How China is Building a Generative AI Parallel World • TechCrunch

Giant tech The leap that machine learning models have shown in the past few months is getting everyone excited about the future of AI — but also TenseIt can have unsettling consequences. ChatGPT’s ability lead intelligent conversations has become a cult in many sectors, thanks to text-to-image tools from OpenAI and Stability AI.

China is a country where tech investors and entrepreneurs are searching for ways to impact the field of generative AI. Open source models are used by technology companies to offer tools to consumers and enterprises. AI-generated content benefits people. The organizers quickly responded to clarify how text, image, and video should be used. The US technology-related sanctions have raised concerns about China’s ability keep pace with the development of artificial intelligence.

Let’s take an in-depth look at how generative AI is making waves around the world as it approaches the 2022 end.

Chinese flavours

Stable Diffusion is a viral platform for creating art, and DALL-E 2 is a DALL-E 2 for creating generative AI. Chinese tech giants are half-way around the world. They have also captured the public’s attention with similar products, adapting to the country’s taste and political climate.

Baidu has been prominent in search engines in recent times Stepping up its self-driving gameWorks Ernie FelgA 10 billion parameter modelThey were trained using a data collection of 145,000,000 pairs of images in Chinese and text. How does it compare to its American counterpart? Here are the results from the “kids eat shumai in New York’s Chinatown” spot claim submitted to Stable Diffusion, against the same prompt in Chinese (纽约唐人街小孩吃烧卖) for ERNIE-ViLG.

I’d say the results were a draw as someone who grew to eat dim sum in Chinatown and Chinatown. The shumai was wrong for neither of them, which is a bit surprising considering the dim sum they were eating. Succulent dumpling, shrimp and pork in a half open yellow wrapper. Stable Diffusion captures the vibe of a Chinatown dimsum restaurant but its shumai is not. However, I see where the machine is heading. ERNIE ViLG is also creating filesAIt is a type of shumai and is more popular in eastern China than the Cantonese.

The rapid test reflects the difficulty in capturing cultural nuances when the datasets used are inherently biased—assuming that a stable spread would have more data on the Chinese diaspora and perhaps ERNIE-ViLG would be trained on a larger collection of rare shumai outside of China.

Tencent is another Chinese tool that has made a lot of noise. Dimensions are different from meThe AI generator can transform photos into anime characters. The AI ​​generator shows its own bias. It was initially designed for Chinese users. However, it has quickly gained popularity in other anime fans’ regions such as South America. But users soon realized that the platform failed to identify black and plus-sized individuals, groups notably missing in Japanese animation, resulting in offensive results from the AI ​​creation.

ERNIEViLG is another large-scale Chinese text/to-image model. TayeIDEA, a research laboratory, is the brainchild of IDEA. Led by famous computer scientist Harry ShumMicrosoft Research Asia is the largest research arm of Microsoft. It was founded by, who also co-founded it. Open source AI model has been trained using 20 million pairs of filtered Chinese text and images. It has 1 billion users. border.

IDEA, unlike Baidu and other profit-driven tech companies, is one of few local governments that has been able to support its work in cutting-edge technology. This gives the center more freedom to search for information and less pressure to achieve commercial success. They are a rising group to watch, based in Shenzhen’s tech hub.

AI rules

China’s Generative AI Tools don’t just use the local data they have learned from, but also incorporate local laws. As MIT Technology Review pointed outBaidu’s text to image model filters out politically sensitive keywords. This is not surprising, considering that censorship has been a long-standing practice on the Chinese internet.

What is more important to the future of the emerging field is the new set of regulatory measures targeting what the government calls “deep synthesis technology,” which refers to “technology that uses deep learning, virtual reality, and other synthesis algorithms to create text, images, audio, video, and virtual scenes.” As is. As is.

These rules could, however, lead to a more responsible usage of generative AI that is already being misused elsewhere. NSFW and sexist content production. For example, Chinese regulations prohibit people from spreading fake news created by AI. The providers can decide how they want to implement this.

China is, surprisingly, at the forefront in trying to regulate [generative AI] as a country,” said Yoav Shoham, co-founder of AI21 Labs, OpenAI competitor based in IsraelIn an interview. “There are many companies that put restrictions on 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 in line with the technology, especially in terms of regulating the automatic generation of content.”

However, there is still no consensus on how to manage this rapidly changing field. Shoham stated that he believes it is an area where we all can learn together. It must be a team effort. It should involve technologists who have a deep understanding of the technology and its limitations, the public sector and social scientists, and anyone else that is affected by the technology, including government officials and the legal and business side of regulation.

AI investment

Many artists in China are using machine learning algorithms to make a lot of money, as they fear being replaced by artificial intelligence. They aren’t the most tech-savvy people in the crowd. They are instead opportunists and stay-at-home moms seeking an additional source for income. They realize this by improving their claims. They can trick the AI ​​into making creative emojis or amazing wallpapersThey can also post the information on social media to increase ad revenues or charge for downloads. The truly skilled can also sell their skills to others who are interested in making money, or even train them.

Some Chinese officials use artificial intelligence in official functions like the rest of the world. For example, light fiction writers can create illustrations for their works cheaply. This genre is usually shorter than novels, and often includes illustrations. Artificial intelligence can be used to design t-shirts, shirts, and prints that are used in other consumer goods. This could be a disruptive use case for the manufacturing industries. Manufacturers can reduce design costs and shorten production cycles by quickly producing large quantities of prototypes.

It is too soon to say how different the West and China’s development of artificial intelligence will be. However, the entrepreneurs made their decisions based upon their early observations. I have heard from a few entrepreneurs that professionals and companies are happy to pay for AI as they see a direct return. Startups are keen to identify potential industry uses. One clever application came out China-backed surreal sequoia (Later It has been renamed to Movio) And the ZMO.ai powered by Hillhouse, which discovered that ecommerce sellers were having difficulty finding foreign models while China kept its borders open during the pandemic. The solution? They developed algorithms that could produce models of all shapes, colors, genders.

Some entrepreneurs doubt that AI-powered SaaS will be able to achieve the same high valuations and rapid growth as their Western counterparts. jasperThen there’s the AI stabilityShe is enjoying. Many Chinese startups have expressed the same concern over the years. They fear that corporate customers in China will not pay more for SaaS than those in advanced economies. Many of them have begun to expand abroad.

China’s SaaS industry is also fiercely contested. “In the US, you can do fairly well creating product-led software, which doesn’t rely on human services to acquire or retain users. But in China, even if you have a great product, your competitors can steal your source code.” Overnight and hiring dozens of customer support staff, who don’t cost that much, to outsmart you,” says the founder of a Chinese AI startup. He requested anonymity.

Shi Yi, founder & CEO Start FlashCloud Sales IntelligenceShe agreed with the statement that Chinese companies tend to prioritize short-term returns over long term innovation. He said that Chinese tech companies are more focused on generating quick cash and getting skilled in applications than they are on talent development. One Shanghai-based investor who chose not to be identified said that he was disappointed that all the breakthroughs in generative AI this season are happening outside China.

Barrieres ahead

Even though Chinese tech companies are keen to train large neural networks, they may not have the right tools. The US government released September’s report. Imposing restrictions on China’s export of advanced artificial intelligence chips. Many Chinese AI startups are focused on the application interface, and don’t require high-performance semiconductors to handle large amounts of data. However, basic research will require computing to take longer and cost more. A major Chinese venture capital company asked not to be named. He said that the good news was that such sanctions have pushed China to invest long-term in advanced technologies.

Baidu sees itself as a leader for AI in China and believes that US chip sanctions are not likely to have an impact on its AI business. According to the company’s executive vice-president and head of AI Cloud Group, Baidu. Du Xin, can be found on Baidu’s site Q3 earnings call. Baidu’s AI cloud business is “not very dependent on highly-advanced chips”. If you do require quality chips, they may already be available enough to support our business in short term.

What about the future? “When we look at it in the medium to long term, we actually have our own developed AI chip, which is named after KunlunThe CEO spoke confidently. Using our Kunlun Chips [Inaudible] In large language models, the performance efficiency of text and image recognition tasks on our AI platform was improved by 40% and overall cost reduced by 20% to 30%.”

It will be interesting to see if Kunlun or other original AI chips can give China an edge in generative AI.

The story has been updated so that Yoav Shaham is now a cofounder of AI21 Labs.