Register a SA Forums Account here!
JOINING THE SA FORUMS WILL REMOVE THIS BIG AD, THE ANNOYING UNDERLINED ADS, AND STUPID INTERSTITIAL ADS!!!

You can: log in, read the tech support FAQ, or request your lost password. This dumb message (and those ads) will appear on every screen until you register! Get rid of this crap by registering your own SA Forums Account and joining roughly 150,000 Goons, for the one-time price of $9.95! We charge money because it costs us money per month for bills, and since we don't believe in showing ads to our users, we try to make the money back through forum registrations.
(Thread IKs: skooma512)
 
  • Post
  • Reply
StratGoatCom
Aug 6, 2019

Our security is guaranteed by being able to melt the eyeballs of any other forum's denizens at 15 minutes notice


Man, the longer they keep kicking the can, the worse the 'finding out' gonna be.

Adbot
ADBOT LOVES YOU

StratGoatCom
Aug 6, 2019

Our security is guaranteed by being able to melt the eyeballs of any other forum's denizens at 15 minutes notice


SpaceDrake posted:

God I am just sat here laughing my rear end off

As so many have already pointed out, this is just going to incentivize bank runs because everyone's going to get bailed out and the folks who bail out early and sell stock early (like what happened when SVB got goosed a few years ago) are going to make utterly ruinous amounts of money. Every bank will be squeezed, therefore every bank is protected, therefore every bank is now a literal mint.

Good job to the Biden administration, I suppose, for choosing the path of total financial chaos and proving that the government could just end poverty at will.

:kheldragar:

StratGoatCom
Aug 6, 2019

Our security is guaranteed by being able to melt the eyeballs of any other forum's denizens at 15 minutes notice


New Thread Title:

[Doomsday Economics] Bank Marathon

StratGoatCom
Aug 6, 2019

Our security is guaranteed by being able to melt the eyeballs of any other forum's denizens at 15 minutes notice



Lmao if this gets them to finally crack down on AI like they should have long ago.

StratGoatCom
Aug 6, 2019

Our security is guaranteed by being able to melt the eyeballs of any other forum's denizens at 15 minutes notice


Frosted Flake posted:

It drives me absolutely apoplectic when I ask for it to depict any historical period I know a good deal about because so much is wrong in very simple ways that you could tell an artist about, or provide reference material to. It does give me a deeper appreciation for history illustrators like Angus McBride though.

The same is true if you know art history because (obviously) it doesn't understand artist or genre, it's just pastiche. Like ChatGPT is "literature" for people who don't like to read, AI creates "art" for people who don't like art.

:hmmyes:

StratGoatCom
Aug 6, 2019

Our security is guaranteed by being able to melt the eyeballs of any other forum's denizens at 15 minutes notice


https://twitter.com/TheCartelDel/status/1636262759947210753

Whelp, just before 2024 :smithicide:

StratGoatCom
Aug 6, 2019

Our security is guaranteed by being able to melt the eyeballs of any other forum's denizens at 15 minutes notice


Frosted Flake posted:

Just like the ChatGPT update, not really.

This poo poo is gonna get worse and worse at this rate.

StratGoatCom
Aug 6, 2019

Our security is guaranteed by being able to melt the eyeballs of any other forum's denizens at 15 minutes notice


Jon Irenicus posted:

remembering the time that cspam found the landlord thread in BFC and having a chuckle. that said I DO NOT endorse raiding the "Fundamentals Are Sound" thread

I don't, tell me more.

StratGoatCom
Aug 6, 2019

Our security is guaranteed by being able to melt the eyeballs of any other forum's denizens at 15 minutes notice


DancingShade posted:

Brics is going to build its own treehouse and the USA & vassals will be outside complaining they weren't invited even though they keep throwing eggs at everyone else's houses.

Then it's going to rain.

The rain being climate catastrophe?

StratGoatCom
Aug 6, 2019

Our security is guaranteed by being able to melt the eyeballs of any other forum's denizens at 15 minutes notice


Oglethorpe posted:

the only thing i can think of that could explode in a chocolate factory is like a boiler or something

Calories are energy, no shortage of flammable poo poo there.

StratGoatCom
Aug 6, 2019

Our security is guaranteed by being able to melt the eyeballs of any other forum's denizens at 15 minutes notice



TWT should be trained to FYAD.

StratGoatCom
Aug 6, 2019

Our security is guaranteed by being able to melt the eyeballs of any other forum's denizens at 15 minutes notice



Well, uh, that explains a lot about mastershakeman. :allbuttons:

StratGoatCom
Aug 6, 2019

Our security is guaranteed by being able to melt the eyeballs of any other forum's denizens at 15 minutes notice


https://twitter.com/stealcase/status/1642019617609506816

StratGoatCom
Aug 6, 2019

Our security is guaranteed by being able to melt the eyeballs of any other forum's denizens at 15 minutes notice


gently caress DOUBLE POSTED

StratGoatCom has issued a correction as of 01:22 on Apr 2, 2023

StratGoatCom
Aug 6, 2019

Our security is guaranteed by being able to melt the eyeballs of any other forum's denizens at 15 minutes notice


Frosted Flake posted:

I don’t want to get too philosophical, but:

“I'm also tired of people saying that ChatGPT4 is not as disruptive as everyone is saying, that the hype will fade, and that good writers will prevail. They ignore what's an obvious reality - if dozens of people report here that they lost almost all clients to Al, then not everything is
as rosy as they want to believe.”

“We live in a profit-driven world where quantity matters more than quality. That means that most companies will only care about publishing decent content every day, and that's where ChatGPT outperforms us. Good writers can deliver the most captivating copies and articles, but they're not free and can't deliver
their work in a matter of seconds.”

but so much copy already read like it was written by machines. This feels like something where capitalism had turned art into craft into commodity, and human writers were “programmed” by the market to write stuff that was all more or less the same anyway.

Obviously I understand the effects on human beings and that sucks, and I agree the writing is trash. The only people heralding this are beep boop tech guys and capitalists, people at the forefront of turning themselves into machines, though the process of capitalism.

This writer judging their work by “Good writers can deliver the most captivating copies and articles” makes my heart sink.

I suppose I keep thinking of whichever AI art model people say works best, in fact you practically have to, when you end a prompt with “ArtStation”. The art was already homogenized, commodified, dictated by capitalism. It already looked like it was mass produced by machines for mass market appeal.

Listacle writing, ad copy, online articles trying to capture google AdSense revenue, that’s sort of the same thing. It’s capitalism doing this, and can anyone say they’re satisfied with the result either way?

https://m.youtube.com/watch?v=XHGEwMgnOnk

Gonna be blunt. I don't trust a lot of these ai stories because the AI moguls conduct themselves like cryptolords (often because they are) and current copyright doctrine don't favor them. They're trying to spread fear and FOMO among marks and competition.

StratGoatCom
Aug 6, 2019

Our security is guaranteed by being able to melt the eyeballs of any other forum's denizens at 15 minutes notice


Paradoxish posted:

The flip side to this is that the people screaming the loudest about AI revolutionizing everything have a huge amount of overlap with the scammers who push poo poo like crypto at every possible opportunity. Browse through sites like Udemy and you'll find a ridiculous number of "prompt engineering" and related AI courses already, because a specific group has latched onto this as their next grift and that group tends to be extremely online and extremely good at creating hype.

It's this. Fundamentally, this poo poo will not work in a lot of cases where copyright (AI 'creative' output cannot be cast iron defendable without enough work that it seriously hurts the value prep at best) or accountability matter, and the law on either of those is not gonna shift because if it does, it breaks the economy that deal with those things nearly completely. These models are crimes.txt in more ways then I can count, and they're trying to offload this piece of computational corium before the law catches up and kicks their rear end. Don't try and cheat using this poo poo, as soon as detection catches up - and if the economy is to stay functional, it will have to, one way or another - you're gonna get birched.

Frankly, if we wanna trade acedotes, I trust a lifer art design lead saying some high power names are explicitly banning it in recent contract amendments, but being necessarily cagy due to NDAs a lot more then reddit acedotes, because it is *entirely* predictable from current copyright law.

Frankly the most Doomsday economics - and thus, most plausible - is that the real money is finding it and getting it off sites and asset chains of title.

StratGoatCom
Aug 6, 2019

Our security is guaranteed by being able to melt the eyeballs of any other forum's denizens at 15 minutes notice


Buffer posted:

Anything you use to detect can be used to tune. There's a feedback loop there. Also, like, what do you think is used in an automated detector? A model! And if the model artifact itself is infringing...

You're also going to have false positives and start flagging human work as ai-generated, which will I guess be used to absolutely crush people in court. The standard there not being accuracy but rather what you can get past a judge once. Like, look at forensics and despair.

But the main effect I think if the court cases go bad will be to kill the FOSS side of this, I don't think the closed side is killable.

The closed side runs into the same issue for copyright defense, but won't have the same problems for offical use becasue they'll put a watermark of some varity in to avoid liability and DMCA/lawsuit the hell out of anyone caught hacking it out or otherwise trying to defeat it, because it puts them at risk of getting the Mouse up in their business.

poo poo will end up being chased into the darkweb and prototyping.

quote:

this isn't a problem because our legal system already has a built-in solution to the copyright question. if the company using potentially derivative work is microsoft or google and has more money to spend on lawyers and bribing judges than the owner of the copyright, then they get to either get away with it or blackmail the copyright owner into selling the rights since their only other options would be to not enforce their ownership or enter into a court battle they can't afford. on the other hand, if the copyright owner is the one with more money and in-house counsel, then the copyright is enforced.

You misunderstand the problem. It's not the training assets, it's that nonhuman poo poo can't be copyrighted, and given that changing that risks the copyright system getting the Clarkesworld treatment... well, they'd be fools to change it.

StratGoatCom has issued a correction as of 16:44 on Apr 12, 2023

StratGoatCom
Aug 6, 2019

Our security is guaranteed by being able to melt the eyeballs of any other forum's denizens at 15 minutes notice


Old Story posted:

Right, but if a significant amount of people lose the ability to make a living producing things that are useless, it could have huge economic consequences

maybe we need to think about decoupling people's ability to survive from their ability to be economically productive or something

Precisely, plus the damage to net functionality from the crapfloods and the attempts to filter them out and ensuring that undefensable poo poo doesn't enter chains of title. It's a blasted mess that no one wins in, even if it's not the creative apocalypse that techbro dipshits try and market it as.

StratGoatCom
Aug 6, 2019

Our security is guaranteed by being able to melt the eyeballs of any other forum's denizens at 15 minutes notice


Buffer posted:


That just means stability can't claim copyright, and maybe you can't copyright direct output. I would just claim copyright when I use it as a tool.


No, that is not in fact how it works. You cannot copyright the product of wholly mechanical processes or instructions, either of which is the case. You can copyright stuff you do with that output, but ONLY the human amendments, which given what AI is free real estate as soon as you teach something to prise out the AI content. The best usecase is for references for concept art, because you have a human artifact to start a chain of title.

This poo poo is not ruddy new, and it's this ignorance that lets the techbros get away with poo poo.

StratGoatCom has issued a correction as of 16:59 on Apr 12, 2023

StratGoatCom
Aug 6, 2019

Our security is guaranteed by being able to melt the eyeballs of any other forum's denizens at 15 minutes notice


Buffer posted:

cool, so because samsung enhances the photos on my phone using a diffusion model I can't copyright them? Purely e2e mechanical and software process except me choosing to click button.

I use that example because that's precisely how a diffuser is used to enhance a sketch or photo. Another workflow is utilizing diffusion models in a way very akin to collage - mixing / welding together different models and weights and prompts across different areas of an image. Humans are *heavily involved* - it is not purely mechanical outside the waifu promptsmiths.

No, because a human lined up the image, and the bits of that genned image may be possible to dig out, as well as the nature of each bit. Its a useless title, not worth the case to defend when you get safer cost effect with existung technique.

StratGoatCom
Aug 6, 2019

Our security is guaranteed by being able to melt the eyeballs of any other forum's denizens at 15 minutes notice


strange feelings re Daisy posted:

AI is eating into concept art, promotional art, and commission work too.

Wonky eyes or 6 fingered hands can be masked out and fixed with a button press now. Power users can save specific characters to reuse and pose them exactly as needed like a 3D model so they don't look like randomized mutants on each generation. There's a huge variety of advanced tools beyond prompting being rapidly released.

Some companies refuse to use AI out of principal or legal fears, but tons are jumping in head first to massively reduce billable hours.

Ehhh, given the rumblings I've heard out of contract and even romance cover art, this poo poo may well be starting to slow because the current doctrine makes it a mess to approach, and the object lesson of Clarkesworld suggests good reason to be very cautious in any amendments to that law that make it easier to protect AI output. Not least because copyright is a really useful thing to have on any asset if money or reputation is involved.

Nothus posted:

AI isn't going to kick over the system and lead to glorious revolution. It's just going to immiserate another layer of the middle class and accelerate the upward accumulation of wealth.

That, and massively worsen the rate of epistemic toxification and the various obtrusive countermeasures needed to avoid making GBS threads things up.

StratGoatCom has issued a correction as of 18:14 on Apr 12, 2023

StratGoatCom
Aug 6, 2019

Our security is guaranteed by being able to melt the eyeballs of any other forum's denizens at 15 minutes notice


Paradoxish posted:

I don't think gpt will ever fully replace the content mills, but I hope I 'm wrong. The ability to poo poo out SEO content without the limiting factors of time or money will essentially destroy the entire internet advertising ecosystem and I'm here for that.

It will destroy the whole net if a lid can't be put on it.

It's also why I don't see a shift on AI output copyrightability in the cards, because it's too productive at making things that might be copyrightable artifacts if allowed in. It makes brute force attacks on the copyright system viable.

StratGoatCom
Aug 6, 2019

Our security is guaranteed by being able to melt the eyeballs of any other forum's denizens at 15 minutes notice


nexous posted:

the end result of ChatGPT will be anti AI people purposefully littering the internet with slightly incorrect data that will poison the well and make the whole thing useless.

an endless arms war between fake data and filtering tech will take place and the only two jobs available will be working for one side or the other

from an office, not remotely

I mean, sort of. Glaze is the first wave of this poo poo, and given that stability had their pet 'opt-out' wanker do a hitjob, I think they genuinely fear it. Even if it doesn't protect in itself, or is easily beaten, the ability TO beat it is a big no-no for the DMCA in the first place, which is something a commercial outfit will shy away from hard.

StratGoatCom
Aug 6, 2019

Our security is guaranteed by being able to melt the eyeballs of any other forum's denizens at 15 minutes notice


Pf. Hikikomoriarty posted:

can you clarify that last point about the dmca?

Circumventing DRM is breaking the law, outside of very specific circumstances, something which the commercial use of AI image generation does not fulfil, is the TL;DR.

StratGoatCom
Aug 6, 2019

Our security is guaranteed by being able to melt the eyeballs of any other forum's denizens at 15 minutes notice


Pf. Hikikomoriarty posted:

ok gotcha, thanks!

so it's like glaze might not work super well, it's like putting a lovely lock on your stuff

but if someone breaks that lock they have done crimes

Precisely, it's the legal analog of a tripwire force.

StratGoatCom
Aug 6, 2019

Our security is guaranteed by being able to melt the eyeballs of any other forum's denizens at 15 minutes notice


Nodelphi posted:

My job has me do a bullshit self-eval for work every year, it’s 10 pages long and wants you to write 7 essays on how you honor each of their corporate ideals.

I used to do one sentence responses and they grumble about it every year. This year I used chatgpt and they recommended me for a promotion.

That just means they're going to start putting out detectors and invigilating you as soon as they get wise, if it's for anything they give a poo poo about.

StratGoatCom has issued a correction as of 20:46 on Apr 12, 2023

StratGoatCom
Aug 6, 2019

Our security is guaranteed by being able to melt the eyeballs of any other forum's denizens at 15 minutes notice


I go to great lengths on how these AI shitters are crypto shitters, and they just up and make it obvious:

https://www.gamedeveloper.com/art/generative-ai-might-make-cross-game-assets-viable

https://www.youtube.com/watch?v=_g-NLYOSPjk

StratGoatCom has issued a correction as of 23:26 on Apr 12, 2023

StratGoatCom
Aug 6, 2019

Our security is guaranteed by being able to melt the eyeballs of any other forum's denizens at 15 minutes notice


Paradoxish posted:

There's an entire ecosystem are tightly connected internet grift and most of it links back to crypto at its core.

Been saying it; it all runs on basically for profit attacks on vital components of the current system that never work, but make for hella pumps and dumps, whether subverting finance law, or copyright law.

Frosted Flake, this is your cue to explain how the more functional socities of old would have curb-stomped this poo poo.

StratGoatCom
Aug 6, 2019

Our security is guaranteed by being able to melt the eyeballs of any other forum's denizens at 15 minutes notice



Well gently caress, he finally went and did it.

StratGoatCom
Aug 6, 2019

Our security is guaranteed by being able to melt the eyeballs of any other forum's denizens at 15 minutes notice


Relevant Tangent posted:

doom econ is perfect because doomsday implies a reckoning that will never come

The current situation is not indefinitely sustainable, sooner or later the current rot will knock something loose that brings the whole thing down.

StratGoatCom
Aug 6, 2019

Our security is guaranteed by being able to melt the eyeballs of any other forum's denizens at 15 minutes notice


https://twitter.com/mhodson520/status/1647300725578366980

StratGoatCom
Aug 6, 2019

Our security is guaranteed by being able to melt the eyeballs of any other forum's denizens at 15 minutes notice


https://twitter.com/BennettTomlin/status/1647290619407093760

StratGoatCom
Aug 6, 2019

Our security is guaranteed by being able to melt the eyeballs of any other forum's denizens at 15 minutes notice


Raskolnikov38 posted:

what the gently caress does this accomplish that a debit card doesn’t

Kill tether, as it requires it to have some rules.

StratGoatCom
Aug 6, 2019

Our security is guaranteed by being able to melt the eyeballs of any other forum's denizens at 15 minutes notice


Paradoxish posted:

They're all like this. I have a pile of AI Udemy courses because I'm evaluating GPT-4 for a client (it won't get even close to doing what they need but they'll probably use it anyway lmao) and they're all the same poo poo. Even the relatively well-meaning ones essentially boil down to instructions on cranking out an absurd amount of garbage-tier poo poo because the effort required to get good results isn't worth it.

I keep saying it, but the usual grifters bandwagoning this poo poo is the best indication that it's not really going anywhere.

Yup; this AI poo poo is crypto all over, but the aftertaste will be worse when the winter returns, because there will be a lot of destructive tools in the hands of net randos with last gen flagship gpus.

https://twitter.com/alexjc/status/1646627483507212290

A relevant thread from someone in the 'industry pushing for some lawfulness so he doesn't get screwed. Take it with those caveats, but worth a read.

StratGoatCom
Aug 6, 2019

Our security is guaranteed by being able to melt the eyeballs of any other forum's denizens at 15 minutes notice


And AI seems to be on a path to winter faster on account of no one sane wanting to make the changes to copyright doctrine to make it viable, and the current security scares.

StratGoatCom
Aug 6, 2019

Our security is guaranteed by being able to melt the eyeballs of any other forum's denizens at 15 minutes notice


mazzi Chart Czar posted:

Jeez.
But,how badly will AI screw up the work flow for your client if they use it all?

Depends on the flow. Models for art and music poo poo out directly strikable content in 1% of jobs. This, along with the previously mentioned protection issues render it nonviable for anything but the start of concepting.

https://twitter.com/ccanonne_/status/1646814299158097921

Buffer posted:

sending people to prison for crawling and archiving websites

common crawl respected robots.txt, lmao

Folks have found :aatrek: and medical info in those models, aside from the GDPR stuff. Plus, the whole 'unauthorized derivative product for profit directly competing with the original maker' is loving nasty, and apparently commercial use for poo poo like LAION is not in compliance with german law.

StratGoatCom has issued a correction as of 23:20 on Apr 15, 2023

StratGoatCom
Aug 6, 2019

Our security is guaranteed by being able to melt the eyeballs of any other forum's denizens at 15 minutes notice


Quinch posted:

surely whatever method you are using to identify copyrightable material could be built into the ai either directly or afterwards so it doesn’t produce stuff that gets you sued? seems like a trivial issue for computers to solve to just have what’s it’s produced ran against the training data again and disregard anything that matches existing work.

That's not nearly so easy as it seems and it runs into the output copyright issues I mentioned before.

Buffer posted:

yes because it uses common crawl, which means the input filter is whatever they have plus a sites robots.txt file. LAION is that dataset, filtered via ML, as of Oct 2022 or whatever.

a lot of people don't setup robots.txt correctly - and there's a lot of things on the public internet that really shouldn't be - and almost no one was looking at or cared about this stuff prior to diffusors

IDK if all of this is illegal under german law or not, but the way some of this works is foundational or at least has been the assumed status quo for a long time - and crawlers do do stuff that's in the public interest.

There's potentially a way to split this by declaring LAION's non-profit / academic pursuit thing bullshit because of the commercial layer and weak-as-hell-firewall RE: interconnectedness with stability and their commercial offerings, so I hope they do that or some other method.
I think LAION's research exception will not survive a legal case, yes, given what happened with Emad's nonsense. Part of the damage the current Scraping Happy Time has caused is damage to that crawler commons because folks have to defend their poo poo from this until the law starts doing its drat job, with proportionate effects on the commons.

StratGoatCom has issued a correction as of 23:29 on Apr 15, 2023

StratGoatCom
Aug 6, 2019

Our security is guaranteed by being able to melt the eyeballs of any other forum's denizens at 15 minutes notice


https://twitter.com/cooperlund/status/1647608609470140417

StratGoatCom
Aug 6, 2019

Our security is guaranteed by being able to melt the eyeballs of any other forum's denizens at 15 minutes notice


Delta-Wye posted:

see, no problem here you just have to be cheaper than automation. welcome to being in the modern labor pool, white collars.

There's also the other issue another thread regular pointed it out: It cannot reliably put out accurate poo poo where that is important, and it's outputs are the most bastardous poo poo imaginable for editing, because you have to sweep it carefully for subtle but serious errors.

StratGoatCom has issued a correction as of 03:38 on Apr 17, 2023

Adbot
ADBOT LOVES YOU

StratGoatCom
Aug 6, 2019

Our security is guaranteed by being able to melt the eyeballs of any other forum's denizens at 15 minutes notice



Good chance the aftertaste from this will render the AI winter to come even worse.

https://twitter.com/EepJeep/status/1647791841402662918

  • 1
  • 2
  • 3
  • 4
  • 5
  • Post
  • Reply