I think you're right. It's pretty much a scam based mainly off of hype and marketing. Actual use-cases are very few and far between. And where's the money? Where's the profit?
all life would be better off without them. I feel the same way about industrialization that is meant to produce surplus at the expense of human life and the planet's resources and eco systems.
I feel the same way about cars. Cost is about 20k or more a year to run them. How the hell can we afford that? Even in the West that is a large part of the GDP per person.
You don't need to bother with elaborate programming schemes. It will be damaging enough to the LLMs as scrapers consume humans' increasingly bad writing. It's halfway to banal gibberish already.
You are hitting on something real! A 2023 paper titled "The Curse of Recursion: Training on Generated Data Makes Models Forget" shows that repeated training on synthetic data (meaning AI learning from itself) leads to increasingly degraded performance, not just stylistically, but in reasoning and factual accuracy. So yes, future LLMs may start producing more generic, less insightful and even nonsensical outputs or a kind of auto-cannibalism of language. "Some groups are already racing to create 'pure human' datasets for future training because they know what's coming," says the LLM I queried about it.
Let's not forget that the money these gouls have ploughed into AI already comes from some of the richest and least moral people on the planet, so when they come crying, telling us we have to accept their dark future vision and they have to make back their Investment, no one forced them to make the bet and unlike the people around the world they are trying to replace, these titans of capital adventure will not go hungry from losing a few quid pro quo.
Let’s not kid ourselves...the billions being funneled into AI aren’t a gamble, they’re a blueprint. These aren’t desperate investors hoping for a return, they’re the architects of a future where profit comes from replacing people, extracting data and controlling every single layer of life. They won’t come begging us to accept their vision, no, they’ll weave it into policy, sell it to us as convenience and absorb and bury dissent under layers of inevitability. No one forced them to invest, but through regulatory capture, media control, and infrastructure they monopolize, they’ve made sure the rest of us will pay the price, not them. They won’t go hungry if their bet fails. But if we don’t resist, we just might. I think the only way is to take it down.
Deception is at the heart of their game. Wanting people to believe the world is behind a screen. Flat, dark, controllable, a miniscule fraction of the real world.
We must look up now and again and realise the truth.
When hearing about teenagers getting relationship advice from the (G) LLMs an image of a lone budgerigar in a cage chirping at a mirror manifests insecure feelings for marketing from the minds of Theil and co.
as an artist, i'm fascinated by all that you're talking about, to the point that as i prepare a print book, instead of setting the type in adobe indesign, i'm rasterizing it in photoshop and placing text as an image. not that it will do much. but in doing print, and using handwriting, hand lettering, and pen and ink drawings, something just says do this. i also think it's harder to do search/find in my text, as i remember it used to be hard to find a printer for my work when i was at simon & schuster because some of the bible belt printers. this was late '90s early 2000s. then everything went to china and now i'll be printing in pennsauken, nj. it's corporate but there's been a lot of censorship in the states as we all know.
I wanted to post some of my writings about alternative work search tactics but don't feel like feeding the scrape machine. Would convert to image if I start posting. Maybe pollute the image somehow also.
This is brilliant. Exactly what this generation needs. And yes, unless the digital material is uploaded somewhere, it's unlikely that your book will get turned into fodder for this machine.
What's the theme of the book? I'm trying to finish an introductory (yet experimental) work in classical philosophy, but I've stalled just before the finish line.
i'm keeping the digital material away from online, and someone would have to transcribe it. to be honest i struggle because paper writing is a whole different form of seduction and attention. so i'm designing it so you have to turn it like dirty magazines of yore, and i'm doing so much detail you get lost in it.
i'd considered doing away with page numbers to facilitate getting lost in pages as you flip through and getting distracted by another thought-- not in an online scattered way, but hopefully in a...serendipitous way.
i've stalled, too. i was going gang busters but i'm back to fighting my landlord and neighbors harassing me BUT it's related to this book.
the book is really a perfect-bound zine in that it's more free form like a zine but it will be a regular thing because there is no ending, there can't be, until i die.
things are changing in the world way too fast and i've no pat answer any more, it's going to be an ongoing maybe 2X a year thing because it'll be discussing what i cannot do on substack nor what i'd want to discuss. it's more HUMAN, with sex, ugliness, horror, and trying to find the angelic again, or God.
so there can be no pat finished book.
when Trump blew it i felt this burst of resignation that suddenly actually became "fxck it" ENERGY. a lot of people gave up the idol and i feel an exciting under current.
Jeff Childers will think it's 78D chess til the end but some of us are done. when RFK got that bunny eyed look i knew it was over.
and maybe also a regular thing (this book i'm calling a MegaZine also as a homage to paper magazines i used to love and look forward to), because i also want to build some kind of real life art scene and want to try and see who gets it, who finds it, and if they be bop riff off it.
so maybe that's why YOU'VE stalled: where to GO???
something's changed and our endings are emorphous now.
another thought: if you're trying to finish an introductor (YET EXPERIMENTAL) work in classical philosophy, maybe you're also called to ...find a different ending and that's why you've stalled when so close?
i'm just at the beginning. i have the art and text but the EDITING... where to start and stop? editing changes the story.
but my theory that i'm trying to awaken sleeping beautiful monsters with is that the abused and feral have the power to face terrors and new things the normals buckle immediately before. i think the ferals and weirdos have to step up and re-make culture to woo the few who need to be inspired or have a new vision.
our vision is insane. i'm obsesses with Devouring Mother and how to turn away. who turns away from Mother? no one can even if you think you can. nope. not even us girls.
so that's why it's an ongoing MegaZine. it's a long ongoing discussion that will evolve with others' input but mostly by my realizations that i test in real life. i'm never starving for story. i'm drowning in crazy stories: i'm living in a dying san francisco full of demons and cannibals.
it's ugly here. but i'm trapped for now so i have to try and make my own paradise in the middle of this hell where i'm being hunted in my own home.
i figure if i can write about this it also becomes a playbook. but i have to test things out. nothing can be theoretical anymore.
there. long answer. the cops just left and i'm stressed and tired. too them 12 hours to come out.
prices are cheaper than me xeroxing, which was what i was preparing to do, when my art director collaborator told me about them and he loves the hard cover full color book they just did for his classic VW project.
I like it. Challenging AI is interesting. It gives you an idea of the algorithm, if you can call it that. It agrees that it's only as smart as the data it's gets and much of the data is compromised. Also, it thinks newer data is more accurate. Where could that go wrong?
I asked it if men could have babies, it said if they had a uterus, they could.
These types of social blocking techniques have been used since as far back as the 80’s to theoretically combat automated text scanning and classification and generally been useless in the long run.
A classic was adding “I’m working on a <weapon>” type message in a signature line, to try to maximize flagging as someone who needs to be monitored, to overwhelm teams who search mails for terroristic activity.
All of these methods have long been understood by IT security people to manage worms, in general - worms are much nastier.
Most tarpits, stochastic token generators, nepenthes traps are well-understood and managed by input scrubbing, and page address management, it’s not hard. If someone doesn’t want anything on their site to be unreadable they only need to stop using a network address other than a number to ensure they are never indexed or reachable.
It’s usually called the darkweb.
It sort of eliminates the utility of indexing altogether, doesn’t it. If fact you can simply use air-gap protection of text by keeping it an un-networked device.
Input scrubbing isn't really an option if you have massively expensive webcrawlers working furiously towards sucking up the last pockets of pristine untapped data, and if the garbage material is only distinguishable through qualitative analysis, well, then you're fucked.
And to be sure, tarpits are old school, but this approach at least seems moderately effective. The coder behind Nepenthes claims that Google's crawlers show up in the access logs millions of times, so unless he's just selling t-shirts, it seems to do some damage at the very least.
Password - lock. If access is wanted grant by request with proof of human. But why “poison” LLM texts? You can make it temporarily more expensive to gather publicly available data but statistical disruption of text is not really that useful.
Why would you use publish text in public without the assumption that derivative works would be made?
Nobody is scanning books in my considerable library. Vast quantities of published material never make it to digital form.
As someone that works in prepress that deals with Adobe day in and day out preparing files for print, I can tell you that while noble, this is assuredly a waste of time. Adobe Acrobat can pretty much instantaneously, and with remarkably accurate results, ocr text in myriad ways. For example, if you took a snapshot of text, or scanned in text, it can make it directly selectable and editable. This can even be done on my aging cellphone without adobe. So, it is not a stretch to assume that these hyperscale data centers and gpu-heavy AI culprits can easily syphon up anything rasterized within its reach and glom it onto its overweight machine mind.
way beyond my tech skills, but knowing people are doing this stuff gives me some hope. AI is a scam, albeit a dangerous one.
I think you're right. It's pretty much a scam based mainly off of hype and marketing. Actual use-cases are very few and far between. And where's the money? Where's the profit?
It's a giant waste of the West's capital. And possibly China's as well. Maybe we're better off if both systems collapse.
all life would be better off without them. I feel the same way about industrialization that is meant to produce surplus at the expense of human life and the planet's resources and eco systems.
I feel the same way about cars. Cost is about 20k or more a year to run them. How the hell can we afford that? Even in the West that is a large part of the GDP per person.
I wouldn't cry if we went back to bicycles and the horse and mule in rural areas.
You don't need to bother with elaborate programming schemes. It will be damaging enough to the LLMs as scrapers consume humans' increasingly bad writing. It's halfway to banal gibberish already.
You are hitting on something real! A 2023 paper titled "The Curse of Recursion: Training on Generated Data Makes Models Forget" shows that repeated training on synthetic data (meaning AI learning from itself) leads to increasingly degraded performance, not just stylistically, but in reasoning and factual accuracy. So yes, future LLMs may start producing more generic, less insightful and even nonsensical outputs or a kind of auto-cannibalism of language. "Some groups are already racing to create 'pure human' datasets for future training because they know what's coming," says the LLM I queried about it.
thanks for the ideas
i was won$der ing if a-not her wa:y to be spi (ky mig ght be to us/e punc—tuat ,ion in a str:an ge wa&y
Let's not forget that the money these gouls have ploughed into AI already comes from some of the richest and least moral people on the planet, so when they come crying, telling us we have to accept their dark future vision and they have to make back their Investment, no one forced them to make the bet and unlike the people around the world they are trying to replace, these titans of capital adventure will not go hungry from losing a few quid pro quo.
Let’s not kid ourselves...the billions being funneled into AI aren’t a gamble, they’re a blueprint. These aren’t desperate investors hoping for a return, they’re the architects of a future where profit comes from replacing people, extracting data and controlling every single layer of life. They won’t come begging us to accept their vision, no, they’ll weave it into policy, sell it to us as convenience and absorb and bury dissent under layers of inevitability. No one forced them to invest, but through regulatory capture, media control, and infrastructure they monopolize, they’ve made sure the rest of us will pay the price, not them. They won’t go hungry if their bet fails. But if we don’t resist, we just might. I think the only way is to take it down.
Deception is at the heart of their game. Wanting people to believe the world is behind a screen. Flat, dark, controllable, a miniscule fraction of the real world.
We must look up now and again and realise the truth.
that'sit, a system on feedback loop that wants us to believe the simulation is more real than the real...
When hearing about teenagers getting relationship advice from the (G) LLMs an image of a lone budgerigar in a cage chirping at a mirror manifests insecure feelings for marketing from the minds of Theil and co.
as an artist, i'm fascinated by all that you're talking about, to the point that as i prepare a print book, instead of setting the type in adobe indesign, i'm rasterizing it in photoshop and placing text as an image. not that it will do much. but in doing print, and using handwriting, hand lettering, and pen and ink drawings, something just says do this. i also think it's harder to do search/find in my text, as i remember it used to be hard to find a printer for my work when i was at simon & schuster because some of the bible belt printers. this was late '90s early 2000s. then everything went to china and now i'll be printing in pennsauken, nj. it's corporate but there's been a lot of censorship in the states as we all know.
I wanted to post some of my writings about alternative work search tactics but don't feel like feeding the scrape machine. Would convert to image if I start posting. Maybe pollute the image somehow also.
Yes, I too was wondering if pages of text copied as images would also jack up the LLM's images!
apparently screenshots of AI-generated nonsense compounds the problem
So now I'm wondering:
What would I title my image files to mess with LLM?
"Horse" or "sky" for a page of text or image of an afro?
This is brilliant. Exactly what this generation needs. And yes, unless the digital material is uploaded somewhere, it's unlikely that your book will get turned into fodder for this machine.
What's the theme of the book? I'm trying to finish an introductory (yet experimental) work in classical philosophy, but I've stalled just before the finish line.
oh you're so sweet. i'm stumbling blind here. this is the link where i've put up pages as examples: https://recordscratchradio.substack.com/
i'm keeping the digital material away from online, and someone would have to transcribe it. to be honest i struggle because paper writing is a whole different form of seduction and attention. so i'm designing it so you have to turn it like dirty magazines of yore, and i'm doing so much detail you get lost in it.
i'd considered doing away with page numbers to facilitate getting lost in pages as you flip through and getting distracted by another thought-- not in an online scattered way, but hopefully in a...serendipitous way.
i've stalled, too. i was going gang busters but i'm back to fighting my landlord and neighbors harassing me BUT it's related to this book.
the book is really a perfect-bound zine in that it's more free form like a zine but it will be a regular thing because there is no ending, there can't be, until i die.
things are changing in the world way too fast and i've no pat answer any more, it's going to be an ongoing maybe 2X a year thing because it'll be discussing what i cannot do on substack nor what i'd want to discuss. it's more HUMAN, with sex, ugliness, horror, and trying to find the angelic again, or God.
so there can be no pat finished book.
when Trump blew it i felt this burst of resignation that suddenly actually became "fxck it" ENERGY. a lot of people gave up the idol and i feel an exciting under current.
Jeff Childers will think it's 78D chess til the end but some of us are done. when RFK got that bunny eyed look i knew it was over.
and maybe also a regular thing (this book i'm calling a MegaZine also as a homage to paper magazines i used to love and look forward to), because i also want to build some kind of real life art scene and want to try and see who gets it, who finds it, and if they be bop riff off it.
so maybe that's why YOU'VE stalled: where to GO???
something's changed and our endings are emorphous now.
another thought: if you're trying to finish an introductor (YET EXPERIMENTAL) work in classical philosophy, maybe you're also called to ...find a different ending and that's why you've stalled when so close?
i'm just at the beginning. i have the art and text but the EDITING... where to start and stop? editing changes the story.
but my theory that i'm trying to awaken sleeping beautiful monsters with is that the abused and feral have the power to face terrors and new things the normals buckle immediately before. i think the ferals and weirdos have to step up and re-make culture to woo the few who need to be inspired or have a new vision.
our vision is insane. i'm obsesses with Devouring Mother and how to turn away. who turns away from Mother? no one can even if you think you can. nope. not even us girls.
so that's why it's an ongoing MegaZine. it's a long ongoing discussion that will evolve with others' input but mostly by my realizations that i test in real life. i'm never starving for story. i'm drowning in crazy stories: i'm living in a dying san francisco full of demons and cannibals.
it's ugly here. but i'm trapped for now so i have to try and make my own paradise in the middle of this hell where i'm being hunted in my own home.
i figure if i can write about this it also becomes a playbook. but i have to test things out. nothing can be theoretical anymore.
there. long answer. the cops just left and i'm stressed and tired. too them 12 hours to come out.
la la la...
(smile)
erika
i'm printing here: https://www.onpressbookprinting.com/
prices are cheaper than me xeroxing, which was what i was preparing to do, when my art director collaborator told me about them and he loves the hard cover full color book they just did for his classic VW project.
Thanks, this is worthy of study. I want to cause AI to feed on itself until it gets mad cow disease. AI is the feral hog of the internet.
I like it. Challenging AI is interesting. It gives you an idea of the algorithm, if you can call it that. It agrees that it's only as smart as the data it's gets and much of the data is compromised. Also, it thinks newer data is more accurate. Where could that go wrong?
I asked it if men could have babies, it said if they had a uterus, they could.
I know men who have a uterus. Gender and sex are not the same thing.
since when?
AI is to the internet what the Black Death was to Europe. A plague.
"Chill boomer.... society will move forward"
Let me introduce you to the cyclical theory of history......
"accelerating model collapse" - that's what I want.
A bit of newfangled machine breaking.
"toxic anti-AI assholes"
Where do I sign up?
“Most people are aware of the Nightshade project for “poisoning the well” of AI training data.”
If by “most” you mean “almost none” then yes.
These types of social blocking techniques have been used since as far back as the 80’s to theoretically combat automated text scanning and classification and generally been useless in the long run.
A classic was adding “I’m working on a <weapon>” type message in a signature line, to try to maximize flagging as someone who needs to be monitored, to overwhelm teams who search mails for terroristic activity.
All of these methods have long been understood by IT security people to manage worms, in general - worms are much nastier.
Most tarpits, stochastic token generators, nepenthes traps are well-understood and managed by input scrubbing, and page address management, it’s not hard. If someone doesn’t want anything on their site to be unreadable they only need to stop using a network address other than a number to ensure they are never indexed or reachable.
It’s usually called the darkweb.
It sort of eliminates the utility of indexing altogether, doesn’t it. If fact you can simply use air-gap protection of text by keeping it an un-networked device.
Input scrubbing isn't really an option if you have massively expensive webcrawlers working furiously towards sucking up the last pockets of pristine untapped data, and if the garbage material is only distinguishable through qualitative analysis, well, then you're fucked.
And to be sure, tarpits are old school, but this approach at least seems moderately effective. The coder behind Nepenthes claims that Google's crawlers show up in the access logs millions of times, so unless he's just selling t-shirts, it seems to do some damage at the very least.
I figure that destruction of the AI jurisdiction is the aim rather than just hiding from it. Any suggestions for how to do that?
Password - lock. If access is wanted grant by request with proof of human. But why “poison” LLM texts? You can make it temporarily more expensive to gather publicly available data but statistical disruption of text is not really that useful.
Why would you use publish text in public without the assumption that derivative works would be made?
Nobody is scanning books in my considerable library. Vast quantities of published material never make it to digital form.
Keep what you publish on paper.
Could an LLM be used to generate the bullshit data?
Yeah, but I guess it would be easier to filter then, using the same approach as these AI detectors (like originality.ai)
I thought that's all they did?
As someone that works in prepress that deals with Adobe day in and day out preparing files for print, I can tell you that while noble, this is assuredly a waste of time. Adobe Acrobat can pretty much instantaneously, and with remarkably accurate results, ocr text in myriad ways. For example, if you took a snapshot of text, or scanned in text, it can make it directly selectable and editable. This can even be done on my aging cellphone without adobe. So, it is not a stretch to assume that these hyperscale data centers and gpu-heavy AI culprits can easily syphon up anything rasterized within its reach and glom it onto its overweight machine mind.