on ai
16:04 30 mar 2026
ok, so ai is everywhere now and it kinda sucks. i want to talk about a few things: for one, ai in general consumer experiences, ai in software, and broader consequences of this 'innovation'. when i talk about ai here, i'm reffering to large language models or image diffusion models, usually pinned under the banner of 'generative ai'
every company seems to be integrating some kind of ai thingy-majig into their product. i dont think anyone likes this at all. im sure these companies will soon realize that this is not a profitable or consumer-engaging use of ai, but just like the last tech-hype trend of crypto and NFT and all that jazz, we the consumers will have to deal with this bullshit for the forseeable future, and the bubble is bigger than crypto stuff, and it's gonna last longer too. oh well.
now, in software: i personally believe that people who are greatly exaggarating how powerful/useful/good these ai tools are for software development are either AI shills in some way affiliated with the model-provider companies, or the software they are working with is simple/useless, or they are utterly deluded. when i say they are shills affiliated with the companies, this not only applies to people saying how great it is, but also to the people saying how dangerous it is: by implying it's in some way dangerous, they are making it seem like it's 'powerful' which obviously makes people want to try it and so on. i believe that ai is not as good/advanced for software developement as people say not because i have some grudge against the companies behind it (although i do, we'll get to that later), or because i think it will remove software development jobs (it won't, not in the way people are constantly implying, and even if it does, i'd rather be a starving artist) but because i actually, properly tried it! not asking chatgpt to generate me a script, or some other simple thing, but actually creating a working project: about 6 months ago, my mum's work provided her with free claude, which came with claude code, she asked if i wanted the login, and curious, i said yes (don't knock it till you try it, right?) i went the full 9 miles and installed all the bullshit on my computer, and decided to see what it could do: i started working on a linux distribution! yes, the original scripts for my distro, derive, were 'vibe-coded' (i put that in heavy quotes, as most of them were altered significantly before i published them, for example the original init script it provided me couldnt create /dev with correct permissions) : the results were utter shit, about 15% of it was usuable, the rest i had to basically rewrite from scratch over the course of a few months because it was utterly unmaintainable (my output is not as fast as the ai, but i am a little more competent) so, my verdict; ai is not useful for creating anything mildly complicated that isn't some unholy webapp made in js and python, because that's what most of it's training data is (it's completely useless at C in my experience). anyway, i'm glad to have claude out my computer now. phew.
now, as of the last few weeks, it appears to be getting better, not as good as people claim it to be, but firefox and linux are beginning to do ai code review. if that works well and helps move stuff along quicker, in theory thats probably fine, even thogh the model providers are evil (see next paragaph), it seems to me like the pandora's box has been opened now and this stuff is gonna be here in some way, shape or form basically forever, so i'm not one of those people who wants to shut it all down, i don't think its possible. so if it is gonna be here, i guess code review isnt so bad? if it points out real issues (it is a pattern-recogising robot after all) and there is significant human oversight, i guess there's not much of (purely software-wise) an issue.i would'nt trust it to write any security critical code, or any code at all really, but i guess it's pretty good at tracing logic? when two major open-source projects i use every day start using ai, it leaves me not really knowing what to think...
now, on broader issues with ai: people commonsly talk about water usage, that's not much better than most social media and so on, the bigger problem is massive data centers next to peoples homes, or worse, the group of insane e/acc cultist idiots running the model providers, who got bullied in school and want to replace the entire human race with robots (no, this isn't a joke) : their vision for humanity issome technocapitalist hellscape where ai becomes so intelligent it can fix all the worlds problems (of course they are all also insane libertarians who think that the only way that can happen is if multiple companies compete at who can drain the world's resources fastest on a completely unregulated open market) : even if ai was good technology, these sure aren't the people i want at the helm of it all.
i guess im echoing what many others have said here, but i'm not really sure why anyone wants this, i mean i guess if pure unfiltered production of text or images is all you want it's great, but why are we attempting to automate away the fun stuff? generating art of some idea in your head isn't a new idea, we had pens, pencils, and paintbrushes! you don't need some image diffusion model to do noise reversals to generate some sonic fanart! i feel like companies are focusing their undoubtedly talented engineers on the wrong things. that's what i would say, of course they are focusing on where they percieve the money to be, why else would they do anything? surely we as a human race can think of some better groundbreaking technology than a matrix multiplication slot machine with a really annoying tone of voice? one can only hope.