“This machine kills AI.”

@scottfgray It's crazy how people don't understand that AI is a tool, not a "get out of work" free card.

I use chatgpt for helping me in programming when I get stuck, but I don't just copy and paste code from it or tell it "Write me the whole program" lol

@charadon @scottfgray And even that is dicey, given the phenomenal amount of illegally obtained and processed work that was used for the AI and where the code you ingest into ChatGPT goes (in case of proprietary software you work on). Not to mention the ethical aspects.

The only model I'd ever consider using for anything important is a self-hosted one that went all and above to ensure ethically correct training data and is open source. Something like Mistral-7B but more ethical (I guess).

@Natanox @charadon @scottfgray What do you think is a way to guarantee that training data was collected ethically correct? Do you believe that training on this more ethical corpus can lead to competetive results like Mistral-7B or LLaMA 3? Is your opinion that one of these models was trained more ethically?
@feliks @Natanox @charadon @scottfgray I think the only way to guarantee a dataset was obtained ethically and legally is to require it be published openly for anyone to investigate. Will this lead to better deep learning models? Of course not, but that's exactly the point: we shouldn't compromise ethics for the sake of a potential benefit for one group.

@feliks @Natanox @charadon @scottfgray Ethical training data won't help the problem described here - prompt "engineers" cannot tweak images or prose based on specific feedback.

You can' fix a massive Dunning Kruger machine that doesn't know its limitations.

@feliks @Natanox @charadon @scottfgray

The company displaying the licence of the works they used to train their pile of linear algebra. That is the only proof

@feliks @Natanox @charadon @scottfgray

I don't care about competitive. Stealing money is very profitable and competitive too. It does not mean I appreciate people getting mugged. Art is pretty expensive, it is a time consuming prospect, and so it programming, and so is writing...

The work of artists and programmers is being used and processed into other works without abiding to the licenses set for transformative work, which AI output definitely is at the very least

@Archivist @Natanox @charadon @scottfgray I believe this hardly to be feasible. Do you?

@feliks @Natanox @charadon @scottfgray

Hardly feasible? Tusky, an open-source and free app for Fediverse, lists 20ish licenses of the code they used. Not only is it feasible, but by law as written it is required.

@feliks @Natanox @charadon @scottfgray

Google Classrooms lists hundreds of licenses of the code from open source project they used. It is feasible and required. Credit must be given where it is due.

@feliks @Natanox @charadon @scottfgray stopped counting at 910 licenses, wasn't half way there
@Archivist @Natanox @charadon @scottfgray You seem to be talking about code while I try to talk about image data. There is a significant difference in workload to these modalities: People tend to tag their software with licenses more than images are being tagged in their metadata. The latter is a massive challenge in processing data properly

@feliks @Natanox @charadon @scottfgray

Untagged images should always assume to be protected. Images with licenses should they be collated in a model should remain protected under their same license. The same is applied with software.

Also AI models do not respect software licenses any more than they do artistic licenses. Which is definitely a problem

@feliks @Natanox @charadon @scottfgray

Derivative work is derivative work, the way it was derived is irrelevant

@feliks @Natanox @charadon @scottfgray

Any AI generated image for which a single CC-BY-SA was used to create the model needs to, according to the license, be tagged with the name of the original author, and be shared under CC-BY-SA. Copyleft exists for images too, so do intellectual property rights

@Archivist @Natanox @charadon @scottfgray I see these license breaches also as a problem, same as you. I can only hope that this leads to a similar moment as with OpenWRT but I doubt it. On the other side I don't see these systems could have been created without the vast amount of training data. In a competitive environment it appears to be a competetive advantage to neglect these licenses. Might shaping of or removing competition from this environment be a valid goal? I'm not sure

@feliks @Natanox @charadon @scottfgray like I said before, mugging people gives you a very competitive advantage in society. So does threatening minorities. So does slavery. So does human trafficking. So does organized crime.

A system that can't exist without committing large scale intellectual property fraud should not exist.

@scottfgray
This is so relatable. I have noticed this too when working with AI that they are sometimes just completely incapable of modifying an exiting piece. They get stuck with something and when I ask it to just modify it, it'll just come back 95% same but with a change I did not ask for. It's just frustrating to use.

@tocisz
this machine kills Al

é

LORENZO MONTATORE= "28 wah

% Artists Against Generative Al - Join
“© Danielle Sanfilippo - 13h -@​
Posting this on behalf of a member who would like to remain anonymous:

I'm an art director and supervisor for a large studio. The studio heads had the bright idea
before I started to hire prompters. Several bros were brought onto the film project. I
absolutely hated myself for not quitting on the spot but stuck with it because it’s
mercenary out there. Have a family to feed etc. I decided to use this time wisely. Treat
them as I would any artist I had hired. First round of pictures of a sweeping Ariel forest
landscape comes through and it’s not bad. They submit a ton of work and one or two of
the 40 are ok. Nearly on brief. So first round feedback goes through and I tell them about
the perspective mistakes, colour changes I want, layers that any matte painting would be
split into. Within a day I get 5 variants. Not changes to the ones I wanted but variations.
Again. Benefit of the doubt I give them another round of feedback making it clear. Next day
it’s worse. I sit there and patiently paint over, even explaining the steps I would take as a
painter. They don’t do it, anomalies start appearing when I say I want to keep the exact
image but with changes. They can’t. They simply don’t have the eye to see the basic
mistakes so the Ai starts to over compensate. We get people starting to appear in the
images. These are obviously holiday snaps.

“Remove the people”

“What would you like them changed to?”

“... grass. I just don’t want them there”

They can’t do it. The one that can actually use photoshop hasn't developed the eye to see
his mistakes, ends up getting angry at me for not understanding he can’t make specific
changes. The girl whose background was a little photography has given me 40
progressively worse images with wilder mistakes every time. This is 4 days into the project.
I'm both pissed about the waste, but elated seeing ai fall at the first hurdle. It’s not even
that the images are unusable, the people making them have no eye for what's wrong, no
thicker skin for constructive criticism and feedback, no basic artistic training in perspective
and functionality in what they’re making.

Yes the hype is going to pump more money into this. They won't go anywhere for a while.
But this has been such a glowing perfect moment of watching the fundamental part fail in
the face of the most simple tasks. All were fired and the company no longer accepts Ai
prompters as applicants. Your training as an artist will always be the most important part of
this process and it is invaluable. I hope this post gives you a boost in a dark time.

@scottfgray just the facts - tellement factuel 👌

@scottfgray Here's the full text from original source:

Posting this on behalf of a member who would like to remain anonymous:
I’m an art director and supervisor for a large studio. The studio heads had the bright idea before I started to hire prompters. Several bros were brought onto the film project. I absolutely hated myself for not quitting on the spot but stuck with it because it’s mercenary out there. Have a family to feed etc. I decided to use this time wisely. Treat them as I would any artist I had hired. First round of pictures of a sweeping Ariel forest landscape comes through and it’s not bad. They submit a ton of work and one or two of the 40 are ok. Nearly on brief. So first round feedback goes through and I tell them about the perspective mistakes, colour changes I want, layers that any matte painting would be split into. Within a day I get 5 variants. Not changes to the ones I wanted but variations. Again. Benefit of the doubt I give them another round of feedback making it clear. Next day it’s worse. I sit there and patiently paint over, even explaining the steps I would take as a painter. They don’t do it, anomalies start appearing when I say I want to keep the exact image but with changes. They can’t. They simply don’t have the eye to see the basic mistakes so the Ai starts to over compensate. We get people starting to appear in the images. These are obviously holiday snaps.
“Remove the people”
“What would you like them changed to?”
“… grass. I just don’t want them there”
They can’t do it. The one that can actually use photoshop hasn’t developed the eye to see his mistakes, ends up getting angry at me for not understanding he can’t make specific changes. The girl whose background was a little photography has given me 40 progressively worse images with wilder mistakes every time. This is 4 days into the project.
I’m both pissed about the waste, but elated seeing ai fall at the first hurdle. It’s not even that the images are unusable, the people making them have no eye for what’s wrong, no thicker skin for constructive criticism and feedback, no basic artistic training in perspective and functionality in what they’re making.
Yes the hype is going to pump more money into this. They won’t go anywhere for a while. But this has been such a glowing perfect moment of watching the fundamental part fail in the face of the most simple tasks. All were fired and the company no longer accepts Ai prompters as applicants. Your training as an artist will always be the most important part of this process and it is invaluable. I hope this post gives you a boost in a dark time.

@scottfgray As an EE who has been told for >20 years that PCB designers will be soon obsolete because auto-routers will do our job faster and better.... Well, I'm still waiting.

Sure, generative AI has made things possible that we didn't think computer could possibly do in our lifetimes. But there is a big difference between generating A picture and generating The picture the customer wanted.

@scottfgray LLMs are useless for the applications commonly found for them. They are tools but only if you're okay with getting extra work to fix, they can't repeat anything with any consistency besides stuff that makes people say "this is wrong".
@scottfgray it really pisses me off that the author described the woman prompter as a girl. Juuuust calling that out.
@RomanceReviews @scottfgray oh there's cultures that call any age Boy/Girl.. but IT has to be equal, I agree
@scottfgray There is a subtle hidden layer to choosing a *2B* pencil that was probably unintentional
.
.
(the main character of the game is called 2B, an android soldier of mankind ordered to fight against the unintentionally intelligent machines)
@scottfgray the killer combination for AI usage is always: somebody who could do it themselves but gets a headstart due to AI. This goes for writing, drawing, etc. the problem: AI is treated as a way for skill-free people to suddenly be enabled (not getting enabled, but already are enabled), which is simply impossible
@shinjiikarus @scottfgray and all the lost inspiration and creativity during the initial part of the creative process...
@mfeilner @scottfgray obviously! Again: you need to be able to do all parts of this anyway and you need to keep that skill sharp and refined, but then generative AI can help you some part of the way (and if it’s only to provide a scribble you completely recreate from scratch for your final work)
@scottfgray AI Tools are the Italian cars of our times.

@scottfgray The generic genericizer can't handle specific demands.

Who would have thunk?

@scottfgray "he can't make specific changes" is the best bit of this.
@scottfgray Very recognizable. I played around with some AI Agent frameworks, and under the hood those are built on hard-coded prompts that attempt to make an LLM produce structured output. Because of the statistical nature of LLM's this is hopelessly unreliable so the developers try to increase the odds of getting usable output by adding encouragement ("You are very good at your job") or even threats ("Give your best answer, your job depends on it!") to the prompts.

Posting this on behalf of a member who would like to remain anonymous:

I’m an art director and supervisor for a large studio. The studio heads had the bright idea before I started to hire prompters. Several bros were brought onto the film project. I absolutely hated myself for not quitting on the spot but stuck with it because it’s mercenary out there. Have a family to feed etc. I decided to use this time wisely. Treat them as I would any artist I had hired. First round of pictures of a sweeping Ariel forest landscape comes through and it’s not bad. They submit a ton of work and one or two of the 40 are ok. Nearly on brief. So first round feedback goes through and I tell them about the perspective mistakes, colour changes I want, layers that any matte painting would be split into. Within a day I get 5 variants. Not changes to the ones I wanted but variations. Again. Benefit of the doubt I give them another round of feedback making it clear. Next day it’s worse. I sit there and patiently paint over, even explaining the steps I would take as a painter. They don’t do it, anomalies start appearing when I say I want to keep the exact image but with changes. They can’t. They simply don’t have the eye to see the basic mistakes so the Ai starts to over compensate. We get people starting to appear in the images. These are obviously holiday snaps.
“Remove the people”
“What would you like them changed to?”
“… grass. I just don’t want them there”
They can’t do it. The one that can actually use photoshop hasn’t developed the eye to see his mistakes, ends up getting angry at me for not understanding he can’t make specific changes. The girl whose background was a little photography has given me 40 progressively worse images with wilder mistakes every time. This is 4 days into the project.
I’m both pissed about the waste, but elated seeing ai fall at the first hurdle. It’s not even that the images are unusable, the people making them have no eye for what’s wrong, no thicker skin for constructive criticism and feedback, no basic artistic training in perspective and functionality in what they’re making.

Yes the hype is going to pump more money into this. They won’t go anywhere for a while. But this has been such a glowing perfect moment of watching the fundamental part fail in the face of the most simple tasks. All were fired and the company no longer accepts Ai prompters as applicants. Your training as an artist will always be the most important part of this process and it is invaluable. I hope this post gives you a boost in a dark time.

@scottfgray Early prompt engineers trying to improve the outcome. 😉
https://youtu.be/HbDnxzrbxn4?si=AvDN4FG45JCWREMo
Elevator Recognition | Burnistoun

YouTube
@scottfgray Ah yes, it "kills AI" in the same way Valve's VAC and Riot's Vanguard "killed cheating"...
That is to say: Temporarily at best.
@finlaydag33k @scottfgray We've been scribbling art on cave walls since the beginning. We still do all the same and we won't stop even when the Skynet's "art" takes over. At least, there is no logical reason for that to happen. So... are you sure about that?
@scottfgray im gonna be impressing it upon others every time this comes up that something that can get 90% of the way there when you don't have the skills to do the last 10% is not doing you any favors, and the diminishing returns on how "good" these models can get seem to suggest it's not getting much better than that. and if you do have the chops to do that last 10%, you don't need the 'ai' to do the first 90%, you would have done it better yourself and not had to clean up a bunch of its mistakes after. there's no point on this gradient where 'ai' is helping anyone
@chrisisgr8 it still pisses me off that my coworkers whine about bugfixing not being real work while being perfectly fine fixing code generated by chatgpt
@robinsyl how the fuck is bugfixing not real work LMAO. this is why so much software is fucked up
@robinsyl @chrisisgr8 As an AV programmer who spends almost a majority of my time trying to fix bugs on a regular basis, I can assure them that bugfixing is a real job, and it's a LOT harder than people think! You really have to dig deep into the code and figure out the cause of the issues. And sometimes, the issues won't show up until long after I moved on!
@scottfgray If you ever feel impostor syndrome, remember there are folks out there who claim "AI prompter" as their job title
@scottfgray
Mechanization is used to devalue the labor of highly skilled crafts people, not eliminate them from the workforce.
@Antiroo @scottfgray In a way, it creates a particular demand for traditional pieces as generated crap has become more saturated. Automation is absolutely a damaging blow to the common person, but for me it's put into perspective where true value lies.

@crocodisle @Antiroo @scottfgray it’s really annoying that I always liked high-effort, supe detailed anime art portraits and JUST started learning how to make them sort of look like that — boom, imitative AI to make everyone suspicious of it.

Guess I have to learn stylised low-detail (but still technically very skilled and crafted) cartoon style. Which would be fine if that’s what I actually want to be able to make.

@MxVerda @Antiroo @scottfgray I think there's still space for high detail anime-style pieces. How it gets shared becomes more prickly for sure. Perhaps sharing process videos or sketches would give watchers the benefit of the doubt.

I never considered that in our lifetimes we'd have such a bombshell to contend with when it comes to finding out what's real or not online. Sorry to hear it's had such a negative effect for you!

@crocodisle @Antiroo @scottfgray ah, nw. I never made money from it and I’ve got friends I can show it to anyway.

I heard those process videos are / will be used for training the next round of imitative AI, so yay.

I appreciate your positivity tho. It feels nice online 😽

@scottfgray I was in a conversation a couple of days ago where a manager expressed "You'd be surprised at how much resistance there is among developers to using Copilot. I guess it's a fear that it's going to take their jobs".

My guy, we just don't want something _else_ making mistakes we'll have to clean up. I can do that on my own.

@scottfgray I can post the whole thing as a reply to your post if you want, so that no one else would have to open the link on that dumpster platform. I used private browsing, but still ominous, that's why I'm saying
@scottfgray
Yeah, AI needs another jump in capability similar to the last time we obsessed over it as a society (the 80s I believe) before it's a credible threat to _any_ competent human. I do think it currently can be a useful tool for a skilled artist (There's at least one professional glamour photographer doing cool things with it as a post-processing step in his photographs) but a replacement for someone that knows what they're doing? Absolutely not.
@nikki
@scottfgray Their company seriously just hired random people with no artistic background but a pretty AI portfolio? Wtf.

@scottfgray I'll never understand how "AI prompter" is an actual job now… how did this many people fall for this con… like, it's not even elaborate

It's fun watching capitalist pigs eat shit though. Happy meal y'all

@scottfgray You can have the best vacuum cleaner and still be shit at vacuum cleaning.
@scottfgray As I always said, AI prompting is only usefull if you are already an artist, have a vision, and are able to manually fixes mistakes or even redo the whole thing but use it as an inspiration. Photoshop and photography reduced the need for painters but not for artists. Their is no reason AI wouldn't be the same. A good artist with ok-ish prompting skills is always going to be better than an amazing AI prompter.
@scottfgray
AI: Fake it till you fake it.

AI may have value in the hands of artists who understand the field and still have all the other skills necessary for the work. It will be another tool in their already well-stocked toolbox.

Thinking this new tool can replace expertise and all other tools is stupid.

@scottfgray

Yep, I realize this super quick when messing with local install of an image generator. Fancy quick pictures but good luck trying to get a super specific quick change on that and still holding the rest. Unless one has a high skill of photoshop bashing.

Big reason I am like, nah, I go back to just drawing by hand. Plus, I got bored so quick with messing with any of those generators. Plus, I find far more enjoyment in exploring and improving my style by practice.

@scottfgray the tech bros are wasting so much time of competent people, it's amazing companies aren't suing them for it.