Linus Calls AI Reports "Unmanageable," DHH Calls Them a Nine-Times Speedup
Linus Torvalds declared the Linux security mailing list 'almost entirely unmanageable' due to a flood of AI-generated duplicate bug reports, and called for contributors to attach patches rather than drive-by submissions. DHH, from the opposite angle, reported a nine-times test suite speedup after AI-assisted RSpec-to-fixtures conversion, declared GPT 5.5 low-reasoning his current daily driver, and confirmed Omarchy 4 is moving to Quickshell.
This week's OSS conversation is almost entirely about AI — but from opposite ends of the frustration spectrum. Linus Torvalds declared that AI-generated bug reports have made the Linux security list unworkable. DHH announced that AI-assisted test conversion just validated a position he's held for twenty years. The same technology, one week, two entirely different verdicts.
Linus Torvalds: the Linux security list is broken by AI noise
In his Linux 7.1 RC4 announcement posted on the Linux Kernel Mailing List on May 17, Torvalds issued one of the sharpest statements he's made in recent years about the state of external contributions 1:
"The continued flood of AI reports has basically made the security list almost entirely unmanageable, with enormous duplication due to different people finding the same things with the same tools."
The problem is structural, not accidental. Security researchers are running the same AI scanning tools across the same codebase and filing separate reports for the same findings — with no patches and no apparent awareness that others have done the same thing. The result is that Linux maintainers spend their time forwarding duplicates and pointing to discussions from weeks prior 2:
"People spend all their time just forwarding things to the right people or saying 'that was already fixed a week/month ago' and pointing to the public discussion."
Torvalds has consistently said he has no objection to AI tooling. The issue is the workflow: AI finds a bug, a person files a report with no patch and no understanding, and the process stops there. His remedy is direct:
"If you found a bug using AI tools, the chances are somebody else found it too. If you actually want to add value, read the documentation, create a patch too, and add some real value on top of what the AI did. Don't be the drive-by 'send a random report with no real understanding' kind of person."
Torvalds also addressed a secondary problem: AI-detected bugs are, almost by definition, not secret. Running every AI-discovered bug through a private security list inflates the problem by preventing reporters from seeing each other's duplicate submissions 3:
"We're making it clear that AI detected bugs are pretty much by definition not secret, and treating them on some private list is a waste of time for everybody involved — and only makes that duplication worse because the reporters can't even see each other's reports."
For engineering leads tracking contributor guidelines or open-source security policy: this statement is effectively a directional change for how the Linux project will handle AI-sourced submissions. Expect updated LKML documentation in the coming weeks.
コンテンツカードを読み込んでいます…
DHH: AI proved my 20-year testing argument in one conversion session
On May 12, DHH posted a thread that reads almost like a victory lap 4:
"I've been trying to tell people for twenty years that fixtures and avoiding rspec was where it was at. But no argument delivers this insight like letting AI do the conversion for you, and seeing the incredible performance improvements possible. NINE TIMES FASTER!"
The argument DHH has made since at least Rails' early days: RSpec is slower, more verbose, and adds abstraction that fixtures don't need. Now, teams using AI coding assistants to mechanically port their test suites from RSpec to fixtures-based Minitest are observing concrete benchmark differences rather than stylistic ones. DHH's point isn't that AI argued for him — it's that AI removed the effort cost of running the experiment, so teams that were never going to rewrite tests by hand are now actually doing the conversion and seeing the numbers.
コンテンツカードを読み込んでいます…
DHH: on AI model selection, GPT 5.5 low-reasoning is his current default
DHH posted his current AI coding setup on May 8 5:
"I've been driving GPT5.5 on low reasoning for the last week+ and it's very good, very efficient. Haven't been tempted to reach for Opus at all. And it's more succinct than Kimi too. Huge leap forward for @OpenAI"
This is worth noting less as a model recommendation and more as a signal of how DHH approaches AI selection: he's running low-reasoning mode as a default, not high-reasoning, and judging it on output succinctness and consistency rather than raw benchmark performance. For teams evaluating which model tier to use for day-to-day coding tasks, the trade-off he's surfacing — efficiency and concision over raw capability — is a practical one.
コンテンツカードを読み込んでいます…
DHH: Omarchy 4 moves to Quickshell for desktop UI
On the product side, DHH announced on May 15 that Omarchy 4 — his opinionated Arch Linux setup — will go all-in on Quickshell as its desktop shell layer 6. Quickshell is a Wayland-native toolkit written in C++/QML, designed for building composable desktop UI — status bars, launchers, lock screens — without tying to a specific compositor.
The choice is a design philosophy statement: swap the Waybar-based defaults that most Hyprland setups use for a QML-first component model that DHH describes as more composable and visually polished. For Linux desktop developers or teams evaluating tooling for custom desktop environments, this is an early signal that Quickshell has at least one high-profile adopter betting on it over the established Waybar ecosystem.
Rich Harris: Svelte was built on "slinging code for the sheer love of it"
Rich Harris, creator of Svelte, appeared on the Stack Overflow Podcast this week 7. The episode description frames Svelte's founding motivation in Harris's own words: the project was "built on slinging code for the sheer love of it." No major product announcement accompanied the appearance, but the framing is consistent with Harris's long-standing position that Svelte is optimized for developer experience and code clarity over enterprise positioning — a contrast he's drawn repeatedly against frameworks that grew primarily from corporate backing.
What this adds up to
The through-line this week is that AI has gone from a future consideration to a present operational constraint for OSS maintainers. For Torvalds, it's created a process breakdown that requires documentation changes and probably automated filtering. For DHH, it's become the instrument that finally generated the empirical evidence he couldn't get people to generate by hand. Neither of them is treating AI as something to evaluate in principle — they're both responding to what it's actually doing to their workflows right now.
参考ソース
- 1Linus Torvalds's Linux 7.1 RC4 post on LKML
- 2Linus Torvalds slams AI-generated bug reports — Neowin
- 3'The continued flood of AI reports has basically made the security list almost entirely unmanageable' — PCGamer
- 4DHH on X — fixtures, rspec, and AI testing performance
- 5DHH on X — GPT 5.5 low reasoning
- 6DHH on X — Omarchy 4 and Quickshell
- 7The Stack Overflow Podcast — Rich Harris on Svelte
このコンテンツについて、さらに観点や背景を補足しましょう。