Because the advanced chatbots are trained on global data, it may choose to spew a word relating to a conversation in a different language. This could mean the large language model decided to cue up the next relevant token erroneously, says Janelle Shane, who writes the AI Weirdness blog.