State of AI in China: Navigating Challenges and Seizing Opportunities
In the wake of ChatGPT’s global impact, China’s tech landscape has been bustling with activity as tech giants and AI innovators strive to match or even surpass their American counterparts.
Chinese tech giants such as Baidu, Alibaba, and ByteDance, along with smaller but prominent AI players like iFlytek and SenseTime — not to mention several university labs — are fervently developing their versions of AI chatbots. This surge of activity underscores China’s ambition to be a front-runner in the AI domain, despite facing significant hurdles.
Global and Domestic Constraints
China’s AI development has encountered notable challenges, primarily stemming from external bans and internal policies.
The US Commerce Department’s 2022 chip ban significantly hindered Chinese companies from accessing cutting-edge Nvidia A100 or H100 GPUs, essential for efficiently training their AI models. Despite these restrictions, Chinese tech giants remain the leading hoarders of AI chips globally. This status is likely the result of acquisitions made before the ban, or through smuggling via the customs of Macau.
OpenAI has heavily restricted users from China from accessing their products. The restrictions led to a Chinese ‘black market’ for ChatGPT accounts, with many, including students and researchers, seeking unauthorized access to this tool. Providers of these accounts use cryptocurrencies and virtual credit cards to bypass the restrictions, charging their Chinese customers an extra US$5 per account on top of the $20 subscription fee. An account with GPT-4 API access could cost up to $2,200 on the Chinese black market when availability was limited in early 2023.
On the regulatory front, China introduced one of the world’s first generative AI regulations to control sensitive content visible to the public. This has placed a burden on Chinese companies to self-regulate the content produced by their bots. For example, anything related to Xi Jinping could not appear in their chats. A company in Shanghai was fined for cloaking ChatGPT as its chatbot through API integration.
LLMs with Chinese Characteristics
Amid these challenges, Chinese tech giants are finding unique ways to thrive.
Baidu, with its Ernie bot, has emerged as a leader in the AI chatbot space within China. Baidu claimed that its Ernie bot had amassed 100 million users as of December 2023, four months after it was authorized to begin offering services to the public.
The company’s robust AI infrastructure, significantly bolstered by Andrew Ng’s tenure as AI lead from 2014 to 2017, has positioned it as a formidable competitor to global players like Google.
In building the Ernie Bot, Baidu’s strategy focuses on mastering Chinese language services, employing top university graduates and industry experts to refine content across various domains.
American companies like OpenAI take advantage of the lower wages in some third-world economies to outsource their data labeling tasks, providing only $170 per month per person in Kenya. In contrast, in Beijing, Baidu is offering a budget of $2,800 per person to fine-tune their corpus for training their bots.
Additionally, by prioritizing sophisticated tasks such as academic paper writing, poetry composition, answering questions related to Western literary history, and generalizing these data to other tasks, Baidu aims to outperform ChatGPT in Chinese language applications.
The English part of the corpus of their bot still heavily relies on materials from ChatGPT, a Baidu recruiter admitted. “But we are sure our (Chinese data labeling) team will be necessary as long as the project (Ernie Bot) is here.”
Emerging Small Startups in AI’s Year One
The landscape extends beyond the reach of tech giants; smaller startups are carving out their niche by harnessing the capabilities of LLMs for innovative applications in China.
HeyGen, a Shenzhen-founded startup now headquartered in Los Angeles, specializing in computer vision, gained attention with a video showcasing a digitally manipulated Taylor Swift speaking Chinese on a talk show. Despite branding itself as an American company, HeyGen has still secured clients like China Daily, a state media outlet.
Utilizing existing video material and AI technology to lip-sync the script, HeyGen’s product enables this propaganda agency to automatically produce content featuring its hosts and journalists speaking English without a detectable accent. This also eliminates the need for them to record any new footage.
In China, AI ventures are colloquially termed ‘AIGC,’ an acronym for AI-generated content, emphasizing the content-centric nature of China’s AI startup landscape. However, many are expanding into other AI use cases beyond merely providing apps for modifying articles or photos.
In the SaaS sector, AI is leveraged for internal training. Some Chinese companies are beginning to feed internal documents to LLMs to quickly familiarize new employees with company policies and codes of conduct. AI-related courses, such as those teaching how to register and use these chatbots, sold well on various podcast platforms.
Navigating the US-China AI Tug-of-War
The 2023 AI boom has yet to spark new unicorns in China. Investors in the country’s venture capital and private equity scene remain cautious due to the sluggish macroeconomic recovery following China’s reopening at the end of 2022.
HeyGen’s founders, the China-born, US-educated Joshua Xu and Wayne Liang, along with many other Chinese AI entrepreneurs, have partially relocated their operations to the United States. This shift aims to avoid harsh regulatory environments in China or attract investments from American VCs.
Companies and entrepreneurs like HeyGen have chosen to base themselves overseas, particularly in America, seeking a better investment environment and the opportunity to tap into superior AI talent.
Conversely, while some companies seek opportunities abroad, others within China’s AI industry commit to domestic development and innovation, in line with the nation’s push for technological self-reliance.
As Kai-Fu Lee, a former scientist at Microsoft and Google who founded 01.AI, stated in a speech, “Self-innovation is an essential path for the development of large AI models in China.”
A bit about the Author
As a political analyst, Roger offers political risk advisory to tech companies and policy recommendations to governments, with a research focus on utilizing natural language processing techniques to dissect political information.
Thank you for reading! The ML Practitioner is edited and curated by Livia Whitermore.
Don’t miss out on cutting-edge ML trends, tips, and discussions. Become part of an exclusive community of forward-thinkers. Stay connected and advance your knowledge with our weekly insights.
🚀 [Subscribe to Our Newsletter] 🚀
Connect and contribute! Find us on LinkedIn to submit an article or share your thoughts.
