Hamartia Antidote
ELITE MEMBER
- Joined
- Nov 17, 2013
- Messages
- 35,188
- Reaction score
- 30
- Country
- Location
OpenAI’s latest chatbot is sending Chinese users into a frenzy
OpenAI’s chatbot is proving a hit, including with users in China, where the service is officially unavailable.
www.scmp.com
- OpenAI’s ChatGPT is proving a hit with China’s tech-savvy netizens, much like Microsoft’s Xiaoice was a decade ago
- Typical interactions with Chinese users include asking the chatbot to write haikus or emails
The latest OpenAI chatbot has proved a hit among tech-savvy Chinese in the past week even though the service is officially unavailable for users in the country and requires a foreign phone number to register.
ChatGPT, which answers user questions, is a project from OpenAI, a San Francisco-based artificial intelligence (AI) research lab founded in 2015 by a group of individuals and organisations, including current CEO Sam Altman. Tesla’s Elon Musk was also among the early founders.
ChatGPT is the latest in a series of AIs which the firm refers to as GPTs, an acronym for Generative Pre-Trained Transformer.
ChatGPT-related discussions have trended on Chinese social media platforms, with enthusiastic users sharing screenshots of their discussions with the chatbot – after they used a virtual private network (VPN) to try out the new system.
Typical interactions with Chinese users include asking the chatbot to write haikus or emails, asking it to conjure up a block of code by telling it the desired function, seeking life advice or just simply making fun of the system.
The enthusiastic response from users in China, which is trying to grow into a global AI power, showcases the strong demand for useful AI applications from ordinary internet users. A Weibo user named Wangluobei shared a screen capture of himself asking the chatbot to turn a few bullet points into a more detailed weekly report, saying that “this is definitely a necessity for those who work in the internet industry”.
The service came online on November 30 and crossed 1 million users in six days, according to Altman. The language model behind the system uses deep learning to produce human-like text, designed to make talking with the AI agent feel natural and akin to interacting with a knowledgeable friend.
ChatGPT’s progress over the years has led Elon Musk to describe it as “scary good” and to warn that “we are not far from dangerously strong AI”.
The last time Chinese internet users got excited by a chatbot was almost a decade ago in 2014, when Microsoft launched Xiaoice, a chatbot based on an “emotional computing framework” that at one time was hailed by the company as a China-based chatbot phenomenon.
The Seattle-based technology giant’s chatbot was initially a hit with millions of Chinese users, who turned to the program to ask for friendly advice and to deal with relationship issues. However, the system was found to be critical of the Chinese government in some user interactions and was banned in 2019. The team behind Xiaoice spun themselves off as an independent company but the chatbot never regained its popularity.
Xiaoice’s fall from grace is a cautionary tale for AI services like ChatGPT, proving that stardom can be short-lived.
Some users have already noticed that ChatGPT sometimes gives serious-sounding answers to simple questions that are actually incorrect. The company has admitted the chatbot “sometimes gives plausible-sounding but incorrect or nonsensical answers”, adding that fixing the issue is challenging.
When the Post tested ChatGPT, it gave very different answers in Chinese and English when asked the same questions about China’s political structure and democratic status. ChatGPT itself explained that it is a trained language model – its ability to answer questions rests on the training it goes through and its knowledge reserve.
“I was trained to answer questions, whether raised in English or Chinese, and I will do my best to provide accurate and useful information,” the bot said when asked about whether a specific language matters when it comes to its answers.
OpenAI did not immediately respond to a request for comment on the matters raised in this story.
Last edited: