional fields can be incubated to serve the digital transformation of all walks of life.
Industry insiders said that the current investment cost of the general large model is very high, and it needs to be effectively combined with specific application scenarios, and the related industry is still in the early stage of development.
Nie Junfeng, researcher of the China Banking Research Center of the Central University of Finance and Economics: We are in the stage where the market is racing, industry rules have not yet been established, and the market pattern has not yet been fully formed. People call it the "100-model war", and from last year to now, there have been nearly 100 financing cases in the large model industry in China. From the perspective of the investment valuation of the entire industry, the new investment in this industry will be about 15 billion yuan to 20 billion yuan in 2023.
Large model talent is scarce, salaries are high The market for hash rate algorithms is activeThe heating up of the large model market is also driving the upstream and downstream industry chain, including data, computing power and algorithms, and talents in related fields, such as large model engineers, are also in short supply in China, and income levels continue to rise.
Liu Xinyang is a large model engineer, his main job every day is to train large models in various industries, before that, he was engaged in the development of search engines. In 2022, he transitioned to developing large-model technology.
Liu Xinyang, a large model engineer of an artificial intelligence enterprise: The large models I trained include the financial industry, the real estate industry, the medical industry, and water conservancy. After all, large models are a newly generated technology, and it needs to be trained to be available.
According to statistics from relevant institutions, the average annual salary of domestic artificial intelligence engineers and algorithm scientists has exceeded 600,000 yuan, and enterprises have a strong demand for large model technology research and development talents, and there are large gaps in some core fields.
Liu Huifu, chief technology officer of an information technology group: Experts who really understand the industry are very much needed, and those who understand artificial intelligence, algorithms, and models are also very needed, because the result of each round of artificial intelligence development is to make those who are good at using tools better survive.
The heating up of the large model market is also driving the computing power algorithm industry, including "big model" chips, computing clusters, servers, communication network hardware, and software such as cloud computing, databases, and virtualization functions. Computing power infrastructure and computing power services are indispensable for large model training, and their performance, quality and stability determine the model training effect.
In the international market, the price of a chip dedicated to training artificial intelligence models has even been speculated to 45,000 US dollars, equivalent to more than 300,000 yuan. The independent research and development of chips in related fields in China is also accelerating.
Liang Zheng, deputy dean of the Institute of International Governance of Artificial Intelligence at Tsinghua University: Computing power is a relatively high threshold for the landing of large models, and the country at this level has been making overall deployment, that is, the construction of our national computing power network, there are some computing centers in various places, which may have been more CPUs in the past, and now it is plus intelligent computing GPUs.
(Client)