Linus Torvalds Says Anti-AI Rules Are "Utterly Stupid" #linux #ai #programming
Read the Portuguese translation →
Author: Better Stack · Watch on YouTube
Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.
Original language: English
Original video: Linus Torvalds Says Anti-AI Rules Are “Utterly Stupid” #linux #ai #programming · Channel: Better Stack · 01:37 · Watch on YouTube
Topics: Software engineering with AI, Open source and software ecosystems, AI safety
Summary
Section titled “Summary”Linus Torvalds used an LLM to understand obscure kernel code, rewrite it for readability, and test it, uncovering the reported bug plus three more that had existed since 2013. At the same time, the Linux kernel is receiving many AI-assisted patches, and maintainers are concerned that some contributors do not understand their submissions. Although kernel policy asks contributors to tag AI assistance, the tags appear on only a small share of changes despite widespread LLM use. Torvalds argues that rules banning AI are misguided: what matters is whether contributors understand their work, not which tools they use.
Key points
Section titled “Key points”- Torvalds used an LLM to make an obscure part of the kernel easier to understand.
- A test harness confirmed the reported bug and revealed three additional bugs dating back to 2013.
- The kernel is receiving a large volume of patches from first-time contributors.
- Torvalds said LLMs can do good work but are often used by people who do not understand what they are contributing.
- Kernel contributors are asked to add an “assisted by” tag when using AI.
- Only 1,800 of 18,500 changes in kernel 7.3 included such tags, while around 90% of contributors were estimated to use LLMs.
- A patch was submitted after the summit to remove the requirement to tag AI assistance.
- Torvalds compared bans on AI to rules against using compilers or editors, calling such rules “utterly stupid.”
- His central point is that contributors should understand their work, regardless of the tools they use.
Chapters
Section titled “Chapters”| Start | Chapter | Summary |
|---|---|---|
| 00:00 | LLM uncovers old kernel bugs | Torvalds uses an LLM to clarify obscure kernel code and tests it, confirming one reported bug and finding three more from 2013. |
| 00:33 | AI patches and contribution concerns | The video describes the influx of kernel contributions, Torvalds’ concerns about contributors’ understanding, and the gap between AI use and assistance tags. |
| 01:06 | Torvalds rejects anti-AI rules | Torvalds mocks bans on AI as tools and argues that what matters is whether contributors understand their submissions. |
Quotes
Section titled “Quotes”the code was too obscure for him to follow — 00:00
LLMs do a good job, but they’re being driven by people with no clue — 00:33
Don’t use compilers in any of your kernel submissions. — 01:06
Do you not see how utterly stupid those kind of rules would be? — 01:06
Concepts and entities
Section titled “Concepts and entities”Linus Torvalds, Linux kernel, LLM, test harness, assisted by tag, asoc array
Full transcript (original language)
Section titled “Full transcript (original language)”00:00 Linus Torvalds used AI on his own kernel code, and it found three bugs that have been sitting there since 2013, and now he’s saying the rules against AI are utterly stupid. First up, a developer reported a bug in part of the kernel called asoc array with a patch to fix it, but in his own git tag, Linus says the code was too obscure for him to follow, so he did what he preaches and uses a modern LLM to help make sense of it. So, he had an LLM help rewrite it so it was actually readable, and then run it through a test harness, which not only confirmed the bug that was reported, but also found three other bugs all going back to the very first commit when he added this library in
00:33 2013. Now, at the same time, the kernel is pretty much drowning in AI patches. At the maintainer summit in Prague, there was nearly 800 first-time contributors in kernel 7.3 alone, and according to LWN, Linus said that LLMs do a good job, but they’re being driven by people with no clue, so the time has come to make it harder for people to contribute. The kernel at the moment also asks that anyone using AI adds an assisted by tag to their commit, but in 7.3, only 1,800 out of 18,500 changes had one of those tags, which is about 10%, but they reckon around 90% of contributors are
01:06 using LLMs now. So, after the summit, a patch was sent to remove the need to tag it. There was some pushback on this move, but Linus replied, “I don’t understand this whole concept of don’t use AI for X. Here’s a new rule for you. Don’t use compilers in any of your kernel submissions.” He then says you better not use an editor for your commit messages either, because it makes it way too easy to backspace and hide your mistakes. Do you not see how utterly stupid those kind of rules would be?” His actual point is pretty reasonable. He expects people to understand what they’re doing, and that understanding
01:37 has very little to do with the tools that they actually use. Subscribe for more AI and dev news every week.