Advertisement
Advertisement
US-China tech war
Get more with myNEWS
A personalised news feed of stories that matter to you
Learn more
In this week’s issue of the Global Impact newsletter, we take a look at the ever-changing artificial intelligence (AI) landscape, and at what progress is being made in China as it struggles against US restrictions on hi-tech imports. Photo: Shutterstock

Global Impact: cut off from high-end chips, how much is the US tech war hurting China’s AI ambitions as Sora highlights worrying gulf?

  • Global Impact is a weekly curated newsletter featuring a news topic originating in China with a significant macro impact for our newsreaders around the world
  • In this week’s issue, we take a look at the ever-changing artificial intelligence (AI) landscape, and at what progress is being made in China
Global Impact is a weekly curated newsletter featuring a news topic originating in China with a significant macro impact for our newsreaders around the world. Sign up now!
Here we go again. Just as soon as US-based OpenAI’s Sora arrived on the scene, a new crisis of confidence struck China’s artificial intelligence (AI) community, which saw the kind of soul-searching in C-suites and among academics reminiscent of the release of ChatGPT.
But this time, instead of a quirky chatbot, Sora turns text prompts into video. The results, at least at first glance, are impressive. So impressive, in fact, that Zhou Hongyi, the founder of Chinese internet security firm 360 Security Technology, called the product’s release a“barrel of cold water poured down China’s head”.
Sora is not available to the general public yet, but early demo videos appear much more sophisticated than the output of similar products. Just as Chinese firms last year felt pressure to show their progress in large language models, some are now trying to assure investors that they have their own video generators in the works. Zhipu AI, for example, promised that its own Sora-like product would be coming later this year.

03:48

Young Chinese singles turn to AI-generated partners

Young Chinese singles turn to AI-generated partners
Some are optimistic about China’s chances. One AI entrepreneur said China would have Sora-like models coming out in a matter of months.
That is not necessarily blind optimism. Open-source models lag behind the likes of OpenAI’s GPT-4 and Sora, but they are rapidly catching up. Many Chinese products are based on Meta Platforms’ Llama 2 model, which is free to use and maybe a year behind OpenAI’s latest and greatest, by some estimates.
The open-source nature of AI makes development in China look more promising. Announced just this month, an effort from Peking University professors and Shenzhen-based AI company RabbitPre seeks to replicate Sora as an open-source model.
This environment is also promising for start-ups that are trying to make it in an industry dominated by tech giants. AIsphere, founded in Beijing by a former ByteDance executive, raised US$14 million in a recent funding round after releasing a video generator in January.

OpenAI’s Sora pours ‘cold water’ on China’s AI dreams

Much of the corporate world is still trying to feel out precisely what the benefits of AI will be and how they will be realised. With video generation, some see an opportunity to upend elements of the video-content mill feeding social media these days.

ByteDance, which owns TikTok, is mobilising resources to catch up in AI, in which it has lagged domestic rivals such as Baidu and Alibaba. CEO Liang Rubo has made recruiting AI talent a top priority, people familiar with the matter told the Post. The company recently hired a key contributor to Google’s video generator VideoPoet, and ByteDance’s website currently lists more than 300 open positions related to generative AI (GenAI).
While still nascent, this technology is already being deployed by some studios in China. State broadcaster China Media Group (CMG) and Shanghai AI Laboratory (Sail) recently teamed up on a series of 26 seven-minute episodes adapted from classical Chinese poetry called Qianqiu Shisong, which were posted to WeChat. The AI was trained on CMG videos and audio, and then used in everything from art design to video generation for the show, according to Sail.

01:45

Chinese AI-generated cartoon series broadcast on state television

Chinese AI-generated cartoon series broadcast on state television

The results look like something on par with the best online Flash animations from the early 2000s. That is enough to get people to watch, and this technology will only get better.

Still, in the global competition among tech titans, the AI quality gap between the US and China remains apparent. And for all the fresh concerns about video-generating AI models, the challenges for Chinese developers have not changed. They remain cut off from the same high-end silicon used to train any other GenAI models.
Restrictions on Nvidia’s tech seem never-ending. These restrictions are now moving into software after the American tech firm issued warnings about the use of CUDA, which enhances AI performance when run on graphics processing units, on third-party chips. The licence agreement tells developers not to use such “translation layers” for other chips – something that is increasingly important for Chinese companies trying to fill a gap left by the absence of Nvidia’s most powerful hardware.

Nuclear weapons and poison pills: Washington, Beijing warily circle AI talks

Huawei Technologies has already been chugging away at this, and its Ascend 910B is considered by some to be equal to Nvidia’s A100 in computing power.
Another avenue left for Chinese platforms, at least for now, is to just use Sora when it becomes publicly available. One Chinese fintech firm claimed that it will have priority access to Sora once it is available on Azure, Microsoft’s cloud computing platform. Microsoft uses OpenAI’s tech for its Copilot AI, and the Chinese firm, Sinodata, has joined the US tech giant’s Cloud Partner Programme.

This is a precarious position, though. Not only are there threats from Washington to prevent American AI platforms from serving Chinese customers, but Beijing has sought to limit access to models developed abroad because of the country’s tight content controls.

02:35

Taiwanese entertainer uses artificial intelligence to bring back deceased daughter

Taiwanese entertainer uses artificial intelligence to bring back deceased daughter
There are legitimate concerns about misinformation spread through AI, and countries are trying to come to some agreement on ethical AI use, but if the example of CMG proves anything, it is that the commercial benefits of products like Sora are already real. Video production is a very time-consuming process, so anything that speeds that up is likely to be a welcome advancement.
It could also just be difficult to stop companies from training their models on any data they can get their hands on. That is the take of Wikipedia co-founder Jimmy Wales. “In some ways, it might be one of the ways to break down some boundaries” between countries, he told the Post.

So, at least in this pocket of the industry, the most cutting-edge models may be more of a nice-to-have tool than an essential one – at least for now.

60-Second Catch-up

Deep dives

Illustration: Lau Ka-kuen
OpenAI’s recent text-to-video model Sora has fired a fresh warning shot to China about its gap with the world’s top artificial intelligence (AI) technologies, triggering questions about why the country has no equivalent product in an echo of the kind of soul-searching local researchers and investors went through after the 2022 launch of ChatGPT.

Just a few years ago, China had envisioned itself eventually dominating the global AI race by leveraging the country’s vast troves of data to develop mature applications for functions like facial recognition. Recent developments in generative AI – which uses large models to produce content like text, images and video – have changed the calculus, making China look like a laggard once again.

Photo: OpenAI
TikTok owner ByteDance is mobilising resources to generative artificial intelligence (GenAI) projects, as the Beijing-based tech giant doubles down on efforts to catch up with OpenAI’s conversational bot ChatGPT and text-to-video generator Sora.

Liang Rubo, who took over as chief executive from co-founder Zhang Yiming in 2021, has set out three objectives for ByteDance with regard to GenAI this quarter: strengthen the recruitment of AI talent, enhance the organisational structure, and improve fundamental research, according to people with access to internal briefings, who declined to be named because the information is private.

Illustration: Davies Christian Surya

The United States and China have a shared interest in sitting down to discuss automated weapons, artificial intelligence and its many potential and unforeseen abuses. Less clear is whether the two global AI superpowers and their huge militaries have common interests or goals coming into the talks, which are expected to take place this spring, according to analysts and experts involved in informal sessions between the two nations.

“The good news, which has been a really, really rare thing these days, is that the AI dialogue is seeing some hope,” said Xiaomeng Lu, director of Eurasia Group’s geo-technology practice, who is involved with US-China “Track 2” talks among former government officials, experts and analysts. “Both sides have an interest in preventing unintended consequences.”

Photo: Handout

Hua Mulan, a legendary heroine in ancient China who took her father’s place in conscription for the army by disguising herself as a man, is now known all over the world thanks to The Walt Disney Co’s 1998 animated motion picture Mulan and its 2020 live-action film with the same title.

Chinese people were reacquainted with the character in the 1950s, when the story was performed by well-known traditional opera artist Chang Xiangyu across the country to raise funds for a military plane for the Korean war.

Photo: Reuters
Employers in China are scrambling for talent with skills in generative artificial intelligence (GenAI), the technology underpinning a new generation of highly intelligent chatbots led by Microsoft-backed OpenAI’s ChatGPT.

Computer vision engineers with GenAI skills are being offered an average annual salary of more than 480,000 yuan (US$66,700), about two-thirds higher than the 290,000 yuan earned by their peers without such knowledge, according to a recent report published by Chinese recruitment agency Liepin.

Photo: Shutterstock
Researchers of Chinese origin comprised more than half of this year’s recipients of Apple’s annual fellowship programme on artificial intelligence (AI), shining a light on a brain drain that threatens the nation’s ambition to become a global powerhouse in that critical technology.

The 2024 Apple Scholars in AIML PhD fellowship – focused on machine learning, a branch of AI concerned with developing algorithms and statistical models for computer systems – shows that 11 of the 21 admitted to the programme were of Chinese origin, based on their names and academic background that include bachelor’s level studies on the mainland, according to a list published on Tuesday by Apple on its website.

Photo: Brilliant Labs

Singapore-based smart lenses start-up Brilliant Labs has become the latest company to introduce a pair of smart glasses, as a slew of tech companies, including several in China, try to make wearing glasses “cool”.

The company’s new Frame smart glasses, announced on Friday, are imbued with a custom artificial intelligence (AI)-powered assistant called Noa, which corrals many large language models (LLMs) to find one that is the best fit for a given query. By putting this tech into slim frames that can be worn all day, Brilliant Labs, founded in Hong Kong in 2019, is jumping on the smart eyewear trend that has also drawn in several Chinese companies looking to capitalise on a market expected to boom this year thanks to the entry of Apple with its much larger Vision Pro mixed reality headset.

Photo: Xinhua

The telecommunications industry is rushing to explore the potential of generative artificial intelligence (AI), and Chinese equipment suppliers Huawei Technologies and ZTE have joined the pack with their own industry-specific models shown at MWC Barcelona this week.

At the event, the world’s biggest trade show for the mobile industry, Huawei launched its self-developed AI model specifically for telecoms service providers.

Photo: Shutterstock
China will press ahead this year with development of large language models (LLMs) – the technology used to train ChatGPT and similar applications – and generative artificial intelligence (AI) systems, despite restrictions imposed by the United States on the mainland’s access to advanced semiconductors, according to analysts from Swiss investment bank UBS.
China is expected to “work within those constraints and try to make progress” by tapping into domestic AI accelerator programmes and being more economical in using computing resources, said Nicolas Gaudois, head of Asia-Pacific technology research at UBS, during a webinar hosted by the bank on Tuesday.
Photo: Handout

A version of this article was first published by The Korea Times in a partnership with the South China Morning Post.

In an office building in southern Seoul, a dozen chips were laid side by side on shelves, each next to their own electric fan to cool them down as they operate.

Global Impact is a weekly curated newsletter featuring a news topic originating in China with a significant macro impact for our newsreaders around the world.

Sign up now!
Post