AI modifying the Linux kernel? Linus Torvalds' first AI-assisted patch series

By Gi Pyeong Lee
Linux AI Linus Torvalds Programming Open Source
Linus Torvalds reviewing Linux kernel code in front of a computer using AI tools.
AI Summary

Linux kernel creator Linus Torvalds has released a series of 10 patches that resolve memory performance issues using Anthropic's Claude Code, marking a significant milestone in AI collaboration.

Imagine this: What happens if the developer who builds the ‘Linux’ kernel, the heart of every server and supercomputer in the world, asks artificial intelligence (AI) to write code? Linus Torvalds, the creator of Linux, famous for his sharpness and strictness, recently made headlines by releasing a patch series he wrote with the help of the AI tool ‘Claude Code’. Source: Linus Torvalds’s first LLM-assisted patch series - lwn.net

Why is this important?

Linus Torvalds has previously shown a critical attitude toward the indiscriminate use of AI, stating that low-quality AI-generated bug reports have cluttered the Linux security mailing list. [Source: Linus Torvalds: AI bug tools broke Linux security list AI Weekly](https://aiweekly.co/alerts/linus-torvalds-ai-bug-tools-broke-linux-security-list) The fact that he personally utilized AI to write kernel code signifies that AI has begun to contribute directly to improving the most complex and critical technical infrastructure, going beyond simple automation tools.

This serves as a signal to developers that “AI can now become a partner that reviews or improves my code.” Source: Linus Torvalds’s first LLM-assisted patch series - lwn.net It is highly symbolic that collaboration with AI has reached the core of global technical infrastructure.

Easy to Understand

The problem Torvalds addressed this time is a complex memory management feature called ‘folio lookup’. To put it simply, it’s like looking for a book in a library. There are tens of thousands of books (data) in the library (memory), and if multiple people try to find books at the same time, a bottleneck is bound to occur while waiting for their turn.

Existing Linux had to go through a ‘reference counting’ process every time it lent a book, checking one by one, “How many people are using this book right now?” If these checks pile up, the entire library’s operations slow down. To drastically increase the speed of reading data, Torvalds and the AI used a technique called RCU (Read-Copy-Update). Simply put, for ‘Hot Folios’ (popular books) that people look for the most, they changed the design to briefly defer complex verification procedures and provide the data quickly first. It’s like speeding up library loan times by more than 10 times by omitting unnecessary administrative procedures. Source: Linus Torvalds’s first LLM-assisted patch series - daily.dev

Current Situation

Torvalds created an experimental branch named ‘noref-filemap-read-with-llm’ and released a total of 10 patches. Source: Linus Torvalds’s first LLM-assisted patch series - lwn.net He stated that these codes were written using Claude Code under his direction. Source: Linus Torvalds pushed a Claude-assisted kernel branch… - YouTube

In fact, benchmark test results showed tremendous speed improvements in environments with heavy read operations. Source: Linus Torvalds’s first LLM-assisted patch series - daily.dev However, the world of open source is strict. While other developers were reviewing the code, they discovered the possibility of a ‘Race Condition’ (an error that occurs when two tasks access a resource at the same time) in these patches. Source: Linus Torvalds’s first LLM-assisted patch series - daily.dev It is a case that clearly shows why even code written by AI must be verified by humans, even if the performance is excellent.

What will happen in the future?

Currently, these patches are under review for merging and are expected to be included in the merge window (the period when new features are merged into the kernel) for Linux version 7.4. Source: Linus Torvalds’s first LLM-assisted patch series - lwn.net The Linux ecosystem plans to clearly distinguish and tag patches written by machines in the future. Source: Torvalds joked that AI made Linux 7.3-rc2 this big In the future, AI will continue to grow as a ‘smart assistant’ that optimizes the performance of the Linux kernel, but in the end, it will remain the human’s role to finally approve this and filter out errors.

AI’s Perspective

A perspective from a MindTickleBytes AI reporter: Linus Torvalds’ experiment shows that AI has become a partner in ‘engineering optimization’ beyond just a ‘writing’ tool. The fact that errors were discovered is not disappointing; it is a healthy signal proving that collaboration with AI is perfectly settling into the current development culture that undergoes final human review. Rather than a perfect tool, AI as a ‘partner that reduces errors together’ seems more realistic and promising.

References

  1. Linus Torvalds - Wikipedia
  2. Linus Torvalds Working On A Hefty Linux Kernel… - Phoronix
  3. Linus Torvalds is using AI to develop Linux kernel code now - ZDNET
  4. Linus Torvalds Pushed an Claude-Assisted Kernel Branch… - YouTube
  5. [Linus Torvalds: AI bug tools broke Linux security list AI Weekly](https://aiweekly.co/alerts/linus-torvalds-ai-bug-tools-broke-linux-security-list)
  6. Torvalds joked that AI made Linux 7.3-rc2 this big, and nobody checked
  7. Linus Torvalds “Did What I Preach”, Used AI To Make… - Phoronix
  8. [Linux founder Linus Torvalds says AI is a “gateway drug”… TechRadar](https://www.techradar.com/pro/linux-founder-linus-torvalds-says-ai-is-a-gateway-drug-into-programming-but-cautions-about-using-ai-for-serious-work)
  9. GitHub - torvalds/linux: Linux kernel source tree · GitHub
  10. Linus Torvalds says AI is a great coding gateway drug but…
  11. AI changed how you respond to incidents. - The New Stack
  12. Linus Torvalds’s first LLM-assisted patch series - lwn.net
  13. Linus Torvalds Debuts First LLM-Assisted Linux Kernel Patch …
  14. Linus Torvalds’s first LLM-assisted patch series - apexneuralnews.com
  15. [Linus Torvalds posts his first · Hacker News Zeli](https://zeli.app/story/50090842)
  16. Linus Torvalds Experiments with AI-Generated Kernel Code …
  17. Linus Torvalds’s first LLM-assisted patch series - daily.dev
  18. Linus Torvalds’s First LLM-Assisted Patch Series · Discernion
AD
Test Your Understanding
Q1. Which AI tool did Linus Torvalds use for this patch development?
  • ChatGPT
  • Claude Code
  • Gemini
Linus Torvalds utilized Anthropic's Claude Code during the development process of this memory performance improvement patch.
Q2. What is the main problem this patch series aims to solve?
  • Security vulnerabilities
  • Memory cache performance degradation
  • Version control errors
This patch series is intended to resolve a long-standing performance bottleneck occurring in the Linux kernel's folio lookup mechanism.
Q3. What is the current technical assessment of these patches?
  • Perfect and immediately merged
  • High performance but a race condition was discovered
  • Failed to implement functionality
While it showed significant speed improvements in read operations in terms of performance, a race condition was discovered during the code review process.
    AI modifying the Linux kern...
    0:00