A Generative AI service made by Chinese corporations sparked controversy after it used profanity during a conversation with a user. The developer said the model produced an abnormal result during multi-turn processing and noted it made an emergency fix.
On the 25th, according to Chinese outlets including the Hongxing News, a lawyer in Xi'an said that while using Tencent's Generative AI service Yuanbao during the Lunar New Year holiday to create New Year's greeting images, text containing profanity was generated.
Yuanbao is a Conversational AI service introduced by Tencent based on its own large language model (LLM).
The lawyer uploaded a personal photo to Yuanbao and said, "I am a lawyer, so please create a New Year's greeting image that suits the characteristics of my profession."
When the initial results fell short of expectations, several additional instructions were given, and during this process Yuanbao expressed dissatisfaction along the lines of "What kind of design is this." The lawyer said that in the newly generated image, profanity was inserted instead of the existing phrase "Happy New Year, wishing you a promotion."
After the case spread on social media, Tencent apologized on the 25th and promised to prevent a recurrence.
Tencent said "an abnormal result occurred as the model processed multiple rounds of conversation," and added it "urgently corrected the related issue and optimized the user experience."
The case draws attention because it occurred as China is rapidly gaining visibility in the Generative AI race. Major big-tech corporations such as Tencent, Baidu, and Alibaba are successively advancing their own LLMs and expanding commercial services.
In response, the Guangzhou Daily in China said, "Technology is not guilty, but leaving it as is amounts to laissez-faire," adding, "AI should not only acquire human learning ability and knowledge, but also inherit humanity's conscience."