Linux creator Linus Torvalds has shared his experience with an AI coding assistant during a particularly arduous debugging process. In a Linux Kernel commit message, Torvalds detailed a "debug session from hell" that involved identifying and fixing a subtle error in the Intel GFX Driver. The AI, he stated, declared the problem "impossible and unsolvable" on multiple occasions, even suggesting that the team simply write a report about the issue instead.
The specific problem stemmed from a "bogus 'round_up()'" that needed to be changed to a 'round_down()'. Torvalds explained that resolving this issue required a significant effort, involving "24 patches adding more and more debug information" and "18 kernel boots" to finally isolate the problem. This particular bug was not a straightforward syntax error but a functional discrepancy, where the code was technically correct but not behaving as intended, a nuance that proved challenging for the AI.
Despite the AI's repeated insistence on the task's impossibility, Torvalds continued to push for a solution. He noted that the AI, while ready to give up, "did keep adding debug code and analyzing it faithfully when I pushed." He credited this persistence, combined with his own stubbornness, for eventually leading to the fix. Torvalds even allowed the AI to write the commit message itself, acknowledging its contribution.
Torvalds' experience highlights the current limitations of generative AI in complex coding scenarios. While acknowledging the AI's persistence in adding debugging code, the incident underscores the need for human oversight and tenacity. Generative AI in coding, often referred to as 'vibe coding', has been a subject of debate. Last year, Collins Dictionary named 'vibe coding' the word of the year, but concerns persist regarding quality control and the potential for AI-generated code to degrade open-source projects. Some developers argue that over-reliance on AI can lead to lower-quality projects that become increasingly difficult to manage over time, emphasizing that a human touch remains essential in software development.