Torvalds Speaks: Future of AI

Let's go for something that's fun and entirely non-controversial. So, let's talk a little bit about AI and large language models. And [laughter] I like you. You're as cynical as I am. That's That's a good sign. That's a good sign. Um, so obviously this is this is the current hype kit on the blog and and if you want to double your salary, add AI to your title. And it it's hilarious to watch the the self-description of of companies and every company on earth has an AI angle to their story as of this year. It is amazing. But what I find so interesting is this idea that Gen AI is going to be the end of you know insert whatever end of programmers the end of authors the end of movie creators the end of so many jobs so you are going to be replaced by an AI model finally [laughter] yeah no I I don't know I I hate the hype at the same time I think AI is very interesting I when I went to university, we were still talking about rulebased expert systems and basian logic and and all of this kind of thing and it it was called AI but it was not in any way intelligent. Um and and I do think that the last few years have been two days. Yeah. But uh I I don't want to be part of the hype and uh I will say that the AI revolution has had even on the the kernel level a few positives. For example, uh a company like Nvidia, who is not exactly famous for being great at at interacting with the kernel community, has actually been much more active and and being involved in the Linux memory management code because suddenly they start caring about Linux when when they are selling a lot of AI hardware. Mhm. I mean, it used to be crypto and and it's still obviously GPUs and and it's being used in in big servers and running Linux. So, it has actually had a positive impact. But my personal opinion is let's wait 10 years and see where it actually goes before we make all these crazy announcements of your job will be gone in 5 years. Well, I I think we already see lasting and and almost irreversible change happening happening through many of these AI tools. Not on the high Gen AI front, but just at the in the tools that make our lives better and easier. I I keep calling it autocorrect on steroids and people are mad at me, but the if if you use an an Gen AI and hopefully an SLM tool to help you with with basically code completion in your editor. Isn't that a a a huge opportunity? So, I'm I'm actually I mean I'm one of those people who are very optimistic about AI and I'm looking forward to the tools to actually find bugs. We have a lot of tools tooling around the kernel and around any software project obviously uh and uh we use them religiously but making the tools smarter is not a bad thing and I to some degree compare it to writing things in assembly which literally I started doing with the initial kernel was I think about 50% assembly language and using a compiler and and using smarter tools rules is just the next inevitable step. So, that's going to happen, but I don't think it's necessarily the gloom and doom that some people say it is, and I definitely don't think it's the promise world that the people who are having their hand out for cash say it is. So, I you need to be a bit cynical about this whole hype cycle in the tech industry. I hope you all realize that before AI, it was crypto. before crypto it was whatever it's it's a cloud native cloud native um so [laughter] there's a loud voice in the front that's not hype okay I mean the hype there's always like a grain of reality behind it but you need to be careful about all the BS around that grain you can't say BS um so so the beautiful science is what he meant yes [clears throat] um So the I look at these tools and I look at you said assembler and then we talk about compilers then we talk about things like sparse a compiler that tries to find certain kinds of bugs. We talk about Julia to do code refactoring. We have had for the last 30 plus years a sequence of tools that helped make development better and more robust. And in in that lineage of things, I I hope there is really cool stuff coming down. Yeah, I mean um we have tools that do kernel rewriting uh with very complicated scripts and pattern recognition and things like that. And and that is actually literally why I think um AI can be a huge help because some of these tools are very hard to use because you have to specify things at a low enough level that the natural reaction would be to hey can we make this a bit easier and automate more of it. Uh so so yes I one one of the interesting angles that this whole large language model and and the the training data brings up is the role that data plays in in our modern world where it we all talk about open source about the source code the algorithms being available but open data really is kind of that the almost the more interesting question no it's not well I mean [laughter] it's not to me and and Actually, I'd like to clarify that I mean the LF obviously has open data projects and to other people it's more interesting and that is I think the whole for me the point of open source is that different people are interested in different things and and uh I was always interested in the low-level nitty-gritty of how the CPU actually works which is why I'm working on the kernel still but uh but yes you're right that in many situations what is important is the is the data that you then use to generate pattern find patterns and generate new interesting information with but but to me that that's not what I tend to do and there is that saying beautiful science in beautiful science out please [clears throat] translate um so you you talk about the things that you love doing and I want to point out it's been more than a decade since you started a project and the world is kind of getting a little antsy. So where is the next Lenus Torvaltz project? Oh no. Um I hope it never happens. And I say that because every single project I've started has always started from me being frustrated with other people being incompetent. Uh so or money grubbing, right? So the reason I started doing Linux was that I couldn't afford the real thing, right? And and I said, "How hard can it be?" And it turns out it can be pretty hard because here I am 33 years later and still working on it. But uh I made the same mistake then it's 20 20 years ago when I said, "Hey, I really don't think source control management is very interesting." and all these people before me, they clearly gotten it completely wrong. So, so I need to do my own. How hard can it be? Uh, and I'm actually hoping to never be in that situation again. Uh, that that there will be somebody else who comes and solves my problems. And I have to say, um, I don't have any huge problems. uh Linux for me solved all the problems I had way back in 90 to maybe 93 uh and and it if it wasn't for the fact that others came around and said hey I need this I would not have continued so so while my products start with something that I need the things that actually keep them going in then is then the fact that hey this is actually useful to other people because if it's only something for me it's not really interesting in the long run. So unfortunately we're out of time. I had a lot more fun questions. I guess I will have to ask you those in Hong Kong. U but for here for Seattle. Thanks everyone. I hope this was fun and [applause]

🚀 Torvalds delves into the transformative influence of Artificial Intelligence on the world of coding. 🚀 Key Topics: * Will AI replace developers soon? * Enhancements in development workflows through machine learning. * Predictions for the future of software development with the integration of AI.