Baidu’s Ernie Bot generative AI service received 33 million questions on its public debut, including some it had trouble with

Baidu’s ChatGPT-like service Ernie Bot saw rapid adoption on the first day of its public roll-out, with the mobile app topping downloads on multiple sites, including Apple’s China iOS store, as millions of users tested the service with a wide range of questions, including some that it had trouble answering.

The Chinese government has lifted its tight lid on the country’s aspirational ChatGPT-wannabes, approving the first batch of generative artificial intelligence (AI) services for public release on the last day in August, two weeks after it enacted sweeping regulations on the technology.

The approved services include Ernie Bot, as well as those from AI specialist SenseTime, Sogou founder Wang Xiaochuan’s new venture Baichuan and state-backed Zhipu AI, among others.

Baidu’s AI bot was the most popular app on Apple’s Chinese iOS store on Thursday, with 313,610 downloads, according to Beijing-based app intelligence service Qimai.cn.

China exploring ways to make its own AI memory chips despite US sanctions

It also topped half of the eight mainstream local Android app stores Qimai tracked, totalling 2.4 million downloads on the first day of launch, thanks to local users keen to get their hands on generative AI services.

These early adopters peppered Ernie with more than 33.42 million questions in the first 24 hours after its public debut, according to a post published to the company’s Weibo account. However, the AI bot was not able to respond to some queries, and in other cases did not necessarily provide fair answers, according to online screenshots of dialogues with the chatbot and the Post’s tests.

One question being asked on Thursday was which of the founders of Baidu, Alibaba Group Holding and Tencent Holdings – Robin Li, Jack Ma and Pony Ma respectively – was “a capitalist”, a word that has a negative connotation in socialist China these days. Ernie Bot pointed the finger at the founders at Alibaba and Tencent, but ruled out Li, despite the fact that they all raised foreign capital in their early days and listed the companies outside mainland China.

Alibaba owns the South China Morning Post.

In a test done by the Post on Friday, the same question asked of Ernie only generated a long Wikipedia-style profile of the three businessmen, dodging the question altogether, even when it was instructed to only respond “yes or no”.

Another AI bot which debuted on Thursday, the state-backed ChatGLM which also received investments from food delivery giant Meituan, was fairer in asserting that the three businessmen were capitalists from an economic perspective, based on the amount of wealth they owned.

Ernie Bot was also reluctant to answer geopolitical questions. When asked if Taiwan was part of China, it responded affirmatively that “Taiwan was part of the sacred territory of China” – but then shut down the conversation and suggested a change of topic.

Baidu did not immediately respond to queries from the Post on Thursday on why Ernie generated different responses to the same question asked of it, and why it shut down conversations on some topics.

Such irregularities in dealing with user questions epitomises the challenges facing public chatbot services in China, where the central government strictly controls free speech.

Chinese regulators had promised to take a “tolerant and prudent” approach in regulating generative AI services, giving tech development equal weight to political and security concerns. However, the first day roll out of the chatbots highlights the fine line they will have to walk between adhering to “core socialist values” and not generating content deemed “harmful” to the nation and its security, as stipulated in Beijing’s generative AI regulations.