The developer under the nickname Biobootloader created the program wolverine , which is debugging Python scenarios and automatizes the correction of errors.
“Start using Wolverine scenarios, and when they” fall “, GPT-4 edits them and explains what has gone wrong. Even if you have many errors, the code will be reinstated until everything is fixed”, – explains the developer.
In the demonstration video, Biobootloader shows a parallel window with the Python code (left) and the results of Wolverine (right). The developer loads his own calculator script, to which it specially adds several errors, and then performs it.
Wolverine launches the code, sees a failure, and then turns to the GPT-4 to try to find out how to correct the error. GPT-4 derives an explanation of errors, shows the changes made, and then restarts the program. When new errors appear, the GPT-4 corrects the code until the script becomes correct, explains Biobootloader.
According to the developer, technology can be applied to other programming languages. For use of Wolverine requires the Openai API key for GPT-3.5 or GPT-4.
Although at present Wolverine is a primitive prototype, such methods illustrate a potential future in which applications will be able to correct their own errors that may arise after deployment. Of course, the consequences, safety and expediency of this method have not yet been completely studied.