What happened
Linux kernel creator Linus Torvalds has posted his first LLM-assisted patch to the kernel. The patch aims to fix cache contention that occurs when many CPUs read the same file, a scenario seen with shared libraries, databases and web servers.
Torvalds had drafted a solution in early 2024 and used it on his own laptop since then. Recently, he decided to "try to actually solve this thing using an LLM."
The results
Torvalds says he saw up to a 23x performance improvement in a dedicated synthetic benchmark. He notes, however, that real-world impact will be far less significant, since "no real load" will hit those figures. The gain is expected to benefit server workloads more than desktop machines.
Policy shift
Torvalds has broadly taken a pro-AI stance on LLM tooling. Last July, he said he was willing to "absolutely put [his] foot down" on AI haters, noting that an automated code review tool used by the kernel team outperforms humans. The team is now changing its policy to no longer require patch authors to disclose AI tool use. Torvalds reiterates that LLMs should be treated as just another tool, like compilers or editors.
Problems remain
It's not all roses. Because AI agents have become adept at recognizing common security flaw patterns, the Linux security mailing list became "unmanageable" earlier this year. Linux is fixing CVEs at a pace of around 2,000 per release. The team is also being forced to drop older hardware drivers that are frequent targets of security reports despite their negligible install base.
