Recent developments in AI usage for coding have revealed challenges, particularly in the realm of Ubuntu development. A Canonical engineer noted that attempts to modernize the Ubuntu Error Tracker with AI led to “plain wrong” code, especially from Microsoft’s GitHub Copilot. Subsequently, Ubuntu developer Skia explored Google’s Gemini AI to create a helper script for the Ubuntu monthly ISO snapshot releases, including the recent Ubuntu 26.04 “Resolute Raccoon” Snapshot 2. However, similar to prior experiences with Copilot, Skia found that Gemini AI produced poor-quality code, resulting in silly mistakes and poorly named variables. These shortcomings demonstrate that AI and large language models (LLMs) still struggle with effective coding practices for large-scale software projects. The pull request involved showcases the Gemini-generated code and subsequent corrections, underlining the ongoing need for improvement in AI-assisted programming tools for better efficiency and accuracy.
Source link
