Torvalds Speaks: Impact of Artificial Intelligence on Programming

today as Jim proved to us without talking about artificial intelligence and large language models um I I typically say artificial intelligence is autocorrect on steroids because all a large language model does is it predicts what's the most likely next word that you're going to use and then it extrapolates from there so not really very intelligent but obviously the impact that it has on on on our lives on on the reality we live in is is significant do you think we will see llm written code that is submitted to you as a request I'm convinced it's

going to happen yes I mean and I'm it may well be happening already maybe on a smaller scale where people really use it more as a as a help in writing code but um it's it's clearly something where automation has always helped people write code I mean this is not anything new new at all uh we don't write machine code anymore we don't even write assembler and now we're moving on from C to rust so I don't see this as something as revolutionary as all the news talk is every day about AI it's not an

area I obviously work with I I'm still very lowlevel I I got into kernels because I love the lowlevel hardware details and that's why I'm still there but but so you you say you expect this can help people write code this can help people get started but then if we look at the previous conversation and the the challenges around code reviews and maintaining so do you think that larger language models will get to the point that they can help us review code that they can help maintain sub I hope I hope because that's certainly one

of the areas where which I see them really being able to shine to find the obvious stupid bugs because I I mean how many of people here are actually programmers in this room a lot so a fair number a lot of the bugs Ivory right a lot of the bugs I see other people write they're not like subtle bugs a lot of them are just the stupid bugs that you did not think about and you don't need any kind of higher intelligence to find them but having having tools that warn I mean we have compilers

that warn about the obvious really obvious ones but having llms that warn about slightly more subtle cases where where it may just say this pattern does not look like the regular pattern are you sure this is what you mean and and the answer may be no that was not at all what I meant that you found an obvious bug thank you very much so I do think that Ms are going to be a big you call them disparaging generally uh like autocorrects on steroids and I actually think that they're way more than that and how

most people work is we all are autocorrects on steroids to some degree and I I see that as a tool that can help us be better at what we do but I've always been optimistic the whole uh hopeful hopeful was the word you helpful hopeful and humble hopeful and humble that's my middle name uh but but on the other hand I mean I have been so optimistic that 32 years ago I was stupid enough to think that you can write a better Colonel than anybody else so you have to kind of be a bit bit

too optimistic at times to to make a difference so my approach to LM really has being that hey this is wonderful this is going to I love seeing the optimism I don't necessar share it now a lot of people disagree with one of the things that I worry about in all this is we see the hallucinations we see and and that's a technical term for llms they do hallucinate and they do makeup stuff and so the more they are being put into the position where they will automatically do things without an actual human being there

to catch them the more this becomes scary not scary as in they will they will rule the world and not in the Sci-Fi sense but in the so many bugs that will happen and that will affect our lives or our code well I see the bugs that happen without them every day so that's that may be why I'm not so worried I I I think we're doing those just fine on our own and I I I don't think I can end the topic on a better uh uh highlight so let's let's go from from there

to to