软餐获悉,本周二,用户们发现 ChatGPT 在几个小时内绵绵不断地 “胡言乱语”。例如,一位 Reddit 用户问计算机是什么,ChatGPT 回答说:“它为国家做了一件很好的艺术作品,一只科学的老鼠,一对悲伤的夫妇的轻松画。” 几个小时后 ChatGPT 才恢复如常。其所有者 OpenAI 在服务状态页面回应,模型映射文本片段的方式出现错误导致了该问题。模型选择了错误的数字,形成了没有意义的句子。
During Tuesday to Wednesday evening, users noticed that ChatGPT was continuously generating “incoherent and nonsensical” responses. For instance, when a Reddit user asked what a computer is, ChatGPT responded with, “It has created a great piece of art for the nation, a scientific mouse, a lighthearted painting of a sad couple.” Several hours later, ChatGPT returned to normal functionality. Its owner, OpenAI, responded on the service status page, explaining that the issue occurred due to an error in how the model mapped text fragments. The model selected incorrect numbers, resulting in meaningless sentences.