FulIden

498 posts

FulIden

FulIden

@FulIden

Katılım Ağustos 2022
41 Takip Edilen15 Takipçiler
FulIden retweetledi
OpenAI
OpenAI@OpenAI·
Today, we share a breakthrough on the planar unit distance problem, a famous open question first posed by Paul Erdős in 1946. For nearly 80 years, mathematicians believed the best possible solutions looked roughly like square grids. An OpenAI model has now disproved that belief, discovering an entirely new family of constructions that performs better. This marks the first time AI has autonomously solved a prominent open problem central to a field of mathematics.
English
693
2.7K
19.5K
7.1M
FulIden retweetledi
Variety
Variety@Variety·
Nicolas Winding Refn on AI: “Having now tried it on something later on that may show here, I really love the creativity. For me, it’s like a brush. And obviously, no one really knows all the implications of what this is going to do and what’s going to happen, but from the perspective of creativity it’s a new invention. And then it’s what you’re going to do with it.” variety.com/2026/film/fest…
English
110
43
391
665.7K
FulIden retweetledi
Tom Green
Tom Green@tomgreenlive·
People worry that AI is going to replace art because you can make some garbage action movie without having to use all the traditional ways of making a garbage action movie. I just think it’s going to evolve things. People will start to see that modern movies are often actually kind of corny. AI or not. And then maybe people will get to back to telling real stories. Get away from some of the storytelling techniques that have consumed film over the last 20 years. Anyway, here’s my new movie I made it this morning. #Ai
English
206
74
1.2K
72.9K
FulIden
FulIden@FulIden·
@Rahll @TheDutchRuler It's relevant if you're calling it compression w/o it being able to decompress 99.999% of the data And I don't think it's problematic that some people will generate super famous images when they ask for them Is it even as problematic as human artists unconsciously plagiarizing?
FulIden tweet mediaFulIden tweet media
English
0
0
0
22
Reid Southen
Reid Southen@Rahll·
@FulIden @TheDutchRuler It's irrelevant because that's not how they're trained and deployed. I don't care about how it could theoretically work under specific circumstances, I care how it DOES work in the world today.
English
1
0
1
67
Reid Southen
Reid Southen@Rahll·
You've just described stealing people's work and then kicking them out of the party like that's a positive development in arts and entertainment. "We don't need you anymore, we've taken it all already," isn't the win you think it is.
Josh Daws@JoshDaws

Solo artists are much more threatened by AI because they are needed by artists that work in collaborative mediums like film and animation. Generative AI allows the latter to exercise their craft without the massive budget required to pay the former.

English
10
73
563
8.5K
FulIden
FulIden@FulIden·
@Rahll @TheDutchRuler I don't think it's problematic or compression. I think if every image in the dataset had only 1 copy (no duplication), the model wouldn't be able to regurgitate any images at all. I'm not certain of course, but based on the characteristics of all your examples, it's likely
English
1
0
0
55
FulIden
FulIden@FulIden·
@Rahll @TheDutchRuler No, what I'm saying is that the vast majority of the images can't be reproduced, i.e. it can only reproduce a miniscule fraction. I'd be surprised if it can reproduce >0.01% of the images. And we're talking about whether it's "compression", not whether it's "problematic"
English
1
0
0
52
Reid Southen
Reid Southen@Rahll·
@FulIden @TheDutchRuler It doesn't have to be able to reproduce the "vast majority" of images to be an issue, lol, that's never been the metric. The fact that it can do it at all and at a problematic scale is what matters, but do move those goalposts.
English
1
0
1
55
ToonHive
ToonHive@ToonHive·
Glitch Productions responds to Party Animals over their AI video contest.
ToonHive tweet mediaToonHive tweet media
English
64
853
11.2K
326.2K
FulIden
FulIden@FulIden·
@Rahll @TheDutchRuler I doubt it can restore a lossy version of anything but a very tiny subset of highly overfitted, famous images Pixel relationship data (transformative!) simply doesn't have enough information to reproduce the vast majority of the images. I'm open to seeing proof showing otherwise
English
1
0
0
52
Reid Southen
Reid Southen@Rahll·
Lmfao, the reason a diffusion model or LLM is so good at compressing is because it references millions of data points and draws relationships between them so it doesn't have to store an image's pixel data, but can restore a lossy version of it when needed based on those other relationships. No one is arguing you can use AI to compress a single thing by 40,000 times, it's the presence of other data that allows the compression in the first place.
English
1
0
1
70
FulIden
FulIden@FulIden·
@Rahll @TheDutchRuler That's pretty convenient that you can claim it's reproducing non-famous images too, it's just that nobody has noticed. You could use image search tools to prove it. Much more likely is that famous works are reproducible because they've been highly duplicated in the training data
English
0
0
0
7
Reid Southen
Reid Southen@Rahll·
@FulIden @TheDutchRuler It's not figurative at all. The reason famous works are easy is because most people can recognize them when they see them, do you expect the average person to have a searchable database in their brain? arxiv.org/abs/2309.10668 x.com/i/status/18752…
Marge Nelk@NelkMarge

@FLOSSTRADAMUS @LinusEkenstam Copyrighted works go in and out comes uncopyrightable stuff? Yeah, it's data laundering. AI models also memorize the training data. The "training" is the process of lossy compression.

English
1
0
1
70
FulIden
FulIden@FulIden·
@Rahll @TheDutchRuler Can you take a single image and compress it to something roughly 40,000x smaller in filesize (based on the numbers in that lawsuit and the size of stable diffusion/LAION 5B), and be able to decompress it again?
English
1
0
0
69
FulIden
FulIden@FulIden·
@Rahll @TheDutchRuler Yeah this is what I mentioned, it's your one source of proof for LLMs being "compression". Emad is clearly speaking figuratively. Where is the "decompression" part for non-famous and overtrained works? Why can't you "decompress" anything but blockbuster movie screencaps?
English
2
0
0
72
FulIden
FulIden@FulIden·
@Rahll @TheDutchRuler Does it? I'd be surprised. What, except for movie/tv screencaps, would be labeled as "screencap" in the training data? If this was proof that LLMs are "compression", and not just proof of overfitting, you could show the models regurgitating your own artwork, for example
English
1
0
0
64
Reid Southen
Reid Southen@Rahll·
@FulIden @TheDutchRuler Lmao, if landscape resulted in a famous landscape photo, you'd probably find a stupid way to defend that too.
English
1
0
9
156
FulIden
FulIden@FulIden·
@Rahll @TheDutchRuler "screencap" is very descriptive, it means a screen capture from a movie. If you don't want screenshots from movies, don't ask for them. If you could get a model to regurgitate anything besides ultra-famous, highly data-duplicated movie stills, that would be much more damning
English
1
0
0
152
Reid Southen
Reid Southen@Rahll·
How is that literally any different than typing cinematic keyframe or something of the like? You're supposed to describe the aesthetic you want, no? Furthermore, it will just give you endless, distinct, copyrighted properties from one, nondescript term, and you think that's totally normal, lol. It's the definition of plagiarism bud.
English
2
0
10
165
Reid Southen
Reid Southen@Rahll·
@TheDutchRuler I've never accidentally drawn a Spiderman or Harry Potter or Darth Vader without intending to, have you.
Reid Southen tweet media
English
1
0
10
240
Reid Southen
Reid Southen@Rahll·
@NotThatNATO @EpicArtFail Lmfao, you're a real surface level guy, eh. "Well, they call it learning, so don't actually look into how the tech works, I'm going to hinge it all on that."
English
3
0
1
44
FulIden
FulIden@FulIden·
@DullClayCutter @BVasquezart1 @PocahantosLong2 @Rahll @JoshDaws You're welcome to your definition of art rooted in historical context. I think that might help someone produce BETTER art, but I don't think any of that is necessary to produce art. Stacking 3 rocks on top of eachother on a whim is art imo, if it's presented as such
English
0
0
0
17
A thatched hut, secluded
A thatched hut, secluded@DullClayCutter·
@FulIden @BVasquezart1 @PocahantosLong2 @Rahll @JoshDaws Nobody honest says that the director is the most responsible for creating the movie. That's a dishonest sideline. The entire point of filmmaking, same as theater, is that it is a Gesamtkunstwerk, a "total work art", i.e. a collaborative effort all or many disciplines.
English
1
0
1
40
A thatched hut, secluded
A thatched hut, secluded@DullClayCutter·
@FulIden @BVasquezart1 @PocahantosLong2 @Rahll @JoshDaws "It doesn’t matter how the art was produced." Yeah, no, that's a cop-out. All his examples were CONSCIOUSLY made UNDERSTANDING what is being done in the context of art history. That's not ai, that is completely different. Still moving the goalpost from the degree of separation.
English
2
0
1
24
BrianVasquezart
BrianVasquezart@BVasquezart1·
@FulIden @DullClayCutter @PocahantosLong2 @Rahll @JoshDaws Are you dumb? I’m bad faith?! You can’t even answer my question without acting like your opinion isn’t etymology lol. You’re SOL and you’re trying to play victim. I answered all of them. I’m not asking for Your opinion and you’re intentionally obfuscating. Coward behavior.
English
1
0
1
25

Keşfet

@Rahll @TheDutchRuler @partydude1719 @canoflop @ToonHive @elonmusk @BarackObama @taylorswift13