LLMs using domestic chips need of hour

There is an urgent need to encourage research and development of large language models trained on domestic chips, in order to build a robust artificial intelligence ecosystem that will ensure sustainable and high-quality development in the AI era, said Liu Qingfeng, a deputy to the 14th National People's Congress, the country's top legislature.

Liu, who is also chairman of Chinese AI company iFlytek, suggested efforts be made toward providing special financial support to enterprises that are developing domestic AI chips and those that use domestic chips for large language model training.

He also called for more support to encourage State-owned enterprises in prioritizing the procurement of AI models developed on domestic chips and promote industry-specific vertical applications based on these AI models.

According to Liu, the domestic computing software ecosystem remains weak, with incomplete supporting tools such as open-source training frameworks, and development platforms. Apart from the company's AI model Spark, all other publicly downloadable large language models are trained on US company Nvidia's chips.

Failing to develop an AI industry ecosystem based on domestic chips is akin to "building a skyscraper on someone else's foundation," Liu cautioned.

He also urged greater efforts to leverage China's extensive AI application scenarios to actively apply large language models in industrial fields, forming a data flywheel that will make the nation the first to reap the benefits from AI in industrial applications.

After China unveiled a new-generation AI development plan in 2017, Liu said the country has accumulated technical reserves and organized teams in the field of cognitive intelligence, making it an important player in the global AI arena.

However, with the emergence of generative AI technology, global competition has intensified and China lags behind the United States in some aspects, he said. Generative AI refers to computer algorithms that produce new text, images, code, videos and audio in a human-like fashion. It is the key technology behind ChatGPT and Sora.

Goldman Sachs Research has forecast in a report that breakthroughs in generative AI can drive a 7 percent, or almost $7 trillion, increase in global GDP and raise productivity growth by 1.5 percentage points over a 10-year period.

Well aware of the opportunities ahead, established tech heavyweights such as Alibaba, Tencent, Baidu, ByteDance, iFlytek and Huawei, as well as thousands of startups in China, are scrambling to develop and embrace large language models.

Chi Xiannian, a senior engineer at the China Center for Information Industry Development, a think tank affiliated with the Ministry of Industry and Information Technology, said finance, manufacturing, governance and transportation were the top industries in China using AI large language models.