China ‘needs to focus on boosting domestic AI sector to close gap on US’ as export curbs on chips take their toll

LLMs are the deep-learning AI algorithms behind chatbots such as ChatGPT that can recognise, summarise, translate, predict, and generate content using very large data sets.

Anhui-based iFlytek is one of the major Chinese firms that have launched their own chatbots, but its chairman admitted in October that Chinese LLMs have to bridge a “real gap” between its product and its US rival, but hoped his firm could become more competitive by the middle of the year.

Export curbs on semiconductor chips, which are used to process the LLMs, are one of the major hurdles to Chinese generative AI development.

The United States has banned companies like Nvidia and AMD from selling their most powerful chips and the equipment used to manufacture them to China in 2022, citing national security concerns, and has been pushing its other countries to follow its lead.

As a result of this push by Washington, Chinese firms should not only focus on LLM breakthroughs, but also how to develop new hardware and software, Zeng Yi, a Chinese Academy of Sciences professor who also heads the China Electronics Corporation, told the panel.

“I believe that artificial intelligence is still the key to the global industrial chain trend. But we must also be prepared to not rely on these technologies in hardware infrastructure, which may be decoupled from us,” Zeng said.

01:45

Chinese AI-generated cartoon series broadcast on state television

The US and China are locked in intense competition to dominate the AI field and China is perceived to have a growing lead in areas such as facial recognition and autonomous driving.

But generative AI has been transformed since OpenAI – the ever-first human-like chatbot that was trained with large datasets covering topics from history to computer codes – launched ChatGPT in November 2022.

However, OpenAI’s introduction of text-to-video generator Sora in February has once again stretched the gap, focusing attention on China’s competitiveness in areas such as computing power that enables AI models to be “trained” using large amounts of data.

Chinese scientists say AI model can mark the best spots for solar panels

Speaking after another AI panel in the forum, Yuan Hui, the founder and chief executive of AI Chatbot firm Xiao-I, said China has been strong in the “application” of AI technologies since the beginning of mobile internet, but it needs to double down on developing the “core and foundational” technology.

Christopher Thomas, a Beijing-based senior fellow at Brookings Institution, echoed this in another panel on AI governance on Thursday, saying: “If you think of the ability to drive artificial intelligence as the ability to control and manage computing, computing is essentially dominated by the United States, with China as a very distant second.”

In a world “where America will do all of its own thing, and China will do all of its own thing … we’re going to waste our money on duplication, rather than solving problems”, he said.

But Yuan was optimistic about a breakthrough. He said no single country or company could fully and independently “own” artificial intelligence as their competitors will develop similar or more advanced chips whenever new products appear.

Liu Cong, the vice president of Chinese AI firm iFlytek, said there was a gap with the US that China needed to close. Photo: Handout

Besides the US chip ban, China also faces domestic challenges such as the huge costs of computing and censorship .

Zeng said a risk averse approach is the biggest challenge to China’s AI development, and questioned whether people were willing to challenge the majority view.

‘’I think the biggest challenge of artificial intelligence in China now is that we may not be good at taking greater risks. In the process of exploring the unknown, we are not willing to pay more costs …. this is the biggest crux of the development of artificial intelligence in China.”

South China Morning Post