BEIJING (Reuters) – Chinese organizations have launched 79 large language models (LLMs) in the country over the past three years as they ramp up efforts to develop artificial intelligence (AI) algorithms, a report says. public research institutes.
The development of LLMs, trained using deep learning techniques on massive amounts of textual data, entered an “accelerated” phase in 2020, according to the report compiled by research institutes run by the Ministry of Foreign Affairs. Science and Technology of the country.
In 2020, Chinese organizations issued 2 LLMs, compared to 11 in the United States, but in 2021 there were a total of 30 LLMs issued in each country, according to the report released on Sunday.
Overall U.S. organizations issued 37 LLMs the following year, compared to China’s 28, according to figures compiled in the report, whose co-authors include the Institute of Scientific and Technical Information of China.
So far this year, China leads with 19 LLMs compared to 18 in the United States.
“Judging from the distribution of large-tongue models released around the world, China and the United States lead by a large margin, accounting for more than 80% of the global total,” the report concludes, according to a statement from press which summarizes the main conclusions of the report.
“The United States has consistently ranked first in the world in terms of the number of large-tongued models.”
The report comes as the country’s AI industry faces significant challenges as US-led export controls prevent Chinese organizations from accessing semiconductors used to train LLMs. , among other advanced computing tasks.
The report analyzed the 79 LLMs developed in China, noting that while there were already 14 provinces and regions where such technology was developed, joint development projects between universities and industry were “insufficient”.
Following the release of OpenAI from ChatGPT, Chinese tech giants from Alibaba to watchdog Sensetime and search engine giant Baidu launched their own versions of chatbots powered by generative AI and LLMs. .
(Reporting by Qiaoyi Li and Eduardo Baptista; Editing by Brenda Goh and Sriraj Kalluvila)