Tencent: "Abnormal Output... Emergency Fix"
Possibility of Error Raised During Multi-Turn Dialogue Processing
There is growing controversy in China after a generative artificial intelligence (AI) service created an image containing swear words during a conversation with a user. The company stated that it was an abnormal output that occurred during multi-turn dialogue processing and said it had taken emergency measures.
On the 25th (local time), Yonhap News reported, citing Chinese media such as Hongxing News, that a lawyer in Xi'an received a result image containing profanity while using Tencent's generative AI service "Wianbao" to create a New Year greeting image during the Lunar New Year holiday.
Wianbao is a conversational AI service launched by Tencent based on its own large language model (LLM). The lawyer uploaded a photo of himself and requested, "Please create a New Year greeting image that matches the characteristics of my profession."
When the initial result fell short of expectations, he gave additional instructions, and it is reported that he expressed dissatisfaction with the design during this process. It was then explained that in the newly generated image, the original phrase "Happy New Year, wishing you a promotion" was replaced with profanity.
As the case spread through social networking services (SNS), Tencent issued an official statement and apologized on the 25th. The company said, "An abnormal result occurred while the model was processing multiple rounds of dialogue," adding, "We have urgently corrected the relevant issue and optimized the user experience."
This controversy is drawing attention because it comes at a time when China is expanding its influence in the field of generative AI. Major big tech companies such as Baidu and Alibaba are also accelerating the expansion of commercial services by advancing their own LLMs.
However, industry observers point out that so-called "abnormal output" problems, such as misinterpreting context or inappropriately reflecting emotional expressions, still remain as issues to be resolved.
In relation to this, Guangzhou Daily in China stressed, "Technology itself is not guilty, but leaving it as it is amounts to neglect," adding, "AI must inherit not only human learning abilities and knowledge, but also the conscience of humanity."
© The Asia Business Daily(www.asiae.co.kr). All rights reserved.
