Generative AI A Powerful Tool but with High Error Rate in ChatGPT's Answers

TapTechNews May 25th news, Generative AI has become a powerful development tool, which can assist developers in quickly generating relevant codes, allowing developers to gradually transition from 'typing code' to 'eviewing code'. However, the latest report points out that the content error rate of programming questions answered by ChatGPT is as high as 52%.

A research team from Purdue University attended the Computer-Human Interaction Conference in early May and submitted a report, examining 517 StackOverflow questions answered by ChatGPT and found that the error rate is as high as 52%.

The survey results show that human programmers prefer the answers of ChatGPT in 35% of cases, believing that the content it provides is more comprehensive and has a clear language style; even worse, in 39% of cases, human programmers did not detect the wrong answers of ChatGPT.

Although this is only a study on one aspect of the function of ChatGPT, it does show that generative AI robots are still prone to many mistakes, and humans may not be able to detect these mistakes.

TapTechNews attaches the reference address: