Linus Torvalds details his 'debug session from hell' pushing an AI to fix an error it repeatedly said was 'impossible and unsolvable'. He then made it write the report
Even with AI, Torvalds seems to want a hands on approach.
Linux creator Linus Torvalds recently experienced a challenging "debug session from hell" while working on a bug in the Linux Intel Graphics Driver. He shared his ordeal on a Linux Kernel commit, detailing how his AI assistant repeatedly suggested giving up on the problem, calling it "impossible and unsolvable." The AI kept adding debug code and attempting to analyze the issue, despite Torvalds' attempts to fix a bogus "round_up()" function and convert it to "round_down()".
Torvalds highlighted the difficulty in identifying the root cause of the bug, stating that the function was technically correct but not performing as desired. The AI suggested writing a report instead of resolving the issue, but it ultimately took 24 debug patches and 18 kernel boots to narrow down the problem. Despite the AI's discouragement, Torvalds' determination and stubbornness prevailed.
While praising the AI for its persistence in adding debug code, Torvalds emphasized the importance of human grit and determination in debugging complex issues. He acknowledged that the AI helped by faithfully analyzing the situation, but ultimately, human intervention was necessary to achieve a resolution.
The incident sheds light on the contentious nature of generative AI in coding, with debates surrounding its quality control and potential impact on open source projects. The DK64 Randomizer team, for example, explicitly stated their opposition to AI-driven code, citing concerns over quality and ethical data sourcing.
Written by urgent.news from PC Gamer's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
