人工智能涉及到的技术-中国人工智能大模型地图研究报告：参数10亿规模以上

在2023中关村论坛上发布的《中国人工智能大模型地图研究报告》显示，当前，中国人工智能大模型正呈现蓬勃发展态势。据不完全统计，截至目前，参数在10亿规模以上的大模型全国已发布了79个。

A visitor experiences metaverse digital virtual human at the exhibition of the 2023 Zhongguancun Forum in Beijing, May 28, 2023. [Photo/VCG]

China has developed at least 79 large-scale artificial intelligence models with over 1 billion parameters each, a research institute said in a rare public statement, amid the worldwide buzz created by OpenAI's artificial intelligence chat bot ChatGPT.

在OpenAI的人工智能聊天机器人ChatGPT引起全球热议之际，一家研究机构发布报告称人工智能涉及到的技术，据不完全统计，截至目前，参数在10亿规模以上的大模型全国已发布了79个。

Industry experts said the United States and China have led the global development of such models, but China still had to narrow the gap with the US in the field.

业内专家表示，美国和中国引领了人工智能大模型的全球发展，但中国仍需缩小与美国在该领域的差距。

More than 14 provincial regions in China have contributed to the research and development of large-scale AI models.

报告显示，中国已有14个省区市在开展人工智能大模型研发。

Up to 38 are from Beijing, followed by 20 from Guangdong province, according to the latest report by the Institute of Scientific and Technical Information of China, which is affiliated with the Ministry of Technology.

科技部直属的中国科学技术信息研究所发布的最新报告显示，人工智能大模型研发主要集中在北京、广东两地，其中北京有38个大模型，广东有20个大模型。

To support the endeavor, local authorities are offering public computing power to meet the rising computing demand from large-scale AI models.

为了支持人工智能大模型研发，各地正在提供公共算力，以满足人工智能大模型不断增长的算力需求。

Beijing and Shanghai municipalities, as well as Guangdong and Zhejiang provinces, currently have the largest number of such models, and they are also the regions that have purchased the most AI servers in the past three years, the report showed.

报告显示，北京、广东、浙江、上海等地的大模型数量最多；同时这些地区也是近三年人工智能服务器采购数量最高的地区。

"The development of China's large-scale AI models is booming, with several technical routes making breakthroughs at the same time," said Zhao Zhiyun, head of the research institute, during the just-concluded Zhongguancun Forum.

中国科学技术信息研究所所长赵志耘在刚刚结束的中关村论坛上表示：“中国人工智能大模型研发正在蓬勃发展，各种技术路线都在并行突破。”

The US and China have led the global development of such models and contributed to 80 percent of the large-scale AI models across the world, she said, adding that China has to narrow the gap to catch up with the US.

她说，从全球已发布的大模型分布来看，中国和美国大幅领先，超过全球总数的80%。中国必须缩小差距才能赶上美国。

For instance, for China's large-scale AI models, natural language processing is still the most active field of research and development, followed by the multimodal field, she said.

例如，对于中国的人工智能大模型，自然语言处理仍然是最活跃的研发领域，其次是多模态等方面。

"But there are few models in the fields of computer vision and intelligent speech, and there is declining cooperation between universities and research institutes with companies," she said.

她说：“但在机器视觉和智能语音领域人工智能大模型很少，学术界与产业界之间的联合研发仍不足。”

"More efforts are needed to boost the overall planning and coordination of resources and accelerate basic research and technological innovation, so as to actively participate in global AI governance to further promote the orderly development of large models," Zhao added.

赵志耘补充说：“未来，我国还需加强资源和研发力量统筹，加快基础研究和技术创新，积极参与全球人工智能治理,以进一步促进大模型有序发展。”

According to the report, half of China's large-scale AI models are open source, meaning that they are made freely available.

根据该报告，中国已有超过半数的大模型实现开源人工智能涉及到的技术，可供免费使用。

Kai-Fu Lee, former head of Google China and CEO of Chinese investment firm Sinovation Ventures, told China Daily: "Open source should be supported but Chinese companies cannot rely too much on open source.

谷歌大中华区前总裁、创新工场首席执行官李开复在接受《中国日报》采访时表示：“开源应该得到支持，但中国企业不能过于依赖开源。”

"It is necessary to establish their own intellectual property and technological advantages as soon as possible to form a moat, or a barrier, because the open source model cannot reach the performance of the self-developed model of foreign manufacturers. In other words, the ability of open source models will become a ceiling that limits Chinese companies."

“必须尽快建立自己的知识产权和技术优势，形成护城河，或者壁垒，因为开源模型无法达到国外厂商自主研发模型的性能。换句话说，开源模型的能力将成为限制中国公司的天花板。”

"At the same time, the open source technologies of major overseas manufacturers have the risk of being shut down. Moreover, due to differences in culture, user habits, and laws and regulations, it is risky to bring a model trained abroad to China for fine-tuning."

“与此同时，海外主要制造商的开源技术面临被关闭的风险。此外，由于文化、用户习惯和法律法规的差异，将在国外训练的模型带到中国进行调整是有风险的。”

Dai Qionghai, an academician at the Chinese Academy of Engineering, said China has strong AI applications but lacks in innovation compared with the US. The country should also boost AI talent training, Dai added.

中国工程院院士戴琼海表示，与美国相比，中国拥有强大的人工智能应用，但缺乏创新。戴琼海补充说，国家还应该加强人工智能人才的培训。