We recently wrote about Torvalds’ atypically subtle and nuanced position on the use of LLM bots in coding. It seems that the reasons have suddenly become a little clearer.
Google’s Antigravity LLM has been winning other friends of late, including Register columnist Mark Pesce, who wrote that “vibe coding will deliver a wonderful proliferation of personalized software.” Some other big names in the world of FOSS have also come out in favor of LLM coding assistants recently, including Redis creator Salvatore “Antirez” Sanfilippo, who wrote “don’t fall into the anti-AI hype.” Said hype is, of course, a subject about which Torvalds opined previously.
Torvalds’ position has been more moderate, which is not entirely like his former self. He is famed for his outbursts at Nvidia, GitHub, third-party companies, and kernel contributors. We could go on, but you get the picture.
Here is a video about this subject: https://inv.nadeko.net/watch?v=hymd3Xc7cCU This channel (SavvyNik) often directly reads and shows parts of the original mailing list source, and if available the relevant part of the interview in video form.
Maybe its not a bad idea to experiment with the tools that lot of people do, so it might help understanding what code it produces. That might be not his original goal, but its a nice side effect I guess.
Say it with me: It’s not vibe coding if you read and understand the code.
That is not vibes, that is knowing what you’re doing and working based off of a suggestion.
This
I also sometimes use AI for small bits and pieces and the vast majority of the time it gets about 10-20% wrong which I then have to fix
There is a net-gain in time on these small bits and pieces, but it’s not really able to do more than that
If I read it correctly, he used a LLM to help him write Python for a hobby project. I think this falls into an open minded, “who cares?”. Come back when he used it for something intended for public or commercial use.
He never would in critical code such as Linux kernel
I wouldn’t be so sure.
Pragmatically minded folks are not always concerned with the details of how something is done.
After all, he often uses code written by people other than himself in the kernel, he works to ensure those people are well clothed and fed, but that’s hardly the point of the linux project. The user experience is not so different from vibe coding.
Because anyone can fall for stupid shit.
Tldr, he’s fine with it because he was lazy
ITT: Anything that makes work easier for the same end goal is laziness.
The guy literally said that he didn’t want to bother learning python and that it’s why he used this crap.
If that’s not intellectual laziness I don’t know what is.
Don’t be a stupid shit. You use things on the daily without full comprehension of them.
If you write code without understanding what you’re doing, to the point where an AI is achieving better results, you should question your choices.
Today we knew that Linus Torvalds is lazy and stupid ✍️
I think it is intellectual laziness to do Python just for its ecosystem when saner options abound.
