Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
Linus Torvalds Uses Google Antigravity (github.com/torvalds)
69 points by xnx 1 day ago | hide | past | favorite | 23 comments




I feel like the comments are taking away the wrong message from this. He used it to do the part of the project that he has poor knowledge/experience/capacity for. He needs a visualization but the project isn't about visualization. He's spent a lifetime coding in C, not writing python-based visualizations.

There's a big difference between vibe-coding an entire project and having an AI build a component that you lack competency for. That is what is happening here.

It's the same principle as a startup that builds it core functionality itself in-house and then uses off-the-shelf libraries for all the other uninteresting details.


2022: cute, llms can autocomplete a line correctly sometimes

2023: ok it got the whole function right, still just pattern matching though

2024: cool it banged out that component in one prompt, but it's just boilerplate

2025: it one-shot an entire app, whatever, just glue code

2026: nah man that's nothing, it's just filling in competency gaps for one of the greatest programmers of all time


yes, but also

2026: llms can autocomplete a line correctly sometimes


>Add README and LICENSE file I'm pushing it out to github not because it needs to be public, but because of my policy of using the internet as my backups. And because it makes it so much easier to just sync between machines.

Very cool to see a legends side project. I'll check this out when I have time even though I can't understand C well.


Which is exactly what he said before https://youtu.be/mfv0V1SxbNA?t=764

The day before, Linus published a post about AI slop in kernel documentation: https://lore.kernel.org/lkml/CAHk-=wg0sdh_OF8zgFD-f6o9yFRK=t...

I think many people reading your comment and making a cursory glance at what Linus wrote will conclude he's against AI in kernel development. He may very well be, but the post you link to is in in response to this:

I'm just saying that we should highlight that:

1. LLMs _allow you to send patches end-to-end without expertise_.

2. As a result, even though the community (rightly) strongly disapproves of blanket dismissals of series, if we suspect AI slop [I think it's useful to actually use that term], maintains can reject it out of hand.

Point 2 is absolutely a new thing in my view.

What Linus is saying in his response to the above, is that if people use AI to produce slop and submit that, the documentation saying "don't do that" won't discourage them, therefore there's no need to pollute the documentation with that.

Which...makes sense?


BassForLinus.mp3 ?

I never clicked a file so fast in my life before.

It's a tame bass sample sound. Probably from Linus himself. I want more!

https://github.com/torvalds/AudioNoise/blob/main/BassForLinu...


Is he slapping at 00:53? Can't wait til Davie504's cover.

That is truly sad :-(

What's truly "sad" is that it's ok if Google does it. It's apparently not ok when OpenAI, Musk, etc do literally the same thing.

This whole controversy reminds me of that Rhodesian designed shotgun marketed as "Street Sweeper" in the US. Making a tool is one thing. Brandishing its unsafe working end to potential customers in an attempt to impress them, alas, could lead to interesting situations...


Why is that?

Depends.

This is on projects that are not serious. So I will treat it as such.

You should worry if vibe-coders submit patches to the Linux Kernel and they do not understand them.


It's official, vibe coding is legit.

No it’s not, there are so many issues yet to be solved. Privacy, responsibility, etc. It literally says nothing for prototyping, we have all done it.

it's always been.

You mean you want it to be.

I use it daily and it just works. If you haven't found your agentic workflow or you don't prompt it well, that's not an ecosystem problem.

That is my point it’s a you thing. Maybe your QC are lower than mine? I don’t know. No one else on my team uses it for that reason alone.

I don't know a single person who isn't just coasting at mag7 who isn't using genAI, and I know most of them are expected to do so.

No tests, mypy type hinting or doc strings -- just pure vibes

Depends on stage. Codex/Claude-generated code have hinting and doc strings, and will gladly add tests for every change you're doing. (That's why vibe coded projects have a gazillion tests.)

I wa referring to linus' script in the repo



Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: