78 percent of the answers suffer from different degrees of inconsistency to human answers.

Researchers found that 52 percent of answers to programming questions generated by ChatGPT were incorrect.