@spid-the-spider replied to your post “if i see one more reblog of that post that...”:
Question: is your stance "generative ai use is fine" or "using the word "virginity" in this context is bad" because I don't understand what you mean. i'm just asking for clarification!
my response got real long, so i'm gonna respond here!
LLM == Large Language Model, ChatGPT, Grok, Copilot, etc are all LLMs
TTIM == Text To Image Model, the models that take in text, and output an image
To me, LLM/TTIM usage is not inherently a harmful thing. It is a nuanced thing. to be clear, i dislike the way LLMS have been integrated into nearly every interface and website. Thats annoying as hell. I wish that would stop.
but more specifically, the usage of virginity in that context is a fucking bonkers choice.
Primarily im deeply concerned with the growing trend I have been seeing of referring to people who use ai in any context as "less pure" or "dirty" -- its a tad sickening, to me. The usage of virginity in this context is particularly ludicrous, especially when the notes of that post are entirely people like, giving what read as church confessions about using ai
begging forgiveness like its church
Like, i see a lot of posts that are like "anyone who has ever used ai should kill themself" or "ai users should have their lives ruined" -- like i undertand it is sarcasm/hyperbole, but youre talking about my coworkers, man, just because they're a tad offline doesnt mean they should die. theyre just people.
ive had college assignments where ive spesifically been told to use AI to contrast it against other information sources to parse out usefulness.
But also--- adjacently, god. LLMs and TTIMS. sorry im about to go on a rant, here we go.
I could talk for a while abt how i think a lot of the way people discuss LLM usage on this website is a bit fearmongery and doesn't get to the point of ANY of the problems theyre worried about.
Copyright issues? the problem is systemic. i could write essays about my hatred of US copyright system, frankly, i have for class.
The US copyright has been broken since disney lengthened copyright terms so far. Copyright is a broken system that deprives writers of their characters and rights to their characters and work on a massive scale. Copyright used to be 25 years here. Then it was lengthened to 45. Then its, what, 20 years paused + 90 years from creators death. can we be so for real and be honest that that jump is insane
do you know how fucked this is. Microfilm doesnt last forever, and so many microfilm collections of newspapers post 1964 cannot be digitized and displayed online because of the copyright extension. ive been losing it. ive been LOSING IT.
and yes, yes, the unauthorized training of large datasets on artists work is fucked, but like--- why the fuck are we defending the copyright machine and cheering for DISNEY winning a copyright lawsuit? like holy FUCK. disney does NOT need more copyright rulings in their favor. its like everyone has lost sight of the greater issue of the copyright machine the moment TTIM training became a thing! like! what are we DOING.
its--- its a massive issue, unathorized training, but like, the copyright system has its own demons and disney winning is a lose for like, everyone else.
Water/power? we're getting something closer to a problem, but the fact of the matter is is that a single ai prompt uses the same amount of energy as a 6 google searches, (or well, with the more advanced LLM models theyre running these days its talking trees of prompts, so lets be generous and say 40 google searches) -- or, if i remember, about the same amount of energy as one lightbulb for 6 seconds, or one second of microwave.
The real water cost in training LLMS is in the crunching of the training data, which, i am going to be so real, NO ONE WHO IS USING AI ON THE USER END CAN CONTROL!!! The corporations creating these LLMS are in a tech bubble - which will burst, and will burn them, but the frontend users do not account for much of that water issue at all.
Of course, this scaled up is an issue, dozens of prompts do add up. but the fact is we live in a digitzed world. Energy usage in this way is going to keep climbing. Building datacenters is a different problem, that i do need to research more, frankly, but the fact is datacenters are being built all the damn time!! cloud storage is nearly standard these days as much as i hate it!! people were not angry about datacenters before LLMs, and its deeply annoying to see people SOLELY angry at the concept of them because of LLMS and TTIMS,
and--- and this is crucial, THE WATER USAGE IS IN THE POWER PLANTS. THE POWER PLANTS THAT WOULD BE RUNNING ANYWAY. and so much of that water usage isnt drinking water! isnt water that would ever go to consumers/wells.
john green has a pretty decent video about this and this article also breaks things down pretty well though i wish it would go more into the training aspect of things
but well, systemically, again, more than that, the US presidental administration is rescinded millions of funding to offshore wind farms back in august
the issue of water is an issue of clean power, and, AS PER USUAL, but everyone is mad about the wrong thing in the wrong way.
it drives me fucking bonkers!!!
its. LLMS can be useful. they can be useful or facinating or even artistic in extrememly small models. Thats what gets me. Because i was tooling around with them back in 2019. But the current end-user models available not only hardly do anything they promise, but theyre a waste of time to engage with! its like--- god. it kills me.
like, every issue people are upset at LLMs/TTIMS about is either--- wierdly puritan moralistic/reactionary, ie ai virginity, or like, wrong at the wrong point of the chain
with the EXCEPTION of training on unauthorized data. and even they. gray area. gray area. My understanding, at least, in the very beginning is like--- have you ever heard the statement "chatgpt was trained on the whole internet" -- thats not nessessarily true, and especially not anymore. Most "ai" (god i hate calling it ai. its not. its not ai. killing every corporate buzzword inventor) manufacturers are now buying spesific datasets from websites offering up training data.
But initally? that inital datasets?
the original training data was from the common crawl datasets.
the common crawl datasets?
yknow who also uses common crawl and is considered a net good, and is in fact the most useful website out there?
The Wayback Machine, and archive.org
If you say goodbye to webworms/common crawl data, you say goodbye to the wayback machines ability to collect website data.
Y'know what was instrumental in finding SO MUCH lost mechanisms content? the wayback machine
Webworms and web crawls have always been legally gray. But well! so is THE FACT I HATE MODERN COPYRIGHT IN GENERAL.
uh. anyway. this got a bit away from me. disliking LLMS/TTIMS is a nuanced thing for me. Theyre annoying in how they have entered every aspect of life but so much of the issues people have with them are symptoms of the wider system at large we live in, and so much of the moral nonsense against them is so close to like, purely reactionary non-logic based it kills me a bit.
dislike ai for actual reasons. not because you lost yoru goddamn virginity to it.
dislike it for the fact its unregulated and sending people into psychosis
dislike it for the fact the companies creating it are operating within a tech bubble
dislike it for how its pushing up RAM prices
dislike it for how its degrading the most easily accessible information for people searching
dont dislike it because of wierd moralistic reasons such as it makes you "impure" or something. anyway
reblogs off. this is for My Blog Only
