News How China is building a parallel-generated AI world • TechCrunch
huge technology The leaps machine learning models have shown over the past few months have everyone excited about the future of AI — but also nervous about its troubling consequences. After Stability AI and OpenAI’s text-to-image tool became a hot topic, ChatGPT’s ability to conduct intelligent conversations has become the new darling of all walks of life.
In China, the tech community has been closely watching advances in the West, with entrepreneurs, researchers and investors looking for ways to make progress in the field of generative artificial intelligence. Tech companies are designing tools based on the open-source model to appeal to both consumer and business clients. Individuals are profiting from AI-generated content. Regulators responded quickly, defining how text, image and video composites can be used. Meanwhile, U.S. technology sanctions have raised concerns about China’s ability to keep up with advances in artificial intelligence.
As generative artificial intelligence takes the world by storm in late 2022, let’s take a look at how this explosive technology is making waves in China.
Thanks to viral art creation platforms like Stable Diffusion and DALL-E 2, generative AI is suddenly on everyone’s lips. On the other side of the world, Chinese tech giants have also captivated the public with their products, adding new flavors to China’s tastes and political climate.
Baidu, which has made a name for itself in the search engine, has been stepping up its self-driving game in recent years, operating ERNIE-ViLG, a 10 billion parameter model trained on a dataset of 145 million Chinese image-text pairs. How does it compare fairly to its US counterparts? The following are the results of the prompt “kids eating shumai in New York Chinatown” for Stable Diffusion, and the results of the Chinese prompt for ERNIE-ViLG (kids eating shumai in New York Chinatown).
As someone who grew up eating dim sum in China and Chinatown, I’d say it turned out to be a tie. Neither got the right siu mai, which in dim sum context are succulent, shrimp and pork dumplings wrapped in half-open yellow wrappers. While Stable Diffusion creates the vibe of a Chinatown dim sum restaurant, its siu mai is closed (but I know where the machines are going).Although ERNIE-ViLG did produce A sort of A type of siu mai, it is a variety more common in eastern China than the Cantonese version.
The quick test reflects the difficulty of capturing cultural nuance when the dataset used is inherently biased – assuming a steady spread there will be more data on the Chinese diaspora, and ERNIE-ViLG may have accepted a wider variety of images of siu mai training, these images are much rarer outside of China.
Another Chinese tool that has caused quite a stir is Tencent’s XYUME, which turns photos of people into anime characters. AI generators exhibit their own biases. It is aimed at Chinese users and has unexpectedly become popular in other anime-loving regions such as South America. But users quickly realized the platform’s inability to identify black and plus-size individuals, groups that are conspicuously absent from Japanese anime, led to objectionable AI-generated results.
Besides ERNIE-ViLG, another large 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’s largest Research branch of Microsoft Research Asia. The open-source AI model is trained on 20 million filtered Chinese image-text pairs and has 1 billion parameters.
Unlike Baidu and other profit-oriented tech companies, IDEA is one of the few institutions that have been backed by local governments in recent years to conduct research on cutting-edge technologies. That means the center may enjoy more freedom to research without the pressure to drive commercial success. Based in the tech hub of Shenzhen and backed by one of China’s wealthiest cities, it’s an up-and-coming company to watch.
artificial intelligence rules
China’s generative AI tools are characterized not just by the fact that they learn from domestic data; they are also subject to local laws. As MIT Technology Review points out, Baidu’s text-to-image model filters out politically sensitive keywords. This was to be expected, as censorship has been a common practice on the Chinese Internet.
Even more important to the future of this burgeoning field is a new set of regulations targeting what the government calls “deep synthesis technology,” which refers to “the use of deep learning, virtual reality and other synthetic algorithms to generate text, images, audio, Videos and virtual sets.” As with other types of internet services in China, from gaming to social media, users are asked to verify their names before using the generated AI applications. The fact that hints can be traced back to a person’s real identity inevitably has restrictive effects on user behavior.
But on the bright side, these rules could lead to more responsible use of generative AI, which has been misused elsewhere to produce NSFW and sexist content. For example, Chinese regulations explicitly prohibit people from generating and disseminating artificial intelligence-generated fake news. However, how this is implemented depends on the service provider.
“Interestingly, China is at the forefront of trying to regulate [generative AI] As a country,” Yoav Shoham, founder of Israel’s OpenAI competitor AI21 Labs, said in an interview. Artificial intelligence may somehow ensure that the legal system or the social system keeps up with the development of technology, especially in terms of regulating automatic power generation. “
But there is no consensus on how this rapidly changing field should be managed. “I think this is an area where we learn together,” Shoham admits. “It has to be a collaborative effort. It has to involve technologists, the public sector, social scientists, people affected by technology, and government, including the commercial and legal sector regulatory aspects, who really understand the technology and what it can and cannot do. “
Artificial Intelligence Monetization
As artists fear being replaced by powerful artificial intelligence, many in China are using machine learning algorithms to make money in a variety of ways. They’re not from the most tech-savvy crowd. Instead, they are opportunists or stay-at-home moms looking for additional sources of income. They realized that by refining the hints, they could trick the AI into making creative emojis or stunning wallpapers, which they could post on social media for ad revenue or simply charge for downloads. The really skilled ones also sell their tips to others who want to join the money-making game – and even train them for a fee.
Like the rest of the world, others in China are using AI in formal work. For example, light novel writers can cheaply produce illustrations for their works, a genre that is shorter than fiction and often features illustrations. An interesting use case that could disrupt the manufacturing sector is using AI to design prints on T-shirts, nail art, and other consumer products. By rapidly producing high-volume prototypes, manufacturers can save design costs and shorten production cycles.
It is too early to understand how the development of generative AI will differ in China and in the West. But entrepreneurs have already made decisions based on their early observations. Several founders told me that businesses and professionals are generally happy to pay for AI because they see an immediate return on investment, so startups are eager to develop industry use cases. One clever application came from Sequoia China-backed Surreal (later renamed Movio) and Hillhouse Capital-backed ZMO.ai, which found e-commerce sellers struggling to find foreign models during the pandemic as China closed its borders. solution? The two companies are working on algorithms to generate fashion models of various shapes, colors and ethnicities.
But some entrepreneurs aren’t convinced their AI SaaS will see the skyrocketing valuations and meteoric growth that their Western peers like Jasper and Stability AI are enjoying. Many Chinese startups have told me over the years that they have the same concern: Enterprise customers in China are generally less willing to pay for SaaS than those in developed economies, which is why many of them are starting to expand overseas.
The competition in China’s SaaS field is also fierce. “In the U.S., you can do very 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 It’s also possible to steal your source code overnight and hire dozens of customer support people to outdo you at little cost,” said the founder of a generative AI startup in China, requesting anonymity.
Shi Yi, founder and chief executive of sales intelligence startup FlashCloud, agrees that Chinese companies often prioritize short-term rewards over long-term innovation. “In terms of talent development, Chinese tech companies tend to focus more on applying skills and making quick money,” he said. A Shanghai-based investor, who asked not to be named, said he was “a bit disappointed that the big breakthroughs in generative AI this year all happened outside China.”
Even if Chinese tech companies want to invest in training large neural networks, they may lack the best tools. In September, the U.S. government imposed export controls on high-end artificial intelligence chips to China. While many Chinese AI start-ups are focused on applications and don’t need high-performance semiconductors that process massive amounts of data, for those doing basic research, using less powerful chips means computing time, said an enterprise software investor. A longer, more expensive job at a top venture capital firm in China, who requested anonymity. The good news, he believes, is that in the long run, such sanctions are pushing China to invest in advanced technology.
Baidu, a company that bills itself as a leader in China’s AI field, sees the impact of U.S. chip sanctions on its AI business as “limited” in the short and long term, said the company’s executive vice president and head of its AI cloud group. , God of War, on its third-quarter earnings call. That’s because “a large part” of Baidu’s AI cloud business “doesn’t rely too much on highly advanced chips.” In cases where it does need high-end chips, it “actually already has enough inventory to support our business in the near term.”
What about the future? “In the medium and long term, we actually have an AI chip developed by ourselves, so we named it Kunlun.” The executive said confidently. “By using our Kunlun chip [Inaudible] In large language models, the efficiency of text and image recognition tasks performed on our AI platform has increased by 40%, and the overall cost has been reduced by 20% to 30%. “
Time will tell whether Kunlun and other indigenous AI chips will give China an edge in the race to generate AI.