Torvalds Speaks: Future of AI

[@lwwells]

Linus: "I don't want to be part of the hype" Video Thumbnail: "AI will REPLACE YOU!" [with a photo of Linus]

[@alext3811]

Isn't it just your normal YT clickbait title?

[@luckyLaserface]

Last sane person in the tech industry.

[@flybyray]

9:55 … Lookin' like a true survivor, feelin' like a little kid And I'm still standin' after all this time Pickin' up the pieces of my life without you on my mind … BUT „because if it is only something for me it is not really intersting in the long run“ An answer to the question: What is the natural instinct to help others?

[@agnescroteau8960]

Pinky promise me

[@miraculixxs]

plus me ofc 😊

[@notthere83]

Eh, there are more. Although they ARE usually encouraged to quit...

[@linuxlizard]

I got my job because of Linux. Simply put, I live by Linux like a lot of sysadmin

[@amoldhamale3202]

"Every project I've started is because of me being frustrated with other people being incompetent" 🤣🤣

[@nasko235679]

He's one of the few people on this planet that can say this without sounding arrogant. Dude has literally walked the walk 😂

[@amoldhamale3202]

Agreed

[@NoSpeechForTheDumb]

​@@nasko235679IMO calling the creators of CVS and Subversion "incompetent" is extremely arrogant, especially considering that the core functionality of Git (Check-in, Branches, Tags etc.) was possible with these already. He just extended the tools and made them more flexible. Same thing with Linux in comparison to older kernels. Thorvalds keeps on insulting the giants on whose shoulders he stands.

[@larssimkins3313]

​@@NoSpeechForTheDumb it sounds like you agree that those features were under-developed and difficult to use in the existing tools. 'Incompetent' is a strong word but it's fair to ask why those devs couldn't streamline their tools to the degree git has.

[@NoSpeechForTheDumb]

@@larssimkins3313 no, completely not. I successfully worked with CVS and Subversion for centuries (!). Those tools were fully capable of everything I needed. Eventually I switched to Git because it was used at my new workplace. It surely has benefits, but I don't see the purported revolutionary aspect. IMO it's overexcited by Thorvalds fanboys. Like Linux, the monolithic kernelblob. LOL

[@JarheadCrayonEater]

"let's wait". As a programmer that started in 1986 and went to the Tattered Cover Book Store in Denver in 1993 to buy Linux in a book, hearing him say "let's wait" is very good advice. Don't be so quick to adopt new things in software. Most are just fads.

[@miraculixxs]

💯

[@ABackendDev]

I bought my first, RedHat 5.1 in a box from Barnes and Noble. Came with the entire CPAN- comprehensive Perl archive network- the og npm!

[@johnmckown1267]

Especially the programming language du jour.

[@bobros5]

It's so interesting that great projects like Linux came from the dissatisfaction of one developer. So believe in your own needs, it often means there are enough people who have the same needs.

[@dan-cj1rr]

Thats the guy i'd love as a leader, down to earth, not sleeping with his investors.

[@marcellerusu]

bro, the editing... don't repeat the clips in a 10 minute video please.

[@kj_sh604]

1:01 he deserved more laughs for that, including from the host. That was very well-timed 😆

[@sebbbi2]

And Crypto was followed by Metaverse/VR. Facebook even changed their name to Meta. There’s always some tech hype going on to lure investors money. They don’t seem to afford to not jump early into each bandwagon.

[@ZombieLincoln666]

Yeah, but at least VR it’s a real product that works.

[@sebbbi2]

Yeah, VR has real professional uses. Consumer VR mostly failed because of locomotion issues. You can’t play most existing games in VR as you can’t freely move without puking. AI has professional uses too. For example our trash is today sorted with AI vision algorith + robot sorter. AI also is used to process particle accelerator data: filter massive data sets, find anomalies, search new possible standard models, etc. AI algos are great for many big data optimization problems. The problem is that marketing hype tells us that everybody will soon wear VR glasses daily and spend most of their day in a virtual metaverse. And AI will replace all our jobs in 5 years and super intelligent AGI will wipe out humanity in the next 10 years. You have to invest today to avoid missing big structural changes similar to internet and smartphones.

[@-Engineering01-]

Metaverse technology as a software is there, just hardware isn't enough to build it. It'll be made someday, in the near future.

[@JakeStine]

Metaverse fad came and went so fast people already forgot it happened.

[@sireveman]

@@sebbbi2 VR headset sales are literally on an exponential curve. Quest 3 is and PS VR are insane this year.

[@danielwalley6554]

Thank God for dudes like Linus, who just come from a place of honest sense.

[@psiryan]

I was looking for midi files to listen to (this is what I used to listen to before MP3s) and was surprised to find some sites with web rings that still function. It brought me back and made me think about using the web then and using the web now. Exploring the net-scape using web rings and link boards is like taking a relaxing and refreshing walk near a gentle stream. Not long after my search for midi files, I had to return to googling for something more urgent, and I realized that modern web usage is not exploratory at all; it's more like hailing a cab in the very busy downtown of a large super urban city. Definitely we need the urgent method sometimes, but I wonder if we do not overrely on it.

[@agush22]

it was surfing the web, nowadays you browse

[@A4Blue-m7n]

Of course Linus has the most logical outlook on AI. I swear all these fear mongering “tech” people are just trying to pump Nvidia stock.

[@Batwam0]

I wanted to see a meme of Linus waving at the camera and saying “Nvidia, thank you!” 😂

[@RedHair651]

Haha me too

[@Basil-the-Frog]

At 8:10, "beautiful science" in results in "beautiful science" out. That's B.S. and it's what you get when you use poor data to train a machine learning model. However, we are getting to a point where we can "generate" data to do training. However, the hard things for people to do are not always easy to train a model to do. This is what Linus was talking about when he said he wanted AI to help him. But that is problematic. A little earlier (at about 7:10) Linus Torvalds mentions the very complex tools used to modify kernel source code. Being able to do this with "AI" would be very useful he says, but I'm guessing it is one of the further-down-the-line places we are going to get. When refactoring/modifying code to do complex things like change the size of a data structure that is accessed directly in assembly language, or low-level C, fixing the offsets and finding all the references to the code can be tricky. Writing algorithmic code (as opposed to a data-driven machine learning model) is exceedingly complex, so I'm guessing it will happen much later. One of the things we ARE seeing is machine learning models that learn by trying. Google has some robots that play soccer and someone else has a robot that sutures wounds. These can learn by trying. Learning how to fix problems like Linus is talking about is very hard to set up. Perhaps this is a good research project for someone.

[@JakeStine]

To be fair, the way LLMs work, you can feed entirely coherent and precise data in, and still get complete BS as output. LLMs are not cognition or reasoning machines. They are models that produce statistically likely sequences of glyphs and tokens (letters and words). The "self-learning" models are just refining that statistics-based response usually at the cost of further reduction in variation of responses. The more statistically precise a response system is, the less likely it deviates significantly from the ideal statistical response. The "iterative models" (which are different, these are the ones that keep trying until something is finally correct) are limited by the quality of tests and criteria that define correctness and also by the # of iterations required to get it right (ballooning costs).

[@thebutler2525]

He's right.. The best play especially to programmers is to only use the AI for specific syntax searches and NOT to fill out or have it write lines of code for you. You won't be a great programmer by doing that sort of thing. I'm reminded of auto-correct... think of how many words you cannot spell because of it.

[@G-3-A-R-Z]

"Beautiful Science is what he meant." 🤣

[@agnescroteau8960]

7:38 😭😭😭😭😭 guys please be competent. I’m supposed to be a simple child bearer!!!!!

[@nazzzz89]

Louis CK has improved his tech talk skills big time.

[@KhopdiTodSaleKa]

lol

[@SealedDoorNu]

It should b assist u not replace.

[@anonymous..-]

The free version Adobe Acrobat PDF viewer is currently trying to convince me to use its AI. It’s Absurd.

[@jojojojo4332]

Is that before or after it’s koning about getting more money

[@Terra101]

Louis CK is a good inverviewer.

[@johnmckown1267]

AI will never replace me. I wouldn't put up with management. They are also AI. Artificial Incompentence. [Supposed to be humourous. I know -- don't quit my day job.]

[@ntal5859]

When he was told could not say "BS" Torvalds should have got up and ended the interview.

[@FrankHarwald]

Why shouldn't people be allowed to say Bachelor of Science?

[@friedpizza262]

yeah the video is a deepfake

[@xpload32004]

Funny this is the guy who told him he couldn't say it, said it himself. What a child

[@ControlProblem]

It was a joke. Linus reacted appropriately.

[@NextLevelCode]

Anyone know the link to the full video or the name of this talk?

[@jcubic]

Where is the original?

[@moe_dk]

Haven't watched in full, but the background fits, and it's from a couple of month ago. The Linux Foundation: Keynote: Linus Torvalds, Creator of Linux & Git, in Conversation with Dirk Hohndel kzbin.info/www/bejne/mYHZg3yNjbOal68

[@BrianMosleyUK]

We're all Gen AI now. 👶

[@philippeannet]

One of a wise man, definitely !! And in reverse engineering, we’re also praying to get some assistance, lots of promises, and still not much to see in the horizon… and personally, the more I dig into the mechanics of GenAI, the more I get convinced that it’s indeed a hype 🤔

[@galaxygur]

What's "cloud native" ??

[@clementtw]

the “beautiful science” was what he meant… lol

[@gmxmatei]

A question: 100 programming languages? Why not only one?... Subject Oriented Programming!! I built it!!!

[@Ferdii256]

Can somebody share the origin talk, please?

[@zoyaa9759]

Really, we need split data and information. The IT actually is DT, there is nobody to inform.

[@justwanderin847]

I would like to see the programs and algorithms as open source. I would like to see the algorithm that they use to censor out "hate", "false truths", and "lies". So will they make public the algorithm to determine what is hate speech?

[@zerotwo7319]

It is a Hammer to make it say the right thing.

[@JayDee-b5u]

There is no algorithm for that. They just train it subjectively. There is no algorithm for 'hate' nor 'love'.

[@justwanderin847]

@@JayDee-b5u If text contain Conservative or Constitution or No Climate Change by Man, Then mark as Hate.

[@JayDee-b5u]

@@justwanderin847 Exactly. Subjective.

[@justwanderin847]

@@JayDee-b5u and that is exactly what they do.

[@acherongoon]

Personally I use AI to assist, and want as described a 4-eyes/paiered coding, otherwise build standardised modules and snippets. I tried and game design, pong, and waa given a solid basic design (both Claude and ChatGPT); simple changes were easy, but (as per uaee feature creep) the program devolved, and as modifications were more targetted to interface design errors and code style breaks increased, ie integration to existing world "High Score" table. But patching to directly interface to a poorly defined tools and webservices it managed baþer rhan I! A MASSIVE TOOL, but OS code/Service layer code needs strong review to ensure code is safe.

[@JoshtheFifith]

Ai would not be able to create git you need linus torvald for this it takes higher level thinking

[@florence-himeap]

"FINALLY"

[@mhchitsaz]

seems like he is still writing in assembly - I use to think he is a visionary

[@paulyz8534]

Linus is God

[@Trev0r98]

what's "calgnated"?

[@Wayfarer941]

Cloud Native

[@friendly.icicle]

Why is this interviewer correcting Torvalds and telling him he can't say BS? Wtf, he's not talking to toddlers, why this censorship?

[@DeepanshuLulla1891]

He is obviously saying it facetiously

[@Sammi84]

The interviewer is a long time colleague of Linus. They have in jokes between them. This is probably one.

[@Basil-the-Frog]

If they have worked together, that explains it. Plus, that way we know a lot of the conversation was planned carefully. I'd expect Linus to want something like that.

[@joshallen128]

​@@Sammi84dirk hondel he's the maintainer of subsurface

[@coach.muhsin]

If the Artificial intelligent is not intelligent. what is that?

[@Lemonsieur-m4m]

Hey Hi instead. Well this is THE purpose of IT. To make things irreversible. Isn't it ? The interviewer is horrendous.

[@nothingmuch67]

7:30 That was sooo cringe 😂😂😂😂

[@Storm_.]

Not a fan of when interviewers pose their questions in such a descriptive way that the interviewee has no choice but to answer within the very specific context provided. Then he keeps interrupting Linus and making cringey remarks.

[@86Corvus]

AI is the NFT of 2024... Exit while you can...

[@ControlProblem]

@86Covus: "AI is NFT." Linus Torvalds: "I think AI is very interesting." "I'm one of those people is very optimistic about AI" Interviewer: "I see a huge opportunity for AI as code completion. Linus: "I am looking forward to using AI to find bugs. Making the tools smarter is not a bad thing."

[@lolopinpin4950]

Have they even used Claude or openai latest products?

[@QwertyNPC]

Oh yeah, Torvalds is famous for speaking about things he didn't try. There's a video of him ranting about vegetable farming and I was like dude, have you ever tried it ?

[@maximenadeau9453]

Asked claude how many A are in strawberry he said 3.

[@bola986]

They think that downplaying AI, will make it to go away. Let's make old computers great again! (I use Linux and Git everyday, but nowadays also OpenAI and Claude)

[@lolopinpin4950]

@@maximenadeau9453 I asked Claude and he said 1. I tried with plenty other words and only got correct answers, fyi

[@raul36]

Y'all are like stochastic parrots, talking about the same thing all the time all damn day, man.

[@BobrLovr]

Maybe stick to talking about OS development 😂

[@Nnonymus]

Well llms could become the os of the future

[@lucasjames8281]

@@Nnonymusimbecile

[@turolretar]

@@Nnonymus I can imagine the horror that would be

[@mecanuktutorials6476]

Copilot+PC is already starting this adoption