ChatGPT capitulates to programming questions
According to researchers at Purdue University, OpenAI’s generative AI chatbot ChatGPT is only of limited use as a source of answers to programming questions. In tests, the tool produced a relatively high number of incorrect results when generating code snippets for software projects—particularly on Stack Overflow.
The Results of the test series were sobering, because the chatbot answered 48 percent of the questions posed correctly, not even half. In addition, 78 percent of the answers were inconsistent on various levels compared to human answers - but they were more formal, more analytical and the tone was less negative. Only just under one in five results (22 percent) had matches with the answers given by humans.
However, ChatGPT scores a plus point in terms of comprehensibility, according to the report. The bot is 65% understandable and the answers are easy to read. However, the abundance of irrelevant information still remains a problem. (pte/swi)
