China is building its own AI at a rapid pace
MICHAEL ZHANG/Getty Images
- Chinese tech giants are racing to catch up with US firms like OpenAI in the AI arms race.
- Over 40 AI models have been approved for public use by the government, per local news reports.
- China has tight controls on AI , with tech companies required to "adhere to socialist values."
China's tech giants are racing to catch up in the AI arms race.
The Chinese government has approved more than 40 AI models in the past six months since it began this process, according to local media reports cited by Reuters, as tech companies seek to make up ground on US rivals like OpenAI.
That includes 14 new Large Language Models (LLMs) approved for public use in just the last week. A senior executive at Chinese tech firm Tencent previously described the country's emerging AI gold rush as a "war of a hundred models" in September.
Leading the way in this "war" is Baidu, a search engine giant sometimes referred to as "China's Google."
The company has released a ChatGPT-rival named "Ernie Bot" which, after a rocky start, it now says has over 100 million users and can go head-to-head with OpenAI's GPT-4 model.
Baidu faces competition from Apple smartphone rivals Huawei and Xiaomi, who have also invested in their own AI models, and from TikTok owner Bytedance.
The latter's AI push has already proven highly controversial. Bytedance has built an unreleased AI voice converter that researchers warn could be used for fraud, and the company was suspended from accessing OpenAI's tools in December after The Verge reported that it had been using them to build a ChatGPT competitor.
Unlike their US rivals, Chinese companies that develop their own chatbots face political as well as technological challenges.
The Chinese government requires LLMs to meet a strict set of rules before being released to the public, with regulations released last year stating that all chatbots must "adhere to core values of socialism" and not challenge state power.
That has meant that many of the country's most advanced AI models have a tendency to avoid sensitive topics, such as the status of Taiwan.
When Bloomberg quizzed a selection of China's ChatGPT rivals, including Baidu's Ernie Bot and the Tencent-backed Minimax, they either tried to change the subject when asked certain questions or refused to answer what they deemed to be "illegal" questions. When asked whether Taiwan is a country, all the bots described it as a part of China.
Taiwan, a self-governing nation that China regards as a breakaway province, has responded by announcing funding for its own AI model, named Taide, that it hopes will be free of Chinese political influence.
Ultimately, it may be geopolitical tensions that hobble China's attempts to become an AI superpower.
The US has announced a series of restrictions on the sale of advanced semiconductors crucial for building and training AI to China, with experts warning that this will hamper the ability of Chinese companies to build advanced large language models.