PC magazine Learn about the amazing discoveries made by computer science researchers at Johannes Gutenberg University and University College London.
“ChatGPT uses sample code to remove errors and fix them better than existing programs designed to do the same thing.
Researchers submitted 40 buggy codes to four different code-fixing systems: ChatGPT, Codex, CoCoNut, and Standard APR. Basically, they asked his ChatGPT, “What’s wrong with this code?” Copy it and paste it into the chat function. In the first pass, ChatGPT worked like any other system. ChatGPT solved 19 problems, Codex solved 21, CoCoNut solved 19, standard APR method solved 7 problems. Researchers have found that answer most closely resembles the Codex. This was not surprising, as “ChatGPT and Codex belong to the same family of language models.”
However, the ability to chat with ChatGPT after receiving the initial answer made the difference, and in the end ChatGPT solved 31 questions, easily outperforming others that provided more static answers. “A strong advantage of ChatGPT is that it allows you to interact with the system in dialogue to specify your requests in more detail,” the researchers report. “We find that in most of our requests, ChatGPT wants more information about issues and bugs. 31 out of 10 bugs have been fixed and surpassed the status quo. – Art…..”
Companies that create bug-fixing software, and software engineers themselves, are taking notice. However, in its current form he said an obvious barrier for tech companies to adopt his ChatGPT for platforms like Sentry is that it is a public database (companies send coveted intellectual property to engineers). This is the last place you want it to be).
