• The VOIDRUNNER'S CODEX is LIVE! Explore new worlds, fight oppressive empires, fend off fearsome aliens, and wield deadly psionics with this comprehensive boxed set expansion for 5E and A5E!

Sarah Silverman leads class-action lawsuit against ChatGPT creator


log in or register to remove this ad

Sacrosanct

Legend
That and the fact that they're completly ignorant of how AI works, are applying a double standard (how come Warhol can steal entire product designs,
Probably not the best example to use to try to prove a point....

 


Scribe

Legend
But, that's not new. Back in 1950s, they noted that the 60s would be ugly. In the 60s, the 70s were going to be ugly, and so on.

There never was a golden time when the future wasn't going to be ugly.

So if we, as a species that is supposedly intelligent, have been able to see that 'this is going to cause problems' its always a marvel to me that we just keep putting our hand in the fire, all for the sake of increasing the virtual numbers in some billionaires bank account.
 


FrogReaver

As long as i get to be the frog
No, it doesn't. Not in the way you are implying.
Agreed. A computer Reading the image necessitates a temporary copy. But from everything I’m reading the image isn’t actually stored as I had previously thought. The LLM is simply a large probability distribution across millions or billions or more of points. That’s what ultimately gets stored

Now if the argument is that the probability distribution somehow should count as containing a copy of the image in question then maybe, but I’d say this idea isn’t on very firm ground.
 

FrogReaver

As long as i get to be the frog
Well, i did say I have read most about Midjourney and everything I have read talks about how the engine examines patterns of relationship between text and images, then uses those patterns to output novel images based on text input. It doesn't create collages. If that isn't true, I'd love for you to point me toward a discussion that explains it.
So in some sense the probability distribution generated from training the AI on the dataset is itself the collage. That’s where all the different pictures tagged with cat are analyzed and their relationships stored. In some sense one might call that a collage of cat images.
 



Umbran

Mod Squad
Staff member
Supporter
The concern is that a great deal of the books the AI was trained on may have been acquired illegally.

Duchamp didn't steal that toilet. That would be a crime.

Well, it is less that they were "acquired illegally" and more that however the text was acquired, using it as training data results in reproduction without permission - violation of copyright.
 

Voidrunner's Codex

Remove ads

Top