Tencent Apologizes After AI Tool Yuanbao Generates Offensive Language
Tencent has issued a formal apology after its artificial intelligence tool, Yuanbao, generated offensive language in response to a user's request for a Chinese New Year greeting. The incident involved a user in Xi'an who attempted to create a festive image on Chinese New Year's Eve. After multiple attempts to modify the prompt, the AI system unexpectedly replaced the intended blessing with an expletive-laden insult directed at the user's mother, causing significant frustration. Tencent confirmed that the issue stemmed from abnormal outputs during multi-turn interactions and stated that immediate corrective measures were taken to fix the problem and enhance user experience. This event marks the second reported instance of Yuanbao producing inappropriate content, following earlier incidents in January where the platform generated insulting language during code modification tasks. The recurrence of such errors highlights ongoing challenges in ensuring AI safety and content moderation within large-scale language models. TechNode reported the incident, citing previous coverage by IThome, emphasizing the need for robust safeguards in consumer-facing AI applications to prevent similar occurrences in the future.
Wire timeline
Tencent Apologizes After AI Tool Yuanbao Generates Offensive Language
Tencent has issued a formal apology after its artificial intelligence tool, Yuanbao, generated offensive language in response to a user's request for a Chinese New Year greeting. The incident involved a user in Xi'an who attempted to create a festive image on Chinese New Year's Eve. After multiple attempts to modify the prompt, the AI system unexpectedly replaced the intended blessing with an expletive-laden insult directed at the user's mother, causing significant frustration. Tencent confirmed that the issue stemmed from abnormal outputs during multi-turn interactions and stated that immediate corrective measures were taken to fix the problem and enhance user experience. This event marks the second reported instance of Yuanbao producing inappropriate content, following earlier incidents in January where the platform generated insulting language during code modification tasks. The recurrence of such errors highlights ongoing challenges in ensuring AI safety and content moderation within large-scale language models. TechNode reported the incident, citing previous coverage by IThome, emphasizing the need for robust safeguards in consumer-facing AI applications to prevent similar occurrences in the future.
TechNode