A new report by AI software company CodeRabbit found that code generated by an AI was far more error-prone than human-written code.