The titel is clickbaitish. It's good that Linus is open to new things and updates his views. As pointed out by the article he is no ai evangelist, but rather pragmatic.
AI help over text isn't that bad...or it's not as bad as the shitty images and videos that people are pumping out. I'm not above asking it for help with something linux-y 'cause I'm a noob and people on forums tend to be a little mean sometimes. :P But I actually attempt to memorize and learn what it's telling me and also I check its sources.
Same here. The AI answer doesn't fix or solve anything for me, but it does give me enough pointers for me to refine my search or find old posts that leads to the right information.
It's too bad that the big AI services are world destroying
Edit: also most of my noob Linux questions I just don't know how to ask. Google search being the shit fest it is now, duckduckgo I feel like doesn't work as well as google when they're quietly enshittifying (I thought google search peaked at 2010s for how good it is finding me things with all the "google-fu" I learned that's for nought now). And quote frankly, I am "afraid" to ask in Linux forum sites (including here). I don't want to gamble getting some asshole Linux guy mouthing me off.
But "90% marketing" and AI being useful isn't mutually exclusive.
I'd say AI is even 95% marketing but I still use it for the other 5%, where it's actually useful. It's just waaaaaay too expensive for the 5% use case usually.
I'd go for smaller, more specialised models instead of trying to fit everything in a broadly incompetent jack of all trades, master of none type of LLM like they're currently doing.
Yes, something like that. If you use a model which is specifically trained on C(++), Assembly and Rust, as well as scientific articles about Linux and the Linux Kernel, has the whole Linux Kernel already baked in, you can probably get faster and more accurate results than if you add the whole Wikipedia in 20 languages.
My LLM doesn't need to answer all the questions, it's enough if it answers the ones I want to ask. AI and the (mathematical) algorithms used for it are very old in terms of computer timeframes (think the 70s to 90s). The new part with LLMs is actually in the first "L": large datasets. I think we have gone too large.
And, with Linux, you have an issue with authorship if you train on copyrighted data. Reproducing code 1:1 from a non-Free textbook might be problematic, so special LLMs trained on Open/Libre texts will be safer in terms of compliance as well.
Being slightly more useful isn't the only problem AI needs to solve. What about morality issue? Sure, slavery is very useful, but some of you, including me, ban Nestlé products. Why do you not have the same standard with products that are produced by, or assisted in production with AI?
I'm not trying to defend him, but it seems like he was moreso saying linux will never be a reactionary tool of principle. It seems more like an admittance that the AI is here to stay and theres little point in trying to pivot everything to reverse an unstoppable trend?
For aiding programming it beats stackoverflow (thank goldness I dislike SO ...)
But be prepared for it to lie, make stuff up, loose context etc.
AI works best if you show it the way. Hold its hand and lead it to where it should go.
Otherwise it'll cry in the corner talking to itself.
Also it cannot create real new data, only glue old datasets together with pva
Ofc, but even before LLMs, executives wanted to get things fast in any way possible, allowing for crappy code to get shipped. LLMs made it far easier and likable since "AI SMART"
I think you may be misunderstanding the situation. He's telling AI critics to stop complaining or fork. If that happens, the main branch will still have AI submissions, and the fork will be the AI-free one.
Important to note: Torvalds is not an LLM evangelist, nor does he just accept AI slop uncritically. The official policy is that LLM written/assisted code is fine, but the quality must hit the bar expected of Linux kernel submissions, and the submitter takes full responsibility for the code correctness whether they personally wrote it or not. Just submitting LLM code with no verification or tests is a great way of getting banned from submitting pull requests and getting a very angry email from Linus.
And also he’s not advocating for everyone to start using it, rather saying that as the primary kernel maintainer, he’s finding it useful to help do a first pass triage on the flood of PRs.
He did not say that. He said that he only uses git and email as primary tools to work on Linux.
He said he does use LLMs for personal projects to test out ideas.
What he said in this case is that people who contribute to Linux can use AI if they wish as long as their code is good and that nobody should attack them for using AI.
29 Comments
jokro@feddit.org · 72 pts · 25d
The titel is clickbaitish. It's good that Linus is open to new things and updates his views. As pointed out by the article he is no ai evangelist, but rather pragmatic.
His sent email
gilokee@lemmy.world · 15 pts · 25d
AI help over text isn't that bad...or it's not as bad as the shitty images and videos that people are pumping out. I'm not above asking it for help with something linux-y 'cause I'm a noob and people on forums tend to be a little mean sometimes. :P But I actually attempt to memorize and learn what it's telling me and also I check its sources.
JayGray91@piefed.social · 3 pts · 25d
Same here. The AI answer doesn't fix or solve anything for me, but it does give me enough pointers for me to refine my search or find old posts that leads to the right information.
It's too bad that the big AI services are world destroying
Edit: also most of my noob Linux questions I just don't know how to ask. Google search being the shit fest it is now, duckduckgo I feel like doesn't work as well as google when they're quietly enshittifying (I thought google search peaked at 2010s for how good it is finding me things with all the "google-fu" I learned that's for nought now). And quote frankly, I am "afraid" to ask in Linux forum sites (including here). I don't want to gamble getting some asshole Linux guy mouthing me off.
communist@lemmy.frozeninferno.xyz · 61 pts · 25d
This is misleading, he's talking about code review tools not vibecoding...
ranzispa@mander.xyz · 0 pts · 25d
Nowhere in the title or the article is vibe coding mentioned.
db2@lemmy.world · 22 pts · 25d
That's the point.
helix@feddit.org · 23 pts · 25d
But "90% marketing" and AI being useful isn't mutually exclusive.
I'd say AI is even 95% marketing but I still use it for the other 5%, where it's actually useful. It's just waaaaaay too expensive for the 5% use case usually.
I'd go for smaller, more specialised models instead of trying to fit everything in a broadly incompetent jack of all trades, master of none type of LLM like they're currently doing.
DevDave@piefed.social · 3 pts · 25d
like preferably a model only trained on one language kind of idea?
helix@feddit.org · 1 pts · 25d
Yes, something like that. If you use a model which is specifically trained on C(++), Assembly and Rust, as well as scientific articles about Linux and the Linux Kernel, has the whole Linux Kernel already baked in, you can probably get faster and more accurate results than if you add the whole Wikipedia in 20 languages.
My LLM doesn't need to answer all the questions, it's enough if it answers the ones I want to ask. AI and the (mathematical) algorithms used for it are very old in terms of computer timeframes (think the 70s to 90s). The new part with LLMs is actually in the first "L": large datasets. I think we have gone too large.
And, with Linux, you have an issue with authorship if you train on copyrighted data. Reproducing code 1:1 from a non-Free textbook might be problematic, so special LLMs trained on Open/Libre texts will be safer in terms of compliance as well.
racketlauncher831@lemmy.ml · 12 pts · 25d
Being slightly more useful isn't the only problem AI needs to solve. What about morality issue? Sure, slavery is very useful, but some of you, including me, ban Nestlé products. Why do you not have the same standard with products that are produced by, or assisted in production with AI?
alcea@feddit.org · 3 pts · 25d
If you take (read: steal) a little its a crime. If you pilfer the internets EnTiRe Dataset, its an innovation.
Wat
Dialectical_Specialist@quokk.au · 9 pts · 25d
I'm not trying to defend him, but it seems like he was moreso saying linux will never be a reactionary tool of principle. It seems more like an admittance that the AI is here to stay and theres little point in trying to pivot everything to reverse an unstoppable trend?
Tenderizer78@lemmy.ml · 3 pts · 25d
Defend him from what?
Dialectical_Specialist@quokk.au · 3 pts · 25d
Defend him against the many that disagree or otherwise anti-support his comments in this article. Apologies if you were just being rhetorical
alcea@feddit.org · 9 pts · 25d
For aiding programming it beats stackoverflow (thank goldness I dislike SO ...)
But be prepared for it to lie, make stuff up, loose context etc. AI works best if you show it the way. Hold its hand and lead it to where it should go.
Otherwise it'll cry in the corner talking to itself.
Also it cannot create real new data, only glue old datasets together with pva
joeljoelle@piefed.blahaj.zone · 5 pts · 25d
hey freeBSD how you doin'
lonesomeCat@lemmy.ml · 6 pts · 25d
BSD gaming when?
geneva_convenience@lemmy.ml · 3 pts · 25d
Because AI used to suck and it no longer sucks as hard.
lonesomeCat@lemmy.ml · 9 pts · 25d
Oh it does, especially in the wrong hands
Eggymatrix@sh.itjust.works · 3 pts · 25d
The point is that in the correct hands it does not.
lonesomeCat@lemmy.ml · 2 pts · 25d
Then how do you stop its wrong-hand usage ?
ButteredBread@sh.itjust.works · 1 pts · 25d
ButteredBread@sh.itjust.works · 1 pts · 25d
MousePotatoDoesStuff@lemmy.world · 0 pts · 25d
Insisting on high standards for code readability and quality - even higher than before GenAI was a thing - could help.
lonesomeCat@lemmy.ml · 1 pts · 25d
Ofc, but even before LLMs, executives wanted to get things fast in any way possible, allowing for crappy code to get shipped. LLMs made it far easier and likable since "AI SMART"
MousePotatoDoesStuff@lemmy.world · 2 pts · 24d
Thankfully, Linux doesn't have executives and shareholders like Windows does... right?
helix@feddit.org · 2 pts · 25d
That's why the Linux Kernel isn't run by executives. They want quality.
middlemanSI@lemmy.world · 2 pts · 25d
If they fork, they leave the main branch AI free, doing their own thing. This way the main branch does not go down the toilet with AI. This is good.
Barbarian@sh.itjust.works · 24 pts · 25d
I think you may be misunderstanding the situation. He's telling AI critics to stop complaining or fork. If that happens, the main branch will still have AI submissions, and the fork will be the AI-free one.
Important to note: Torvalds is not an LLM evangelist, nor does he just accept AI slop uncritically. The official policy is that LLM written/assisted code is fine, but the quality must hit the bar expected of Linux kernel submissions, and the submitter takes full responsibility for the code correctness whether they personally wrote it or not. Just submitting LLM code with no verification or tests is a great way of getting banned from submitting pull requests and getting a very angry email from Linus.
BartyDeCanter@piefed.social · 4 pts · 25d
And also he’s not advocating for everyone to start using it, rather saying that as the primary kernel maintainer, he’s finding it useful to help do a first pass triage on the flood of PRs.
ranzispa@mander.xyz · 13 pts · 25d
He did not say that. He said that he only uses git and email as primary tools to work on Linux.
He said he does use LLMs for personal projects to test out ideas.
What he said in this case is that people who contribute to Linux can use AI if they wish as long as their code is good and that nobody should attack them for using AI.