Developer creates ‘self-repair’ software that fixes itself thanks to AI – Ars Technica

Zoom in / An AI-generated and humanoid image of “computer programming Wolverine”.

Bing Edwards/Midjourney

Debugging a faulty program can be frustrating, so why not let AI do it for you? That’s what a developer working on “BioBootloader” did by creating it Wolverinea program that could give Python programs “regenerative healing capabilities,” according to reports Hakkadai. (Yes, just like the Marvel superhero.)

“Run your scripts with it and when it crashes, GPT-4 edits it and explains what went wrong,” BioBootloader wrote in a tweet. accompanied Explanatory video. “Even if you have many bugs, it will restart frequently until everything is fixed.”

GPT-4 is a multimedia language model created by OpenAI and released in March. It is available for ChatGPT Plus subscribers and in the form of an API for beta testers. He uses his “knowledge” about billions of documents, books, and websites ripped from the web to perform word processing tasks such as authoring, language translation, and programming.

In the demo video for Wolverine, BioBootloader shows a side-by-side view window, with Python code on the left and Wolverine results on the right in the terminal. It loads a custom calculator script in which it adds some errors on purpose, and then executes it.

“It runs it, it sees the crash, but then it goes and talks to GPT-4 to try and figure out how to fix it,” he says. GPT-4 displays an explanation of the program’s errors, shows what changes it’s trying to make, and then restarts the program. When new errors are seen, GPT-4 fixes the code again, and then it works correctly. In the end, the original Python file contains the changes added by GPT-4.

See also  Google's Pixel 8 Pro may look like a round Pixel 7

The code is Available on githubThe developer says that this technique can be applied to other programming languages. Use of Wolverine requires an OpenAI API key for GPT-3.5 or GPT-4, and usage fees apply. Currently, the GPT 3.5 API is open to anyone with an OpenAI account, but access to GPT-4 is still restricted by Waiting list.

Recently, there are several experiments involving GPT-4 in loops, such as Auto-GPT And BabyAGItry giving GPT-4 more.agentCapabilities that allow it to spin up more GPT-4 instances (agents) to perform several tasks simultaneously or act independently.

Although currently a rudimentary proof of concept, technologies like Wolverine illustrate a potential future where applications might be able to fix their own bugs — even unexpected bugs that might pop up after deployment. Of course, the implications, safety and wisdom to allow this to happen have yet to be fully explored.

Leave a Reply

Your email address will not be published. Required fields are marked *