Will China’s approval of the release of AI chatbots spur innovation?

Will China’s approval of the release of AI chatbots spur innovation?

Credit: Nikkei Asia

The race to deliver China’s answer to ChatGPT is entering its next stage: the application war. But regulatory constraints and other structural hurdles threaten to hold back progress even as technology players in the West surge ahead.

In the past week, several big tech companies, including ByteDance, Baidu and the U.S.-sanctioned facial recognition firm SenseTime, launched their own chatbots to the public after winning approval from the authorities.

Levels of enthusiasm vary. Baidu has been “all in on AI” since ChatGPT, the chatbot developed by Microsoft-backed OpenAI, took the world by storm. ByteDance, by contrast, released its Doubao chatbot in app stores without much fanfare, and according to an executive, the company “views it simply as a new product.”

Tencent Holdings, which operates the WeChat super app, has taken a relatively slower approach in launching its large language model, or the foundation on which generative artificial intelligence applications are built. The company plans to reveal its Hunyuan model on Thursday, people familiar with the matter told Nikkei Asia. The company hinted that such an announcement was coming in a social media post on Wednesday, but did not mention Hunyuan explicitly.

Alibaba Cloud, meanwhile, only said the company “has completed the filing process for its large language model” and is “waiting for the official launch.”

Unlike their U.S. rivals, Chinese companies can only offer their generative AI services to a restricted number of users until they obtain approval to roll them out to the general public. With an increasing number of users now trying out these services, their input is expected to contribute to the refinement and improvement of these products.

In the U.S. and Europe, by contrast, generative AI tools catering to various industries are already widely available, and new ones are constantly being developed. There are applications for health care, tourism, marketing and advertising, finance and manufacturing, to name just a few areas. The increasing demand for AI is also creating a windfall for U.S. chipmaker Nvidia, whose second-quarter revenue more than doubled on the year to $13.51 billion.

U.S. tech companies are also continuing to build new versions of their AI models and further refine their algorithms. OpenAI released its latest large language model, GPT-4, in March.

Baidu, the first Chinese tech company to announce it was building China’s answer to ChatGPT, said last month that the next version of its own generative large language model, Ernie 4, will be out by the end of this year.

Training a large language model requires a significant amount of computing power, advanced algorithms and large data sets. That last part is particularly sensitive in China, as chatbots that return answers the Communist Party doesn’t like could lead to severe consequences for companies.

The Provisional Administrative Measures of Generative Artificial Intelligence Services, which stipulate generative AI service providers must uphold the values of socialism, came into effect on Aug. 15.

One key part of the provisional measures, originally stating that service providers should be responsible for the legitimacy of the data sources used in pre-training their models, was modified in the final version to state that they should carry out pre-training in accordance with the laws. This alteration partially addressed one of the biggest concerns raised by tech companies, as they believed it would be impractical for them to ensure the “legitimacy” of the training data sets, a person close to China’s top cyberspace regulator told Nikkei Asia.

Some state media see business opportunities in the regulation. Three leading tech companies told Nikkei that the People’s Daily, the official newspaper of the Chinese Communist Party, approached them to sell a set of “safe text archives,” priced at 3 million yuan ($410,000), to use in training large language models, as its reports are commonly considered the most politically correct in the country.

A source at one of the companies said it declined to purchase the package due to the “ridiculously” high price and because there was no guarantee that it would provide 100% safety.

The People’s Daily did not immediately respond to Nikkei’s request for comment.

Meanwhile, more comprehensive legislation is expected to come as China’s State Council, or the cabinet, is set to submit a draft artificial intelligence law for review by the country’s top legislative body this year.

Shawn Yang, managing director at Shenzhen-based Blue Lotus Capital Advisors, said the string of approvals for big companies to launch their AI chatbots implies the next battleground will be at the application level, such as tailor-made plugins for specific sectors like finance and health care.

“This industry is still in its relatively early stages and requires enough people to try it out in order to continue optimizing it,” Yang said. “Previously, people used to believe that AI was capable of everything, but now the public may only perceive it as capable of handling some document processing and translation. Therefore, companies need to explore more application scenarios.”

To reach the Innovate Tech Show editorial team on your feedback, story ideas and pitches, contact us here.

Leave a reply

Your email address will not be published. Required fields are marked *