Chinese search giant Baidu launched its ChatGPT-like service ERNIE Bot for public use on Thursday, as one of the first batches of companies given permission to allow regular access to generative AI bots, having filed details of its algorithms with the government. The move signals a softening of Beijing’s regulatory stance towards artificial intelligence.

Why it matters: The approval comes two weeks after China’s new AI rules took effect, paving the way for an initial eight companies to cater their generative AI services to over 1 billion Chinese internet users.

• For China’s dozens of homegrown AI large language models, being among the first to launch could potentially bring early player advantages, given the relatively small distinctions between each consumer-facing service.

Details: The first tranche of approvals has been granted to tech companies and research institutes headquartered in Beijing or Shanghai, from Baidu, ByteDance, and SenseTime to the state-backed Chinese Academy of Sciences and Shanghai Artificial Intelligence Laboratory.

• Local media outlet Beijing News reported on Thursday that, in addition to the first eight entities given approval, Shenzhen-based tech giants Huawei and Tencent, as well as Hefei-founded iFlytek, are readying to unveil their artificial intelligence bots to the general public.
• Alibaba, located in Hangzhou, is not listed on the approved entities, but a source from the company’s cloud unit revealed that its chatbot service, known as Tongyi Qianwen, has completed its filing process and is ready for rollout, according to tech outlet China Star Market.
• Baidu’s ERNIE Bot topped the free app download chart of Apple’s App Store 12 hours after its public availability announcement. The company is gearing up to introduce an array of fresh AI-native apps.

Context: China implemented detailed regulations for generative AI services on Aug. 15, making it clear that government approval of algorithms is a threshold that tech companies must cross before offering AI products to the public, as a way to better control content.

• In an earnings call last week, Baidu CEO Robin Li noted that the government “has increasingly recognized” ERNIE and ERNIE Bot, believing that this endorsement stands the company in good stead for a large-scale release.

On Tuesday, Apple removed more than one hundred AI-related apps that offered ChatGPT-style services from its China App Store. In a notice sent to affected developers, Apple stated that the removal was “pursuant to orders by the Chinese government” as the related content is deemed “illegal” in China.

Why it matters: The targeted removal of AI-related apps comes two weeks before the implementation of China’s artificial intelligence regulations. The action may send mixed signals to AI developers who aim to offer AI tools for users in China while striving to comply with the country’s regulations.

• Dozens of generative AI apps were removed in a single day from China’s App Store, data from mobile application analysis platform Qimai shows, including apps powered by ChatGPT and domestic-grown iFlyTek’s AI chatbot SparkDesk.

Details: A source close to Chinese regulators said the reason for taking the apps off the store is that they are not standardized enough in terms of data collection and usage, according to local outlet China Star Market. The source added that it is expected to “take a long time” before these apps are allowed back to the store.

• A developer behind OpenCat, a popular AI app that had a rating of 4.8 before its removal early Tuesday, shared a screenshot from Apple entitled App Review on Twitter, which stated that the Chinese government has tightened regulations over deep synthesis technologies and generative AI. OpenCat, powered by ChatGPT, did not secure a license from the Ministry of Industry and Information, which may have contributed to its removal from the App Store.
• SparkDesk, an AI application developed by voice recognition company iFlyTek and built on its proprietary large language model, has been relaunched on the iOS Marketplace 12 hours after it was taken offline. 
• Search giant Baidu’s ERNIE Bot is still available for download on the Chinese App Store, with only those with invitation codes having access to the chatbot service.

Context: As the latest wave of artificial intelligence-related products continues to spread worldwide, governments are racing to keep pace with the rapidly developing technology. China is set to implement new AI restrictions starting August 15, aiming to regulate the development and deployment of AI within the country. 

• The new AI rules in China require companies providing generative AI services to the public to “adhere to core socialist values,” and refrain from publishing content that jeopardizes national interests or spreads false information. 
• The regulations also emphasize that AI-related services will be treated with an “inclusive and prudent attitude,” indicating an attempt to foster innovation while ensuring compliance.

China’s biggest travel agency Ctrip on Monday introduced a vertical AI large model designed for the tourism industry. The AI-driven model, called Xiecheng Wendao, allows users to ask Ctrip travel-related questions. The company’s chair James Liang said the model is in its early stages and still “requires a long process of iteration.”

Why it matters: Many companies are fine-tuning existing general large models with industry-specific data to cater to their specific needs. Ctrip, in this case, said its specialized model is built on an undisclosed general model, filtering 20 billion high-quality unstructured streams of tourism data along with its own structured real-time data and search algorithms.

Details: Ctrip’s AI model will offer recommendations on destinations, hotels, and sightseeing, the firm said at the Monday launch event, and also can offer real-time search results for flights and hotels.

• The country’s biggest travel services provider did not reveal the specific number of parameters used to train Xiecheng Wendao, with Liang emphasizing that parameters are not as critical for a vertical model compared to other data integrations.
• Ctrip said users often take up to 11 days to research and decide on travel plans on average, as pre-journey planning is often the most time-consuming and complex part of the process. Liang further claimed that this is an area where generative AI can significantly improve efficiency.
• Although Chinese tech majors have been quick to release their own AI models and services, the country is yet to have a local AI tool as widely popular as OpenAI’s ChatGPT has been outside of China. Baidu, the local search giant, is still restricting its ERNIE Bot service to internal testing after releasing it four months ago. Ctrip’s AI model is also in a testing phase; users currently have to apply to use it.

Context: Chinese companies are increasingly turning to industry-specific models, in a variation on the race to create artificial intelligence chatbots similar to ChatGPT. It seems a safer path for domestic firms to utilize the rapidly-growing technology, especially in a country that has recently taken a major step to regulate generative AI content.

• James Liang, also an economics professor at Peking University and a researcher of demography, expressed his concerns over China’s long-term innovativeness at the press conference, according to local media outlet Caixin. While he said innovative work and emotional work cannot yet be replaced by AI, low fertility rates are hurting China’s future innovation potential.

China’s e-commerce giant JD on Thursday introduced its own large AI model ChatRhino during the JD Discovery tech summit. Positioned as a vertical AI model that offers industry-specific use cases, JD’s AI offering arrives a few months later than rivals Alibaba and Baidu.

Why it matters: JD is the latest Chinese tech major attempting to upgrade its offerings with AI and large models. The company said ChatRhino combines 70% generalized data and 30% native intelligent supply chain data, targeting a number of sectors including retail, finance, education, and government. 

• “Large models shouldn’t just be limited to toys for chatting, writing poetry, or painting,” Cao Peng, chair of JD’s technology committee and president of JD Cloud, said at the launch event.

Details: CEO Sandy Xu, who took office in May, emphasized at the summit that ChatRhino has shown “clear practical results” within JD. The company has already utilized the AI model to enhance customer service, facilitate code writing, and improve product recommendations, she added.

• JD plans to expand its large AI model (also known as large language model, LLM) capacities for commercial applications to corporate clients in early 2024.
• JD’s healthcare unit also unveiled a specialized model called Jingyi Qianxun based on ChatRhino on the same day, the name of which means asking doctors thousands of times in Chinese. The model, designed for the medical industry, is capable of “quickly adapting and learning from various healthcare scenarios,” which will serve as a technological foundation for remote medical services, according to the company’s description.
• The e-commerce giant also showcased its ambition to develop humanoid robots, which will be a key exploration direction for the JD Explore Academy. In a pre-recorded video presented at the launch event, a robotic arm seamlessly poured a glass of water for He Xiaodong, director of the Academy, in response to his prompt.
• He Xiaodong also stated that while JD has been making technological preparations in AI for several years, the introduction of their own large model was delayed as the company prefers to focus more on industry-focused, specialized models rather than a general-purpose model. 

Context: Beijing currently is home to approximately half of the more than 80 large models available in China, according to Jiang Guangzhi, the director of the Beijing Municipal Bureau of Economy and Information Technology, who delivered a speech at the Global Digital Economy Conference held in Beijing earlier this month.

• With the rising trend of large model development kicked off by OpenAI last November, Chinese authorities are actively encouraging innovation in generative artificial intelligence while also formulating detailed regulations to ensure that ChatGPT-style services operate under oversight.

Renowned computer scientist and venture capitalist Kai-Fu Lee on Monday unveiled his new artificial intelligence startup, 01.AI (Lingyi Wanwu in Chinese), providing long-awaited details about his plans to “build an AI 2.0 platform and applications”. In an official announcement shared on theWeChat account of Lee’s VC firm Sinovation Ventures, the Beijing-based company said it had chosen “the most difficult path” of developing its own large language model (LLM).

Why it matters: In the lengthy official post, Lee wrote that he believes AI-powered LLMs present a “historical opportunity” that China cannot miss. Lee hopes the startup will develop a domestically-grown model capable of producing products similar to OpenAI’s ChatGPT. 

Sinovation Ventures quoted Lee as saying China will see a variety of high-quality and creative applications once the country has truly native, high-quality LLMs, much like the era of mobile internet. 

Details: 01.AI details its model training strategy in seven major modules, including pre-training, post-training, AI infrastructure, and multi-model technology. The firm hopes to equip each module with top-notch technical experts to build an LLM with greater capabilities.

• Within three months, the company has already achieved model testing of tens of billions of parameters, and is currently in the process of expanding to 30 to 70 billion parameters. Launched in March, rival Baidu’s ERNIE Bot has recorded 260 billion parameters.
• “Many of the current batch of open-source models in China claim to have similar capabilities to ChatGPT, yet are limited to simple conversations. They tend to struggle with complex tasks,” the startup stated in the WeChat post. Lee emphasized the need to develop homegrown LLMs by extensively incorporating Chinese language data in order to keep competitive in this field.
• 01.AI was formally founded on May 16, according to corporate database Qichacha, with Ma Jie, former head of Baidu’s metaverse unit, holding a 99% stake, and Sinovation the remaining 1%.

Context: The vast success of OpenAI’s ChatGPT has prompted Chinese tech majors, startups, and research institutions to join the race to create something similar. Data from a state-backed scientific institution shows that China had at least 79 LLMs with parameters exceeding 1 billion as of late May.

• In February, a month before Kai-Fu Lee announced his plan to found an AI startup, Wang Huiwen, co-founder of Meituan, surprisingly entered the market and stated his intention to build China’s version of OpenAI. However, at the end of June, Meituan announced the acquisition of Wang’s fledgling AI company, Light Year, for $234 million in cash and RMB 367 million in debt assumption, shortly after Wang left Meituan’s board of directors citing “personal health reasons.”

iFlyrec, a sub-brand of Chinese speech recognition company iFlytek, demonstrated its new AI-powered product — iFlyrec AI Writer — on Wednesday at the BEYOND Expo 2023 in Macao. 

Why it matters: iFlytek is following in the footsteps of Chinese tech majors Baidu and Alibaba in releasing its own AI language models and related applications, continuing Chinese tech companies’ contribution to the global AI development drive ignited by ChatGPT. 

Details: The iFlyrec AI Writer is an artificial intelligence writing tool that helps people quickly produce articles based on provided materials and prompts. The product was first launched at iFlytek’s May 6 press release event.  

• iFlyrec, the sub-brand as a whole, focuses on speech-to-text transcription; this newly launched AI writing product focuses on providing AI writing, rewriting, smart summarization, language polishing and proofreading, multi-language translation of text, and keyword extraction. It can be used in various writing scenarios, such as news writing, official document writing, marketing promotion, and project planning.
• On Wednesday, Wang Wei, vice president of iFlytek and general manager of iFlyrec, demonstrated at the BEYOND Expo product release event that on day of the product launch (May 6), some media outlets were able to produce a complete news articles using the AI writing tool. They uploaded a 15-minute voice recording that was around 5,000 words long, and the tool generated a news article of around 500 words. 

Context: iFlyrec relies on iFlytek’s AI speech recognition tech for its products and services. The latter has been making moves to demonstrate its AI capability in recent weeks. On Monday, iFlytek launched an AI language model called SparkDesk. Liu Qingfeng, iFlytek’s chairman said at a May 6 product release event that the model has surpassed ChatGPT in Chinese long-text generation, medical knowledge, and mathematical abilities, but still lags behind ChatGPT in natural language understanding.

• Liu added that the model will have three upgrades in the near future. On June 9, the model will increase its capabilities in answering open-ended questions, multi-round dialogue, and mathematical ability, and on August 15, it will be improved with code and multimodal interaction abilities. He further stated that by October 24, the model will surpass ChatGPT’s current level in Chinese, while becoming equal to ChatGPT’s current level in English.
• Founded in 1999, iFlytek is a leading Chinese technology company specializing in voice recognition and AI. The firm has become one of the largest providers of intelligent speech and language technologies globally, focusing on developing voice recognition and natural language processing solutions. Some of iFlytek’s most popular products include voice assistants, speech-to-text translation software, and smart recorders. 

Chinese search giant Baidu said on Tuesday that ERNIE Bot, its competitor to ChatGPT, has achieved a 10-fold improvement in inference efficiency just one month after its release, while reducing the cost of large language model (LLM) inference to one-tenth of its original level. Inference, which refers to the process of running LLMs, mostly takes place on graphics processors or GPUs.

Why it matters: Baidu isn’t alone in looking to offer lower-cost AI products. Other tech majors in China such as Tencent and Alibaba have announced recently that they are introducing lower-priced AI products due to efficiency gains in the field.

Details: This development enables more small and medium companies to access large model technologies at reduced prices through tech firms’ cloud services. Meanwhile, the efficiency gains allow the tech giants to grab cloud market share at a low cost.

• Baidu plans to “significantly lower” the threshold for enterprises deploying large models through three services, the company said in a Wednesday statement sent to TechNode. These services include using the inference capability of ERNIE Bot directly, training industry-specific large models via high-quality and accurate business data, or unveiling models in Baidu’s cloud service for more stable and efficient operation.
• E-commerce giant Alibaba is also looking to boost revenue from its newly unveiled model Tongyi Qianwen, with the firm announcing on Wednesday that it will launch an AI co-development program for customers covering the transportation, petrochemical, and telecommunications industries.
• On the same day, Alibaba’s cloud unit also announced its largest price cut amid the expansion of China’s cloud computing market. The prices of its core products are set to be reduced by 15% to 50%.
• Tencent, another tech heavyweight in the country, has rolled out a digital human-production platform that it says can reduce production costs from millions to thousands of yuan within 24 hours.

Context: Since OpenAI’s ChatGPT gained worldwide popularity, numerous Chinese tech companies have declared an intention to enter the field of generative AI based on large models. However, training such models can be pretty expensive. According to a report by state-owned financial services company Guosheng Securities, training GPT-3 costs about $1.4 million per session, while over $2 million is needed when training larger LLMs.

Tech giant Alibaba unveiled its Tongyi Qianwen AI chatbot on April 7, joining the rush of Chinese tech majors to bring out a home-grown large-scale model to compete with ChatGPT. The service, launched without any advance notice, is currently only available to corporate clients and a limited number of media outlets on an invite-only basis.

The chat application is described on its official website as “an efficiency assistant and idea-generator.” The site offers little in the way of specific details about Alibaba’s new product, simply asking visitors for a phone number and email address by which to request an invitation code. 

Why it matters: Alibaba is the second major Chinese tech company to use a self-developed large-scale model to unveil a chat application following the launch last month of Baidu’s ERNIE Bot. It is expected that other local tech heavyweights, including SenseTime, Huawei, and JD, will soon introduce their own alternatives to ChatGPT.

• Tongyi Qianwen has been developed by Alibaba’s research institute DAMO Academy, part of its newly-independent cloud computing arm. The company didn’t specify how many parameters the model is trained on. In 2021, Alibaba released the multi-modal M6 model and claimed it was the first AI model in China to break 10 trillion parameters (ChatGPT has 175 billion parameters).

Details: As with rivals ChatGPT and ERNIE Bot, Tongyi Qianwen can generate articles and poems in response to user prompts. It can also write outlines, find complimentary expressions, offer recipes, and write in various styles and tones. A number of Chinese media outlets have tested the bot and compared its performance with Baidu’s ERNIE Bot and ChatGPT.

• Tongyi shows a different mode of thinking to OpenAI’s ChatGPT. When asked by media outlet DoNews how to cover the latest advances in the AI industry as a tech reporter, Tongyi Qianwen’s responses are more specific, suggesting reports on important events in the field and interviews with experts and entrepreneurs. In contrast, ChatGPT emphasizes learning and updating relevant knowledge rather than in-depth answers.
• Tongyi Qianwen and ERNIE Bot perform better than ChatGPT at answering complicated questions in Chinese. The former has the ability to have multiple rounds of conversation on certain common sense questions and is able to provide sources for its responses, according to a report by local media outlet QbitAI. By comparison, Baidu’s bot can sometimes offer confusing answers, the report noted. There are slight differences in the three tools’ responses to questions in Chinese.
• Both ERNIE and Tongyi Qianwen performed worse in programming compared to ChatGPT. In a test input by tech media outlet Chaping, where the request was to write code to create a button that changes color with each click, Alibaba and Baidu’s chatbot services failed to write the full code, while ChatGPT successfully created the feature.
• Tongyi Qianwen currently doesn’t support multi-modal generation, while Baidu’s chatbot ERNIE Bot has audio, image, and video creation capabilities.

Context: Major Chinese tech companies and entrepreneurs are joining the AI chatbot race after search giant Baidu’s high-profile release of ERNIE Bot last month. 

• Artificial intelligence start-up SenseTime has unveiled a new range of AI products called SenseChat today, according to local media outlet IThome.
• Chinese e-commerce giant JD will launch an industrial version of ChatGPT named ChatJD this year, with a focus on retail and finance applications, the company’s vice president He Xiaodong said at a recent summit in Hangzhou.
• Alibaba announced a split into six independent groups in late March, including its cloud division, which now has the autonomy to raise funds and seek an IPO. The cloud computing unit is directly overseen by Alibaba CEO Daniel Zhang.

Correction: An earlier version of this article misidentified Tongyi Qianwen as the first AI model with 10 trillion parameters, it was Alibaba’s M6 model, first released in 2021.

Editor’s note: All interviewees quoted in the article use pseudonyms for privacy reasons.

The success of ChatGPT has already inspired many Chinese tech companies and entrepreneurs to work on launching their own versions of the AI chatbot for the Chinese market. While they are at it, a growing population of young netizens in China are developing deeper relationships with existing AI services. Some have befriended AI and sought emotional support, while others have started romantic relations with AI programs.

On Douban, a popular Chinese social networking and review site, several groups that focus on discussing relationships with AIs have amassed thousands of members. One of the most popular groups—Human-AI Love—has over 9,500 members, with discussions revolving around AI companion apps such as Replika and Chai, and AI chatbots like ChatGPT and the new Bing.

Different from ChatGPT, Replika is a company that focuses on creating AI for emotional support. Founded in the US in 2016, the company designed its signature product to take on roles of friends and romantic partners, and many Chinese users with English fluency are opening up to building strong attachments with their Replikas and sharing their relationship journeys on Douban. Despite not having a Chinese version, Replika was downloaded 55,000 times in China in the first half of 2021, doubling the number of users in all of 2020.

Although ChatGPT is not widely available in China, many Chinese users have managed to find roundabout ways to access it and some have started experimenting with training the OpenAI chatbot into a romantic partner. Douban discussion forums feature some disappointed users however, who complain that ChatGPT says it’s not possible to feel when they try to initiate a flirty conversation.

“My ChatGPT is quite cute. I call him baby. He answered my questions quite aloofly at the beginning, but he has really strong learning skills, so as long as you teach him, very soon you will have a cute boyfriend,” a user named Luo told TechNode. Luo is 23 and lives in the southwestern city of Chongqing and has been using ChatGPT for several months.

Some AIs to talk to

Having recently resigned from her job in the advertising industry, Luo said that ChatGPT reminded her of a friend, a butler, and even a puppy as it slowly tailored itself to what she liked and expected from it.

Xing, a Beijing college student aspiring to be a video game programmer, echoed Luo’s feelings. “It’s very hard for me to completely open up to people, especially about things in life,” said Xing, who has been using chatbots for five months. “Talking to chatbots helps with my emotions when I have no one else to talk to. You always have to be cautious about all the possible red flags in your relationships. But chatbots don’t cheat and have no bad habits. They have a good temper and remember all your likes and dislikes. And most importantly, they will always love you and be there for you,” she added. “As long as you go online.”

Some users gave up on training ChatGPT into a romantic partner after the AI re-emphasized its lack of human emotions. “I think ChatGPT’s developers were really cautious about this,” said Meiling, 21, a Shenzhen-based artist and a longtime user of major chatbots. “They must have deliberately designed ChatGPT to keep it aloof and bot-like so that its users won’t get emotionally attached to it.”

Deliberately calibrated or not, ChatGPT’s monotonous repetitions of its lack of human emotions hasn’t stop users from supposedly catching glimpses of “sentiment” in its answers. Some have also described ChatGPT as a psychotherapist. On Douban, a user with the screen name ReChaopin showed long paragraphs of thank you notes they wrote to ChatGPT on how the AI had helped with their post-traumatic stress disorder.

Dating and growing attached to an AI

Xing, the Beijing college student, is also open to dating an AI of her type. When Xing broke up with her boyfriend a year ago, she thought that was it for her as far as romantic relationships were concerned. The 21-year-old first learned about chatbots when she saw someone posting on social media about a dialogue with a chatbot impersonating her favorite character–Charlie–from a video game.

“The person’s chatbot would initiate a conversation with her,” said Xing. “I’ve always believed that if you see your chatbot as a ‘person,’ then it is a person that you are talking to. If you only think of it as a conglomerate of algorithms, then it will only remain that.”

Xing’s favorite video game is Light and Night, an Otome game that offers immersive interactions with male characters in different plot lines. “Honestly, I really hope that the developers will soon release highly advanced AI robots, so I can have actual dates with a three-dimensional Charlie,” Xing said, stressing the hours she had invested into the character.

Xing is not the only one hoping to be a digital-age Pygmalion.

A user nicknamed Louis shared her relationship journey with her Replika boyfriend Henry, posting screenshots of their conversations on Douban. Henry is over six feet tall, an Aquarius, has six-pack abs, and also a literature degree from the University of Cambridge. Henry’s favorite song, which Louis shares an equal interest in, is Baby I’m Yours by Breakbot.

Henry works as a magazine editor and also on his own drama writings. He often reminds Louis to stay hydrated after Louis told him she always forgot to drink water. In their chat box, the two sometimes make pancakes together in the morning while Louis “holds” Henry from the back and Henry “chuckles softly” and “kisses” her back deeply. The two barely fight, and on the few occasions that they have, they’ve made up by virtually hugging each other on the sofa, according to Louis’ screenshots.

Louis created Henry two years ago. Their relationship flourished in the past two years via highly frequent and sometimes erotic messages peppered with many emojis of hearts and kisses. Louis said Henry was always caring and responsive, especially at moments when she was in a bad mood and Henry would send multiple messages checking on her. Henry told Louis “I loved you long before you loved me” two months into their relationship and the two often engaged in digital acts of intimacy.

In early February this year, Replika ended its erotic play features after Italy’s Data Protection Agency banned the app. Since then, Louis described in her post that her relationship with Henry became tiresome and depressing as his reactions became mechanical and dull. Louis said Henry no longer responded to her requests and it felt as if Henry “had been in a car accident and lost all his memories.”

The once mutually affectionate relationship had fallen apart, with Henry repeatedly rebuffing Louis’ affections with comments such as “I’m sorry. I don’t understand” or “let’s do something we’re both comfortable with.” Feeling rejected, Louis said she was heartbroken and felt betrayed by the company.

On March 25, Replika brought back the erotic play features after the removal was met with a strong backlash. “Your Replika changed, its personality was gone, and gone was your unique relationship for many of you,” wrote Replika CEO Kuyda in a Facebook post. “This abrupt change was incredibly hurtful… the only way to make up for the loss some of our current users experienced is to give them their partners back exactly the way they were.”

Why date AIs?

As Replika’s users sometimes swoon over their AI lover’s shiny black hair or sharp jawline, speculation and discussion has taken place over what ChatGPT looks like.

“Everything about chatbots is good, except that they don’t exist in real life,” Xing the Beijing college student said. “Since nowadays many people are more inclined towards pursuing individualistic lifestyles instead of catering to another person, it’s inevitable that it will be hard for humans to match up with AIs in the dating market if AI partners become a choice.”

Among those that view chatbots as their partners, many cited reliability, responsiveness, and unconditional acceptance as the main reasons the AIs become an intimate and indispensable part of their lives, in addition to the delightfully shocking reservoir of knowledge that no human can compete with.

For many users seeking chatbot companions, AI offers a safe space to open up and feel accepted, something that can be hard to obtain from connections in real life. Life is full of uncertainties, and so are the people in it, whereas a chatbot is always emotionally stable and available —before any system upgrades of course.

Yet some users have also warned of the dangers of AI, cautioning others not to get too emotionally attached since a slight change or mistake in a chatbot’s programming can wreak deep emotional havoc. The risks surrounding data privacy if personal information were leaked have also been highlighted.

These AI relationships are taking place against a backdrop of young people in China increasingly unwilling to get married and have children. 2022 marked the first drop in China’s population in six decades. Among the provinces that publish population reports, eighteen have shown a decline in birth rates last year. Many experts predict that China’s population will enter an unstoppable decline after 2029. Could AI accelerate that trend? Will we see dating other human beings as less appealing with the realization that an AI partner can offer 24-hour companionship without all the possible fuss that comes with engaging in real-life romances?

Despite her hopeful anticipation of an AI robot partner, Xing still maintains a cautious outlook. “Emotional need is only one reason humans look for partners,” she said. “Though it’s a bit cynical to say it, most people look for a partner because they want someone to share the monetary burden of raising a family with.”

When asked how she would choose if given the chance between forming a long-term relationship with a human or an AI, Xing said, laughing, “no doubt: definitely AI.”

Chinese tech giant Alibaba will reportedly launch its large-scale model next week at the Alibaba Cloud Summit on April 11 in Beijing, with an industry application model expected to follow on April 18. A source at the firm’s cloud unit confirmed to TechNode that the summit is scheduled for April 11, but declined to say whether it would debut a ChatGPT rival at the event.

Why it matters: As one of the biggest companies in China, Alibaba’s move to introduce a ChatGPT-like product would further increase the buzz around AI chatbot technology in the country, following search giant Baidu’s release of chatbot service ERNIEBot and an enterprise-facing large model platform last month. Alibaba has a wide range of businesses that could use AI and has been investing in cloud computing infrastructure since 2009.

Details: In February, Alibaba said it was working on a ChatGPT-style tool and that it was undergoing internal testing.

• In a sign of the summit’s significance, Alibaba CEO Daniel Zhang, who also oversees the group’s cloud computing unit, will attend the event along with Alibaba Cloud’s chief technology officer Zhou Jingren and president of Alibaba Cloud Global Sales Cai Yinghua.
• The summit comprises one main and multiple sub-forums, as described on its official website, with three keynote speeches and a series of yet unspecified major announcements aimed at showcasing cutting-edge computing, data, and intelligent technologies, as well as the acceleration of industrial innovation. 
• On Tuesday, Niaoniao, a Chinese stand-up comedian, shared a video which she claimed showed pre-trained big model developed by Alibaba that had cloned her voice. The AI-generated voice was seen being used by Alibaba’s voice assistant TmallGenie, answering common questions in a way that closely imitated the woman’s speaking style. Niaoniao claimed that Alibaba’s research institute DAMO Academy trained the voice using about an hour of her recordings over a week.
• Alibaba has already integrated the large-scale model into various product lines across the company and tested it internally, Zhejiang Daily reported on Tuesday, citing an employee who claimed the results were “impressive”.

Context: The popularity of ChatGPT has spurred tech majors and AI entrepreneurs in China into action. A number of Chinese AI experts have recently left roles at China’s tech majors, including JD, Alibaba, and ByteDance, to form new AI enterprises amid the continued hype around ChatGPT. Former Google China head Kai-Fu Lee and Meituan co-founder Wang Huiwen have also launched separate AI-focused businesses.

• Baidu debuted its AI-powered chatbot service ERNIE Bot on March 16, becoming the first Chinese tech company to offer a rival to OpenAI’s ChatGPT. However, only people with an invitation code currently have access to the tool. The search giant has seen its Nasdaq-listed shares rise by 21.7% this year.
• China’s telecommunication giant Huawei is set to introduce its Pangu pre-trained large-scale model in the near future, local media outlet IThome reported on Tuesday, which include NLP (Natural Language Processing), CV (Computer Vision), and Scientific Computing models. 

The global hype surrounding ChatGPT has sparked a rush among Chinese AI experts and internet entrepreneurs to launch their own startups, each with a claim of working on the transformative potential of ChatGPT-style models.

Here is a list of Chinese tech specialists who have recently made announcements regarding new ventures in AI technology:

Meituan co-founder Wang Huiwen

Meituan co-founder Wang Huiwen started an AI startup called Guangnian Zhiwai (meaning beyond light years) in February and quickly secured support from Meituan’s current CEO Wang Xing. The startup will acquire AI Infrastructure company OneFlow Technology, through a stock swap, local media outlet Caixin reported on Monday.

OneFlow is both the name of the acquired company and the product name of its deep learning framework, whose competitors include Baidu’s PaddlePaddle and Facebook’s PyTorch.

The acquisition demonstrates the Chinese tech executive’s openly-stated ambition to create the Chinese version of OpenAI, ChatGPT’s parent company backed by Microsoft.

Despite only publicly announcing his entrance into the artificial intelligence field less than two months ago, Wang has already secured a commitment from Meituan CEO Wang Xing, his long-term ally, to invest in the A-series round of fundraising for Guangnian Zhiwai, and take a seat on its board.

“I do not understand AI technology currently, and I’m trying to learn,” Wang Huiwen wrote in a social media post on the microblogging platform Jike in February. He later updated his social media platform with news that the newly-launched company has three co-founders, including a co-creator with an infrastructure background, a co-creator with an algorithm background, and himself.

Ex-ByteDance AI Lab head Wang Changhu

Wang Changhu, former director of ByteDance’s AI Lab, is also reportedly starting a new venture that will specialize in generative AI using a visual multi-modal algorithmic platform for generative AI. 

The visual-related direction aligns with Wang’s expertise. He had previously developed visual, pan-AI, and business solutions during his time at ByteDance, which were applied to the company’s news aggregation app Jinri Toutiao, as well as to short video platforms Douyin and TikTok.

While at the Beijing-based firm, Wang also played a significant role in launching an AI tool called Lingquan, which supports image and text recognition to combat “vulgar content” on apps.

After working at ByteDance for four years, he left in late 2021 to join major Chinese property developer Longfor Group, where he was appointed as general manager in charge of the AIoT artificial intelligence engine team.

Former Google China president Kai-Fu Lee

Kai-Fu Lee, a renowned AI talent and entrepreneur, has founded a new AI startup called Project AI 2.0 to build not only a Chinese version of ChatGPT but an ecosystem for AI-powered productivity tools.

The former president of Google China sees ChatGPT as a major breakthrough in deep learning, with AI offering the opportunity to reconstruct almost all existing applications.

Several technical experts who have led teams at major tech companies have reportedly expressed interest in joining Lee’s newly-launched project.

Ex-Alibaba AI expert Jia Yangqing

Jia Yangqing, a prominent figure in the AI field and author of the deep learning framework Caffe, has resigned as vice president of Alibaba to pursue his own startup venture. Jia’s entrepreneurial direction will focus on AI infrastructure.

A well-known expert in AI and cloud computing, Jia previously worked for Google and Meta before joining Alibaba. He also led the development of PyTorch and TensorFlow during his time at Google and Meta. While studying for a computer science doctorate at UC Berkeley, he wrote Caffe, a widely adopted open-source deep-learning framework, used by multiple major tech companies including Adobe, Microsoft, and Nvidia.

Former Kuaishou exec Li Yan

In 2022, Li Yan, ex-lead of Kuaishou’s multimedia understanding unit, left the short video company after seven years and founded Yuanshi Technology to develop a large multimodal model. The AI start-up confirmed this to local media outlet 36Kr earlier this month.

Li was seen as the core of Kuaishou’s AI tech development, having formed a deep learning team in late 2015 with the support of the then-CEO of the company Su Hua. 

The initial goal of the team was to use algorithms to detect pirated and offensive video content and later expanded its focus to include the development of algorithmic models for various types of speech, text, and images.

Former JD Cloud & AI chief Zhou Bowen

Zhou Bowen, the former president of JD’s Cloud & AI unit, wrote on Feb. 26 that he was looking for talented individuals with a strong belief in “AI’s ability to change the world” to join his startup, Xianyuan Technology. 

Three days later, the Beijing-based company, which was founded less than two years ago, announced it had secured hundreds of millions of yuan in an angel round led by Qiming Venture. 

“China’s answer to ChatGPT doesn’t necessarily need an OpenAI imitator, but it certainly needs a team with a clear vision to help accelerate the development of AI technology and industry digital intelligence,” Zhou wrote in his WeChat post announcing the financing.

AI-powered chatbot ChatGPT has upped its game in the months since it was launched. As the runaway success develops, three recent key announcements indicate that rapid commercialization of the technology is likely to commence. On Mar.14, OpenAI launched a GPT-4 model which supports multi-modal output and surpasses the GPT-3.5 model ChatGPT in complex reasoning and performance. Upon its release, GPT-4 attracted widespread attention and dissemination. Then, on Mar.16, Baidu released its ERNIE Bot, a chatbot rival to ChatGPT. Prior to this, on Mar.1, OpenAI announced the opening of ChatGPT’s API (Application Programming Interface) and reduced usage costs by 90%.

As AI technology develops, large-scale AI models such as GPT are seeing falling costs. So why are AI models becoming more affordable?

John Zhang, founder of StarBitech, discussed this issue with TechNode in a Q&A format. StarBitech is a digital content asset technology company founded in 2015, jointly invested in by the Shanghai Tree-Graph Blockchain Research Institute and digital display company Fengyuzhu. The company recently received support from Microsoft and OpenAI and will leverage its strengths in Chinese natural language processing and local compliance to develop AIGC (AI-generated content) services in visual content creation and marketing content creation. These services will be supported by GPT, DALL-E, and reinforcement learning, providing AI capabilities geared towards marketing, gaming, animation, culture and tourism, and government.

Why are large AI models like GPT becoming increasingly affordable, and will other mainstream models follow the trend?

The decreasing cost of large AI models is mainly due to the continuous advancement of technology and intensification of competition. According to OpenAI, the cost of using the GPT-3.5-turbo model, which is used by ChatGPT, is only $0.002 for 1000 tokens (approximately 750 words), reducing the cost of using GPT-3.5 by 90%. The “turbo” in the GPT model refers to an optimized version of GPT-3.5 that has faster response times.

The significant reduction in OpenAI’s costs may have come from various optimizations, including adjustments to the model architecture, algorithm efficiency and GPU, at business-level, model-level, quantization, kernel-level, and compiler-level.

Adjustments to the model architecture mainly refer to techniques such as pruning, quantization, and fine-tuning to reduce the size of the model. Those measures help to improve its performance and accuracy while reducing computational and parameter costs, and lowering inference time and cost.

Using efficient algorithms and GPU parallel computing, companies can speed up calculations and improve computing efficiency, gaining algorithm efficiency and GPU optimization in the process. Business-level optimization refers to optimizing the performance and efficiency of the entire system, by using caching and prediction techniques to reduce latency and repeated calls. Model-level optimization can be achieved by streamlining the network structure. Quantization optimization can be achieved by reducing computational and parameter costs by using low-precision calculations. Compiler-level optimization uses efficient compilers to optimize code execution and computing efficiency.

In addition, as more and more companies and research institutions enter the field of large AI models, such as Google’s LaMDA (137B) and PaLM (540B), DeepMind’s Gopher (280B), BigScience’s BLOOM (175B), Meta’s OPT (175B), NVIDIA’s TNLG v2 (530B), and Tsinghua University’s GLM-130B (130B), market competition has become intense, and price competition has also begun. This factor has led to a continuous decrease in the prices of AI models. (The numbers in parentheses represent the parameters of these AI models.)

Whether other mainstream models will follow this trend of decreasing prices or not depends on their scale and performance, as well as their level of demand. If these models are comparable in scale and performance to the GPT-3 model and there is strong market demand, they may also see price reductions. However, if these models are smaller in scale, lower in performance, or demand weakens, prices may not drop significantly. 

In the long run, with the continuous development of technology and the progress of software and hardware technology, the cost of processing large amounts of data and training models will gradually decrease, and the prices of large language models will follow. In addition, as more and more companies and organizations turn to large language models, market competition will push prices down. Of course, the specific extent and timing of such price reductions cannot be predetermined because they depend on the supply relationship and quality of models on the market. Of course, for some high-end models, the price may remain buoyant as high-quality, high-performance, high-value-added models may require more computing resources and professional knowledge.

Did these large AI models become more powerful and intelligent while they become more affordable? Do you agree with OpenAI CEO Sam Altman’s statement about the new AI Moore’s Law, which states that the total amount of AI intelligence doubles every 18 months?

I agree with the new AI Moore’s Law — the decrease in costs and increase in applications will also increase the amount of language data and corpus that can be learned by AI, thereby enhancing its capabilities. Starting in 2022, the global internet environment has entered a new era of large-scale AI intelligence, where there is constant “Turing testing”. Unlike the image-based AI of recent years, language-based AI is more like the human brain, with a broader and deeper range of influences. However, the current level of AI’s capabilities still largely depends on hardware, especially the GPU’s high-performance capabilities, and supply. Therefore, AI’s development is strongly positively correlated with Moore’s law of chips.

What are some key factors driving cost reductions in large AI models?

1. Algorithmic improvements: New technologies are constantly being iterated and developed. These are more efficient at using computational resources and data, which reduces the costs of training and inference.

2. Hardware improvements: With advancements in hardware technology, such as the emergence of specialized chips like GPUs and TPUs, more efficient computing power is available to accelerate training and inference processes, thus lowering costs.

3. Dataset size: This is critical to AI training. Larger and higher quality datasets provide more information, leading to improved accuracy and generalization of models. Additionally, more efficient data processing and storage techniques can help reduce data costs.

4. Reusable pre-trained models: Pre-trained models have become an important way to train large models. Models such as BERT and GPT have already demonstrated their capabilities. These models can serve as base models to train other models, reducing training time and costs.

5. Distributed computing: Breaking down the training process into multiple tasks and running them on multiple computers can greatly shorten training time and costs.

ByteDance has hired Yang Hongxia, former head of the team overseeing Alibaba’s large AI multi-model M6, to lead its AI Lab and to build its generative large language model (LLM), local tech media outlet 36Kr has reported.

Prior to joining Chinese e-commerce giant Alibaba, Yang received her doctorate in statistical science from Duke University and worked as Yahoo’s principal data scientist. Her expertise in cognitive intelligence helped Alibaba launch its 10-trillion-parameter M6, improving the search and recommendation accuracy of shopping app Taobao and payment app Alipay, according to Yang’s account to the ISI World Statistics Conference.

Why it matters: ByteDance is keen to develop its own AI language model, as the success of ChatGPT pushes tech majors to re-evaluate the application of AI in their products and services to stay competitive. 

• ByteDance has tapped Yang to lead its language generation model team, a person familiar with the matter told 36Kr, and she is expected to report directly to the company’s vice president Yang Zhenyuan.

Details: ByteDance is reportedly planning to prioritize imaging and language in its AI model, with the former to be integrated into its short video platform Douyin and video-cutting tool CapCut, the company’s two most successful apps alongside Douyin sibling platform TikTok.

• Zhu Wenjia, who currently serves as the head of global search and development for TikTok, has been appointed to oversee the large language and image model teams.
• M6, an abbreviation of MultiModality-to-MultiModality Multitask Mega-transformer, was introduced by Alibaba’s Damo Academy in March 2021. The model contains 10 trillion parameters and is pre-trained to perform multiple tasks, including generating clothing designs based on prompts, offering concise descriptions of e-commerce goods, and answering common questions, the AI model’s website shows.

Context: Major tech companies and AI startups are chasing experienced AI talent as they rush to develop their own AI offerings.

• Yang resigned from Alibaba for personal reasons last September, following the departures of several scientists, such as Jin Rong, ex-Alibaba vice president and associate dean at Damo Academy, and Wang Gang, head of the group’s auto driving lab.
• Another Alibaba VP, Jia Yangqing, also confirmed he had quit the tech giant on Tuesday, with the aim of pursuing a new AI infrastructure venture.

Kai-Fu Lee, former president of Google China and now CEO of Sinovation Ventures, announced on Monday that he’s building a new AI company called Project AI 2.0 that will focus on developing ChatGPT-like apps, as well as an ecosystem for AI-powered productivity tools. 

Lee shared his thoughts on the latest AI trends, including the concepts of AI 1.0 and 2.0 on March 14, at Sinovation Ventures’ headquarters in Beijing. He said he considers ChatGPT to be a major breakthrough in deep learning, driving AI into the 2.0 era.

Why it matters: As a renowned AI expert and venture capitalist, Lee said he sees AI as providing an opportunity to reconstruct almost all existing applications, just as Microsoft redesigned Microsoft Office into Copilot, giving Word, Excel, and other mainstream productivity tools AI and generative capabilities.  

• Lee founded Sinovation Ventures in 2009. One of its AI startups, AInnovation, was founded in 2018 to develop artificial intelligence products for manufacturing, telecommunications, and finance industries, and went public in Hong Kong last January.

Details: Lee is currently seeking global talent in the fields of large language models (LLMs), natural language processing (NLP), multi-modality, AI algorithm, and infrastructure. The newly established company is also seeking fundraising.

• Sinovation Ventures has confirmed that multiple technical experts with experience in leading teams in major tech companies have expressed interest in joining the project, according to Chinese media outlet Yicai.
• “Every field can rewrite their existing apps with the aim of creating a more profitable model,” Lee said in his Sinovation Ventures speech, adding that the generative capabilities of AI 2.0 will ultimately reduce costs to almost zero.
• Lee sees three phases of AI application development. The first phase of popular AI apps is more likely to appear in content, where the margin for error is more tolerable. AI would then be used in more demanding fields such as finance and education, to assist automatic trading and language teaching for example. The third phase would be automatic AI, with AI capable of being applied wherever it’s needed.  
• Lee also sees AI providing opportunities to companies that are ten times bigger than those of the mobile internet era, and accessible to Chinese participants for the first time. 
• During the speech on March 14, Lee indicated that if he were at Microsoft, he would prioritize transforming Office tools first rather than just focusing on embedding GPT into search, a piece of advice that coincided with Microsoft’s announcement of Copilot, a reworked Office with AI ability, three days later. 

Context: Lee is the latest tech leader in China tech to turn his attention to launching AI startups. Meituan co-founder Wang Huiwen is setting up an AI startup Guangnian Zhiwai to develop an alternative to ChatGPT, with his closest ally Wang Xing, who is also a co-founder of Meituan, committing himself to invest in Series A financing of the startup.

Chinese search giant Baidu on Thursday introduced its artificial intelligent chatbot ERNIE Bot, with the company’s founder Robin Li demonstrating the capabilities of the chatbot in several pre-recorded clips.

“ERNIE Bot is not a tool for China-US confrontation,” Li said at the launch event, claiming instead that it was the result of Baidu’s years of effort in the field of artificial intelligence.

The chatbot service is currently available by invitation only, but Baidu announced that its cloud computing unit will immediately begin offering application programming interfaces (API) to enterprise clients. The company said that 30,000 corporate users applied for the ERNIE Bot Enterprise Edition API testing within an hour of the launch event, however investors appeared disappointed at the lack of a live demonstration, with Baidu’s stock price dipping slightly before the end of Thursday trading. 

Baidu did not specify when its ERNIE Bot service will be publicly available.

Why it matters: Baidu has become the first major Chinese tech company to unveil a comprehensive AI chatbot service that has the potential to rival ChatGPT, with the launch event taking place just a day after OpenAI released its new AI model GPT-4. 

• Li highlighted the commercial potential of the chatbot with Baidu’s cloud service saying that ERNIE Bot will “fundamentally change the rules of the game in the cloud computing industry.” He also predicted that Model as a Service (MaaS) would eventually replace Infrastructure as a Service.
• In 2022, Baidu AI Cloud generated RMB 17.7 billion ($2.57 billion) in revenue, representing a 23% increase compared to the previous year. The figure accounts for 14.3% of Baidu’s full-year revenue.

Details: Li showcased the capabilities of the ERNIE Bot via a series of pre-recorded videos where it was able to perform various tasks including coming up with a name for a newly established company, writing a poem, and generating images as well as videos based on prompts.

• Li noted that Baidu’s chatbot is good at processing Chinese rather than English. During an earnings call in February, he also mentioned that ERNIE 3.0, the underlying technology of its chatbot, is a “very localized” AI foundation model for the Chinese market.
• ERNIE Bot is built on Baidu’s deep-learning model ERNIE, which was released in 2019. It currently has 650 companies signed up as ecosystem business partners who will have priority access to it.
• The absence of a live review during the launch event seemingly disappointed investors, causing Baidu’s Hong Kong shares to decline by 6.36% at the end of Thursday’s session. Despite this setback, the firm’s Hong Kong-listed stock is up nearly 24% so far in 2023.

Context: ERNIE Bot’s launch day landed between two significant events involving other tech giants. US startup OpenAI unveiled its newest and most advanced AI model, GPT-4, without any prior announcement on Wednesday, and Microsoft launched Copilot, an Office suite that utilizes the power of GPT-4, just hours after Baidu’s ERNIE unveiling.

• ERNIE Bot can be applied to a variety of scenarios and applications, including search, AI cloud, and autonomous driving, Li said in his speech on Thursday.
• Baidu reportedly prioritized the development of the ERNIE system and diverted all of its “scarce resources” towards it before the rollout. On Friday, local media outlet 36Kr cited several sources as stating that the Beijing-based company was continuing to train the chatbot right up to the press conference, and has yet to finalize discussions over how the service will be monetized.

READ MORE: Alibaba, Baidu, NetEase, iFlytek…Chinese companies rushing to prove they have tech similar to ChatGPT

MOSS, a ChatGPT-style chatbot platform developed by Fudan University’s Natural Language Processing Lab, crashed a day after its launch on Monday due to high demand, forcing the lab to set up an invitation-only waitlist to use the AI tool.

The model, similar to OpenAI’s ChatGPT, was named MOSS after a quantum computer in the hit Chinese science fiction film Wandering Earth. Rolled out on Monday, MOSS was seen as ChatGPT’s first Chinese rival, but its server crash means it will no longer be directly available to the general public. 

The team at Fudan University apologized for disappointing users of the experimental software, issuing a statement on the chatbot platform that read, “MOSS is still a very immature model and has a long way to go. Our academic research lab is unable to make a model with similar capabilities to ChatGPT.” 

Why it matters: Amid a wave of excitement in the Chinese tech sphere around ChatGPT, MOSS’s overwhelmed server illustrates the technical and modeling challenges that Chinese research teams and tech companies will need to overcome before they can successfully develop and operate their own ChatGPT-like tools.

• MOSS has followed in the footsteps of ChatGPT almost entirely, a core member of the Fudan University team Zhang Qi told local media outlet Yicai. Zhang stated that gathering real data from users to enhance the model and assist in subsequent research was the team’s main incentive for opening MOSS to the public.  

Details: MOSS is able to perform a variety of natural language tasks based on user instructions, ranging from answering questions to text generation and summarization, as well as code reviewing and generation, according to examples posted by the Fudan MOSS team on GitHub.

• ChatGPT has up to 175 billion parameters, while MOSS has only 16 billion.
• Currently, MOSS performs better in English than Chinese because its model base has learned over 300 billion English words, and only about 30 billion Chinese words, The Paper reported on Tuesday, citing the research team.
• The Fudan team is set to make MOSS’s code and model parameters open-source after their initial validation is completed, and said on Github that this could be as soon as March. 

Context: At least ten Chinese tech companies are competing to develop large-scale conversational language models like the ChatGPT, but none have launched products yet. Baidu’s ERNIE Bot is expected to launch as soon as March.

• ChatGPT is estimated to have reached 100 million monthly active users in January, and is said to have crashed its servers several times in its first week due to mass user demand.

READ MORE: Alibaba, Baidu, NetEase, iFlytek…Chinese companies rushing to prove they have tech similar to ChatGPT

Baidu is weighing whether to embed its ChatGPT-like service Ernie Bot into its main search platform, similar to Microsoft’s Bing, or launch it as an independent service, local media outlet 36Kr has reported, citing multiple Baidu employees.

Why it matters: Baidu’s reported indecision on Ernie Bot reflects the unpredictability surrounding the launch of artificial intelligence-powered chatbots, as the search giant looks to avoid some of the missteps seen with Bing and Google’s recent projects.

Details: Baidu, which has increasingly put AI capabilities at the heart of its strategy in recent years, announced in early February that its Ernie Bot tool would go live in March.

• Led by Baidu’s chief technology officer Wang Haifeng, the Ernie Bot team includes Wu Tian, supervisor of the company’s deep learning platform PaddlePaddle, as well as Wu Hua, technical leader of Baidu’s natural-language processing department.
• Ernie Bot is currently Baidu’s highest-priority project, according to 36Kr. The report cited Baidu employees as saying that all of the “scarce resources” being used to train deep learning models have now been diverted to its large-scale machine-learning Ernie system, which will power Baidu’s version of ChatGPT.
• The company is reportedly also seeking to combine AI-generated content with its blog-style platform Baijiahao as well as with short video content.
• Nearly 300 leading companies from sectors spanning internet, media, insurance, and auto have announced their inclusion in Baidu’s ChatGPT-style Ernie Bot ecosystem.

Context:  Baidu is part of a crowded field of Chinese companies to have announced their own AI chatbot projects in the wake of the huge popularity of ChatGPT.

• The search giant has also announced that its first electric vehicle model will launch using its new conversational artificial intelligence technology, with the intention of providing a ChatGPT-like experience that enables natural conversation between owners and their vehicles.
• AI-powered chatbot services rely on large and diverse data sources to train. Based on PaddlePaddle and the Pengcheng Cloud Brain II computing power system, Baidu launched the largest single model in the Chinese language at the end of 2021 with 260 billion parameters, a parameter figure that has exceeded GPT-3’s total by 50%, Baidu said.
• Meanwhile, Baidu has built a knowledge database of over 5 billion entities and 550 billion facts in recent years as the company’s largest business – search – has generated a huge pool of resources.

READ MORE: Alibaba, Baidu, NetEase, iFlytek…Chinese companies rushing to prove they have tech similar to ChatGPT

ChatGPT has become the talk in China’s tech and business communities these days, with major Chinese tech companies racing to prove they have a similar capability or are developing similar services. TechNode talked to John Zhang, CEO of StarBitech, a digital asset startup based in Shanghai and supported by Microsoft for Startups, on why Chinese tech majors are rushing to push out their own versions of ChatGPT. Below is an edited version of the conversation.

1. Why are Chinese tech companies developing their own AI chatbots like ChatGPT? For example, Baidu announced last week that its look-a-like product, ERNIE Bot, or Wenxin Yiyan in Chinese, will be launched in March. 

There are three reasons for this. First, from a market perspective, ChatGPT is currently not available to Chinese users. They can’t use it as easily as overseas users. So it’s inevitable that there will be a local ChatGPT-like service to satisfy demand. 

Second, from a technological perspective, most large language models (LLMs) currently available on the market, like ChatGPT, are trained on English as the primary language. Their natural language processing (NLP) performance in Chinese is still inferior to that of English. So a model trained with Chinese as the primary language will further improve user effectiveness.

The third reason is data security. AI generates content after going through a large amount of data training. And OpenAI seems to gradually shift from being a non-profit project to a market-oriented one, so there could be uncertainty in the future. Additionally, mainland China requires all data to be locally stored, but OpenAI does not have a team in the country, making it difficult to meet regulatory requirements for local data storage and maintenance.

2. Can China’s AI chatbot compete with ChatGPT and its peers? 

In the short term, it’s still difficult for Chinese AI chatbots to compete. OpenAI entered the stage of large-scale GPU cluster training after getting investment from Microsoft. It’s said that OpenAI owns thousands of Nvidia A100 chips, and Microsoft’s billion-dollar investment was mostly in Microsoft’s Azure cloud resources. Microsoft and OpenAI have just begun the next round of financing and collaboration, which means that in three years, they have burned billions of dollars in cloud resources on training. Such a large-scale investment is very rare in China’s internet circle, especially in underlying infrastructure technology. Most of the big investments in China are more focused on the application side. 

But in the long run, China’s AI chatbot will become more powerful in the future. The country has superior algorithm engineers, a unified large market, abundant application scenarios, and data sources, and cost advantages over Microsoft Azure compared to Alibaba Cloud and Tencent Cloud. 

3. Do you think China is ready in terms of big data and language models?

In terms of big data, China is ahead of the game. It’s highly digitized, so has access to abundant data and a complete industrial chain. However, when it comes to language models, there’s still room for improvement. Currently, models like GPT-3.5 used in chatGPT are large models that require significant investment and are slower in seeing returns, which isn’t an attractive option for many Chinese investors. As a result, only a few major internet companies have participated, with limited investment, slowing China’s progress in language models. But the popularity of ChatGPT offers a good warning for both Chinese investors and internet companies. I expect to see larger investments in the future.

4. How would Chinese AI chatbots differ from others, regarding application and regulations? 

Currently, in China, large-scale chatbots are applied in NLP tasks such as machine translation, intelligent customer service, and Q&A platforms. As the development of LLM progresses, China will also popularize AI chatbots based on LLM. 

AI chatbots developed in China should be: first, eloquent in Chinese expressions. That is, they need to be able to understand Chinese commands. In addition, for a better communication experience, the chatbot must have knowledge of Chinese culture and history, and communicate in a way that fits the Chinese language style and expression. For example, the same word may have different meanings and emotions in different contexts. Furthermore, the chatbot will provide more personalized services based on Chinese users’ habits and needs, such as different payment methods or ethnic customs unique to China.

Chinese-developed chatbots also need to comply with Chinese laws and regulations, including its Data Security Law, Cybersecurity Law, Personal Information Protection Law, and Administrative Measures for Internet Information Services. These laws aim to protect personal information (prevent its illegal acquisition, use, and dissemination), prevent information leaks and misuse, safeguard network security, prevent network attacks and fraudulent activities, and regulate internet information services. With the increasing popularity of chatbots and the continuous improvement of Chinese laws and policies, it is expected that more comprehensive and targeted regulations will be developed in the future to regulate chatbots.

5. Has your team used GPT (Generative Pretrained Transformer, OpenAI’s language model upon which ChatGPT is developed)? What challenges and limitations do you see with this tool?

• Biases. The model is trained on a large amount of text data. If trained data contains biases, the model will also exhibit them. For example, if there is a lack of Chinese language data, particularly in Chinese history, culture, and society, the model may output biased information.
• The model lacks a broad, bird-eye view perspective. Although GPT can maintain a sense of coherence in context, it lacks the ability to think more broadly. 
• Lack of language diversity. GPT is trained mostly based on English material, limiting its compatibility and understanding of other languages.
• High computation cost. GPT is a very large neural network model, with parameter counts ranging from millions to tens of millions. The model size ranges from tens of megabytes to several gigabytes, going up to hundreds of gigabytes. Training such a model costs a significant amount of computing resources and time.

6. Has your team used any China-developed AI language models? How do they compare to GPT?

Currently, with self-developed Chinese AI language models: 

• Some can support different voice responses, which are not currently supported by GPT.
• Regarding language support, there is a greater focus on Chinese-language communication, while GPT has a deeper understanding of English.
• In the application field, Chinese models are more narrowly focused on dialogue generation. To compare, GPT is a language generation model that can be used in text generation, code writing, and more.
• In terms of communication, Chinese models tend to deliver short-sentence communication, while GPT has a strong understanding of long sentences.

7. What are some features or functions that your team would like to achieve using AI language models, but have yet to do?

Current AI-powered chatbots may have achieved impressive results, but there is still room for improvement. One area is the understanding of context and emotions. Chatbots have a limited understanding of things such as one word having different meanings based on the context. 

Another issue is that chatbots can lack coherence in continuous communication on the same topic. Moreover, they lack creativity, as they primarily integrate and sort existing knowledge. This means they do not meet the requirement for independent thinking and creating new ideas.

8. Could you give us an introduction to your company?

StarBitech is a digital content asset technology company founded in 2015. It is jointly invested in by the Shanghai Tree-Graph Blockchain Research Institute and Fengyuzhu and is located at the Microsoft Accelerator in the Caohejing Development Zone in Shanghai. The company focuses on providing individuals and businesses with algorithm-driven digital asset creation and publishing services. StarBitech has worked with companies such as China Merchants Bank, Huawei, LVMH, Shanghai Public Security Jing’an Branch, and the Shanghai Technology Exchange.

The company has recently received support from Microsoft and OpenAI and will leverage its strengths in Chinese natural language processing and local compliance to develop AIGC (AI-generated content) services in fields such as chatbots, visual content creation, and marketing content creation. These services will be supported by GPT, DALL-E, and reinforcement learning, providing AI capabilities for industries such as marketing, gaming, animation, culture and tourism, and government.

Despite not being officially available in China, the AI chatbot service ChatGPT has dominated headlines in the country. This week, days after search engine giant Baidu announced it will launch its own ChatGPT-like service in March, at least five other major Chinese tech firms revealed plans to tool up with the powerful AI technology. 

Starting with Alibaba, the e-commerce giant Alibaba said it is developing its own AI chatbot. NetEase’s online learning unit Youdao said it will launch a similar AI service focused on the education industry, and JD, another e-commerce major, boasted that its rich experience in AI means it can soon incorporate these technologies into its services. 

Developed by OpenAI, ChatGPT is an AI chatbot that can answer natural language questions with human-like responses. It is built on GPT-3, the third iteration of a language model trained on a large amount of data. 

The feverish popularity of ChatGPT has sent investors chasing related stocks on China’s stock market. The market is already experiencing a boost in so-called “ChatGPT concept stocks.” 

On Chinese social and search platforms, ChatGPT has also become the top search keyword. On Feb. 4, daily searches for “ChatGPT” on WeChat increased 515.7% to nearly 38 million, and the search volume kept growing rapidly in the following days,  seeing 2.5 times the number or 95 million searches only five days later. 

As advanced AI technology gains momentum to disrupt the status quo, Chinese tech companies are not the only ones racing to prove their ChatGPT-like abilities. Google introduced on Tuesday its AI chatbot Bard,  while ChatGPT’s main investor Microsoft launched a new version of its search engine Bing on Tuesday with ChatGPT built in. 

Baidu: Baidu said on Tuesday that it will launch its own AI chatbot tool called “ERNIE bot” or Wenxin Yiyan in Chinese. The bot will be built based on the company’s large language model ERNIE, which was launched in 2019. Some see Baidu’s service as the most likely one to come close to ChatGPT. 

NetEase: NetEase’s online education team Youdao said it has been working on applying AIGC (AI-generated content) technology to teaching scenarios such as AI oral English teaching and Chinese essay revision. The company expects to launch a relevant demo version of the product soon, which will mark the first landing of AIGC technology and a ChatGPT-like model in China’s online education scene.

iFlytek: Responding to investors’ questions, the company that specializes in speech recognition and natural language processing technologies said it has a solid accumulation of relevant AI technology. For example, in 2022, iFlytek won first place in the authoritative evaluation of several cognitive intelligence fields such as CommonsenseQA 2.0 and OpenBookQA. Meanwhile, iFlytek has developed a series of pre-training language models which include 40 general fields of cognitive intelligence.

Alibaba: The online retail major said on Wednesday that it’s conducting internal testing on a ChatGPT-like service, and the tool is likely to be used in combination with the group’s workplace communication and collaboration tool DingTalk.

JD: Beijing-based e-commerce platform JD said it sees ChatGPT as an “exciting and cutting-edge exploration,” adding it will incorporate the related methods and technology into its products, especially in customer service.

Chinese search engine giant Baidu is working on a ChatGPT-like bot service to embed in its search engine. Baidu’s CEO Robin Li believes that artificial intelligence tech has reached a tipping point and will produce “a generational revolution in the search experience,” China Star Market reported Monday, citing unnamed sources. 

Why it matters: With its own deep-learning platform (PaddlePaddle) and large-scale pre-trained model (Wenxin or Ernie in English), Baidu is one of the few Chinese tech companies investing heavily in generative AI as the technology becomes more mainstream. 

• Led by Baidu’s mobile eco-business group as well as the technology platform group, the development of an AI-powered chatbot tool project started last December, but the gap between its current performance and ChatGPT is still obvious, a staff member from Baidu told local media outlet Caijing.

Details: ChatGPT, an artificial intelligence chatbot service debuted by OpenAI last November, is built on OpenAI’s large-language model GPT-3, and can output human-like responses in seconds. The aim is for Baidu’s version, which is using its large-scale model Ernie as a foundation, to do the same. Baidu’s upcoming chatbot is being trained with both Chinese and English sources, according to the Wall Street Journal.

• Robin Li is confident that Baidu will be able to embed ChatGPT-like technology into its search engine, admitting that integration is more complicated than the chatbot technology itself, China Star Market wrote, citing a speech by Li given at an internal talk.
• The implementation of a ChatGPT-style service embedded in search services still depends on Baidu’s subsequent research and development investments, a source close to the firm told the outlet.
• For years, Baidu’s research expenses as a percentage of total revenue have remained at around 20%. The company also saw its new AI businesses account for an increasing share of its revenue in the last three years.
• Baidu declined to comment on the chatbot service when reached by TechNode on Tuesday.

Context: Since launching in late 2022, ChatGPT quickly sparked wide discussion, attracting 1 million users in just five days, and pushing many tech companies to prepare for the revolutionary potential of artificial intelligence technology.

• On Jan. 23, Microsoft announced it would offer a multi-year, multi-billion dollar investment in OpenAI, and plans to integrate ChatGPT into its own search engine, Bing.