The chatbot is not officially available in mainland China or Hong Kong, but users are finding ways round the problem. The service is being used for everything from coding to fitness – and even to cheat in exams on Marxist theory
ChatGPT is a language model trained by the research and development company OpenAI to generate humanlike text. The service attracted 1 million registered users in its first week after launching last month.
However, the service is not currently supported for users with IP addresses in mainland China or Hong Kong – as well as in countries such as Iran, Russia and parts of Africa.
The company has not yet explained why it is not supported in China.
But young Chinese users have found ways round the problem such as using VPNs or overseas phone numbers to access the service. In other cases, community developers have set up a mirror site, while others add ChatGPT to their browser’s plug-in, and compare it with the results given by the search engine.
Users have been sharing tips on how to use the chatbot, with some using it as a movie critic, career counsellor, dentist, investment manager or dream interpreter.
Others are using it as a personal trainer, with one Weibo user writing: “It gave me a suitable exercise plan and nutrition advice depending on my current fitness level, goals and lifestyle habits.”
A Chinese programmer shared his experience of using ChatGPT to write code in Visual Basic to automatically manipulate Excel files on a video he uploaded.
“I got the code I wanted by simply describing my needs, it greatly improved my work efficiency,” he said.
Perhaps inevitably, others said they were using it to cheat in exams. “I used it to write a 3,000-word thesis for a course in Marxism, and it successfully passed the plagiarism detection,” one student posted on Weibo.
Chinese technology companies and computer scientists are closely monitoring public interest in the field of artificial intelligence generated content (AIGC), and following closely behind.
“ChatGPT seems to open a window to artificial general intelligence. Foundation models tend to have the ability of reasoning, induction and understanding of common sense,” said a computer science professor, who spoke on condition of anonymity.
He added while the technology was now showing the ability to develop “semantic understanding”. He said that in the past it could be trained to generate a cat, but might not understand what a cat is – but this may no longer be the case.
However, the technology is not without flaws. ChatGPT answers users’ questions by learning from massive amounts of text on the internet, and making guesses based on probability.
However, unlike a search engine, it still cannot answer questions based on real-time information and sometimes the answers it gives will sound plausible but will be incorrect.
It has also been “trained” using data from 2021 and before, so cannot answer questions about things that happened in the last two years.
Similar AIGC applications include Dramatron, a script writing tool from DeepMind, and Google’s medical chatbot Med-PaLM, which passed the US Medical Licensing Examination with the highest score in history last month.
Chinese tech-companies are also developing their own models. Baidu is developing a service called ERNIE 3.0 Zeus for natural language processing and ERNIE-ViLG 2.0 to generate pictures from descriptive text. It has also developed a video editing tool called VIMER-TCIR.
“The various comprehension and generation capabilities involved in ChatGPT can all be found in the ERNIE model,” Wu Hua, the technical leader of Baidu’s natural language processing department said last week.
“Chinese companies pay more attention to the needs of users and truly applying AI models in products.”
The professor said Chinese developers have not come up with their own version of ChatGPT because they “focus on solving realistic problems and promoting the actual implementation of AI” instead.
Source: South China Morning Post