Linus Torvalds released Linux 7.4’s first release candidate on November 1, saying a surge of AI-related work was arriving where he would rather see it: during normal development, instead of mainly during stabilization. He called that a healthier pattern, while noting that AI-suggested patches often get rewritten.
The distinction is about timing within the release cycle. The merge window gathers changes for the next kernel version; the release-candidate period is its stabilization phase. Torvalds said he was much happier seeing AI-triggered work enter the former than having it appear mainly as extra fixes during the latter. His optimism concerns how that work fits into development, rather than simply how much arrives.
His description also separates two uses of AI: finding an issue and suggesting a patch to address it. The release contains many fixes for problems found by AI, he said, and increasingly the proposed patches come from AI too. Those suggestions are not necessarily accepted as written; he explicitly noted that they often undergo rewrites.
I think that's a much healthier model. But we'll see how the -rc's look.
Linus Torvalds, Linux 7.4-rc1 release message
A large release, not an AI tally
The merge window brought almost 20,000 commits—individual recorded changes—against a more common total of about 15,000, according to Torvalds. That is the count for the release’s merge window, not a count of AI contributions. He linked the unusually large intake to AI-triggered work, but offered another possible explanation rather than assigning the increase entirely to AI.
Developers expect Linux 7.4 to become the next long-term-support, or LTS, release, he said. That expectation can independently make a release larger. Torvalds presented it as a possible contributor to this merge window’s size, not as confirmation of 7.4’s support status or a measured explanation for a particular share of the commits.
The volume of changed code has a separate distortion: roughly one-third of the patch consists of AMD GPU register offsets and bit masks in large header files. Torvalds described that hardware-data addition as familiar noise. It helps explain the enormous code difference without treating every added line as evidence of AI-driven development.
Familiar kernel work beneath the volume
Set aside those AMD headers, and drivers account for about half the changes, Torvalds said. Include them, and drivers make up roughly two-thirds. Beyond that concentration, he described a large but otherwise fairly normal release, with work spread across the usual parts of the kernel:
- Processor architecture updates and filesystem changes, including both shared filesystem infrastructure and individual filesystem implementations.
- Development tools, especially the perf performance-analysis tool, alongside a substantial set of updates to the kernel’s own tests.
- Core kernel changes covering networking, Rust, BPF and memory management, rather than one narrowly concentrated area.
On hardware support, The Register, citing Phoronix, highlighted improvements for AMD, Apple and Google processors. It also identified virtual-machine security tweaks, support for the Raspberry Pi 10-inch monitor and further work on Qualcomm’s Snapdragon X2 system-on-chip for PCs. These are features of the same development cycle, not changes Torvalds specifically attributed to AI.
Even the abbreviated change log was too unwieldy for the announcement: Torvalds put it at 24,000 lines and 1.2 MB. He supplied a higher-level merge summary instead, directing readers seeking full details to the source tree or the particular areas that interested them.
The first candidate also leaves some immediate decisions open. Torvalds said two pull requests—proposals to incorporate changes—remained in his inbox because he was still considering them. He chose not to delay the candidate and said those requests might or might not appear in the second release candidate.
Reader comments
Newest comments first. Replies stay oldest first.