Why is DALL-E a scam?
twitter.com
twitter.com
I find it quite amusing that there seems to be overlap between people who think IP/copyright law needs an overhaul, and people who think that Github Code Pilot and DallE are exploitive.
I use them, I do not think they are evil. I can also appreciate the asymmetry of using code with restrictive licenses or private to train an ML vs the consumer visibility over, for instance, windows source code. In the case of DallE I find it fascinating and useful, and can likewise appreciate the asymmetry of having the compute power and placing restrictions on which prompts are allowed,
It is not difficult to see how those asymmetries can be seen as exploitative. I don’t have a side but the thought is easy to entertain.
Edit: grammar
If I draw a picture and release it into the world; I'm familiar with that process. I know, it'll be viewed and consumed by human eyeballs and I understand to some degree their capabilities and motives, or maybe I pretend I do. The point is, the release of the content into the wild was done with some understanding of its consequences.
Now, along comes DALL-E, it's an unfamiliar, new class of a consumer. It'll swallow anything you produce and turn it into profit for another human, right before your eyes. It'll also do so in such a way without _recognizing_ the contributors to its success. DALL-E is a black box where the effort of creators goes in and someone else pocket is made fatter.
The key point for me is that DALL-E would be worthless without it consuming other peoples work. Folks here have mentioned how they've consumed content over their lives that others have produced... yeah, but, with enough time you could have got there yourself. DALL-E simply cannot. It is absolutely necessary for it to consume the content of others.
Also, in the paid version, all rights accrue to the user, not to OpenAI, so that assertion is false.
I wish we had 2 different words for these things. Artist has a different ring than graphic designer.
One deserves to be protected, and I don't think will go away anyway, the other may well be automated away, but will invoke the former to protect itself.
Cosmopolitan leaked how the process looks like: https://www.cosmopolitan.com/lifestyle/a40314356/dall-e-2-ar...
https://hmg-h-cdn.hearstapps.com/videos/creation-loop-165578...
At the end of the day dalle is a fancy Content-Aware Fill on steroids(style translation), pasting mishmash of copyrighted data scrapped off the web.
if it were only trained on public artwork (where copyright has expired) would that still be a problem?
The history of private property is rife with abuse and exploitation. We should not rush to extend such ideas into new arenas.