, "By using our Kunlun chips in large language models, text and image recognition tasks on our AI platform are 40% more efficient and the overall cost is reduced by 20% to 30%."

Chinese tech companies rush to match Stable Diffusion and DALL-E 2, but tough times lie ahead

The huge technological leaps machine learning models have demonstrated over the past few years have everyone excited about the future of artificial intelligence—but also nervous about its troubling consequences. After Stability AI and OpenAI's text-to-image tools became hot topics, ChatGPT's ability to conduct intelligent conversations has become a new favorite in all walks of life.

OpenAI CEO tweeted warning of clear limitations of AI tool ChatGPT

China’s tech community has been closely watching progress in the West, with entrepreneurs, researchers and investors looking for ways to make progress in the field of generating artificial intelligence . Tech companies are designing tools based on the open source model to attract both consumer and enterprise customers. Individuals are profiting from AI-generated content. A series of regulatory regulations responded quickly, defining how text, image and video compositions could be used. At the same time, U.S. technology sanctions have raised concerns about China's ability to keep up with advances in artificial intelligence.

As generative AI takes the world by storm in late 2022, let’s take a look at how this explosive technology is making waves in China.

Chinese characteristics

Thanks to “artistic” creation platforms like Stable Diffusion and DALL-E 2, generative AI is suddenly on everyone’s lips. China's tech giants have also captivated the public with similar products, adding novelties to cater to Chinese tastes.

Baidu , which rose to prominence among search engines, has been stepping up its game in autonomous driving in recent years, operating ERNIE-ViLG, a 10 billion-parameter model trained on a dataset of 145 million Chinese image-text pairs. How does this model compare to its American counterparts? The following are the results of searching "kids eating shumai in New York Chinatown" in the Stable Diffusion model, and the results of searching (children eating shumai in New York Chinatown) in the ERNIE-ViLG model.

Stable Diffusion

The results are arguably a tie for the two results above, neither eating the right siomai, and the quick test reflects the difficulty of capturing cultural nuances when the datasets used have inherent biases - assuming stable diffusion there will be more data on the Chinese diaspora, and ERNIE-ViLG may have been trained on a wider variety of siomai images that are rarer outside of China.

Another Chinese tool that has caused a stir is Tencent 's two-dimensional photo generator: "Me in Another Dimension" , which can turn photos of people into anime characters, but the artificial intelligence generator shows its own bias: It is aimed at Chinese users and has become unexpectedly popular in other anime-loving regions such as South America. But users quickly realized that the platform was unable to identify black and plus-size individuals, groups that are conspicuously absent from Japanese anime, resulting in objectionable AI-generated results.

In addition to ERNIE-ViLG, another large-scale Chinese text-to-image model is Taiyi, which is the brainchild of IDEA, a research lab led by renowned computer scientist Harry Shum, who co-founded Microsoft Research, the largest research arm of Microsoft outside the United States. The open-source AI model is trained on 20 million filtered Chinese image-text pairs with 1 billion parameters.

Unlike Baidu and other profit-oriented technology companies, IDEA is one of the few institutions supported by local governments in recent years to engage in cutting-edge technology research. This means the center may enjoy more research freedom without the pressure to drive commercial success. Headquartered in the tech hub of Shenzhen and backed by one of China's wealthiest cities, this is an up-and-coming company worth keeping an eye on.

AI Rules

China’s generative AI tools are distinguished not only by the fact that they learn from domestic data, they are also governed by local laws.As MIT Technology Review points out, Baidu’s text-to-image model filters out some keywords, which is to be expected.

Even more important to the future of this emerging field is a new set of regulations targeting so-called "deep synthesis technology," which refers to "the use of deep learning, virtual reality and other synthesis algorithms to generate text, images, audio, video, and virtual scenes." Like other types of Internet services in China, from games to social media, users are using generated AI Apps have previously been asked to verify their names, and the fact that prompts can be traced back to a person's real identity inevitably has a restrictive effect on user behavior.

Deepfake performs face forgery

But on the bright side, these rules may lead to more responsible use of generative AI, which has been abused elsewhere to create NSFW and sexist content. For example, Chinese regulations explicitly prohibit people from generating and spreading artificial intelligence-generated fake news. However, how this is implemented depends on the service provider.

“It’s interesting that China is at the forefront of trying to regulate [generative AI] as a country,” said Yoav Shoham, founder of AI21 Labs, an Israel-based OpenAI rival, in an interview. “There are various companies that are putting limits to AI… Every country I know of has efforts to regulate AI or to somehow make sure that the legal system, or the social system, is keeping up with the technology, specifically about regulating the automatic generation of content.”

“What’s interesting is that China is at the forefront of trying to regulate [generative artificial intelligence] as a country,” Yoav Shoham, founder of Israel-based OpenAI competitor AI21 Labs, said in an interview. "There are all kinds of companies that are limiting AI... Every country that I know of is trying to regulate AI or in some way ensure that the legal system or the social system keeps up with the technology, especially when it comes to regulating automation."

But there is no consensus yet on how this rapidly changing field should be managed. "I think this is an area where we are learning together," he said. "It has to be a collaborative effort. It has to involve technology experts who really understand the technology and what it can and can't do, the public sector, social scientists, people affected by the technology, and government, including the business and legal sectors." 7Work Together."

Artificial Intelligence Monetization

As artists worry about being replaced by powerful artificial intelligence, many in China are leveraging machine learning algorithms to make money in a variety of ways, and they do not come from the most tech-savvy population. Instead, they are opportunists or part-time workers looking for additional sources of income. They realized that by improving the tips, they could make the AI ​​produce creative emojis or stunning wallpapers, which they could post on social media to increase ad revenue or charge for downloads directly, and the truly skilled would also sell their tips to others who wanted to join the money-making game—or even train them for a fee.

Like the rest of the world, workers in China are using artificial intelligence in their formal jobs. For example, light novel writers can cheaply create illustrations for their works, a genre that is shorter than novels and often features illustrations. One interesting use case that could disrupt the manufacturing sector is the use of AI to design prints for T-shirts, manicures, and other consumer products. By quickly generating high-volume prototypes, manufacturers can save design costs and shorten production cycles .

It is too early to know how the development of generative AI differs in China and the West. But entrepreneurs have already made decisions based on their early observations. Some founders told me that businesses and professionals are generally happy to pay for AI because they see a direct return on investment for versus , so startups are eager to pioneer industry use cases.One clever application came from Sequoia China-backed Surreal (later renamed Movio) and Hillhouse Capital-backed ZMO.ai, two companies working on algorithms to generate fashion models of all shapes, colors and ethnicities.

But some entrepreneurs are not convinced that their artificial intelligence products will achieve the same skyrocketing valuations and breakneck growth that their Western counterparts, such as Jasper and Stability AI, are enjoying. Over the years, countless Chinese startups have had the same concern: China's corporate customers are generally not as willing to pay for SaaS as those in developed economies, which is why many of them are starting to expand overseas.

China SaaS Competition in the field is also fierce. "In the U.S., you can do well by building product-led software that doesn't rely on human service to acquire or retain users. But in China, even if you have a great product, your competitors can overtake you overnight by stealing your source code and hiring dozens of customer support staff who don't cost much," said the founder of a generative AI startup in China.

Shi Yi, founder and CEO of sales intelligence startup FlashCloud, agrees that Chinese companies often prioritize short-term returns over long-term innovation. "In terms of talent training, Chinese technology companies tend to focus more on applying skills and making money quickly," he said. One Shanghai-based investor, who asked not to be named, said he was "a little disappointed that this year's big breakthroughs in generative AI are all happening outside of China."

Hurdles Ahead

Even if Chinese tech companies want to invest in training large neural networks, they may lack the best tools. In September 2022, the United States implemented export controls on high-end artificial intelligence chips to China. While many Chinese AI startups focus on applications that don’t require high-performance semiconductors to process massive amounts of data, for those doing basic research, using less powerful chips means longer and more expensive computations. The good news is that U.S. sanctions are driving Chinese investment in advanced technology in the long run.

Baidu Executive Vice President and Head of Artificial Intelligence Cloud Group said that as a company that bills itself as a leader in China's artificial intelligence field, the impact of U.S. chip sanctions on its artificial intelligence business will be "limited" in the short and long term because "a large part" of Baidu's AI cloud business "does not rely too much on highly advanced chips." In the case that it does need high-end chips, it "actually has enough inventory to support our business in the near term."

What about the future? "In the medium to long term, we actually have our own developed AI chip, named Kunlun .", "By using our Kunlun chip in a large language model, the efficiency of performing text and image recognition tasks on our AI platform has increased by 40%, and the total cost has been reduced by 20% to 30%."

Time will tell whether Kunlun and other local artificial intelligence chips will give China an advantage in the race to generate artificial intelligence.