Recently I used chatGPT 4 to write a small project in Perl which involved reading of an exam file (multiple choice), comparing it to a master exam file, and scoring the student exam including fraud detection and statistics.

chatGPT did a great job of providing low-level code such as regex for pattern matching.

However, chatGPT failed miserably when asked for the program logic of a larger task such as reading the whole exam file and categorizing it into questions, answers, and hints.

What I found stunning was that it got the program logic about 80% correct.

However, even when giving it the whole output it seemed unable to adjust the code and sometimes even made it worse. It was like chatGPT hit a roadblock.

Why is chatGPT unable to refine its code even when prompted with the output and specific instructions on how the output should look vs. what it looks like now?



You must log in to answer this question.

Browse other questions tagged .