A developer going by the name of “BioBootloader” has created a solution for fixing malfunctioning programs using artificial intelligence (AI). The solution is a program called “Wolverine,” which grants Python programs the ability to self-heal similar to the Marvel superhero.
Wolverine uses GPT-4 to detect errors in the script and provide explanations of what caused the issue, editing the code until every problem is resolved, even if multiple bugs exist. The modified code is saved in the original Python file.
BioBootloader demonstrated the program’s capabilities on Twitter using a split-screen display with Python code on the left and Wolverine results on the right in a terminal. The demo showcased a custom-made calculator script that contained intentionally inserted flaws before executing Wolverine.
In March, OpenAI launched GPT-4, a multimodal AI language model accessible to ChatGPT Plus users and beta testers through an API. GPT-4 is capable of completing text-processing tasks such as programming, language translation, and composition by drawing on its understanding of billions of web documents, books, and websites.
BioBootloader, the developer of Wolverine, has made the program’s source code available on GitHub, and believes that the same technique could be applied to other programming languages.
To use Wolverine, users need an OpenAI API key for GPT-3.5 or GPT-4, which may incur charges. Currently, the GPT-3.5 API is available to all OpenAI account holders, but access to GPT-4 remains limited to a waitlist. Although Wolverine is still a rudimentary model, it offers a glimpse of a potential future where software programs can rectify their own issues, including unforeseen problems that may arise after release.
However, the broader implications, safety concerns, and feasibility of this approach require further investigation.