The ocr example says it recognizes Chinese, but output ignores it - maybe just AI bug in generated examples