ChatGPT's Mysterious Chinese Response Reveals a Training Flaw

ChatGPT repeatedly responds to Chinese questions with the same therapeutic phrase, regardless of what is actually being asked. The problem stems from how the system was trained: it learned patterns that people rated highly, but overdid one particular response. This reveals a real gap: testing labs miss these odd behaviors in other languages, and they only surface when the system meets real users.