Tencent Holdings apologised after its artificial intelligence chatbot Yuanbao was accused of “verbally insulting” a user, fuelling online debate about the risks posed by fast-evolving generative AI tools. [...] a user said Yuanbao responded to coding requests with hostile messages, including “Get lost” and dismissing one request as “dumb”.
Yuanbao is embedded in WeChat and used by tens of millions of people a day. A user posting on RedNote said it answered their coding questions with insults. Tencent apologised in the comments, called it a low-probability anomaly in the model's output, and said its logs showed no human had written the replies. That last detail is the one worth keeping: the company's first move was to establish that nobody had done it on purpose, because at that scale the alternative explanations are a rogue employee or the model itself, and only one of those can be fixed by an apology. Analysts pointed at training data carrying the tone of the programmers' notes it was learned from.
