As nice as it would be to have a total AI commons, the reasoning is kinda bullshit. Copyright office is arguing this technology is different because it's "random" and "not possible to predict", but it's actually deterministic and merely hard to predict. Expecting this to get overturned the moment AI gets used by a major film studio.
Yep this is a good read on the situation, I don't even think overturning it is the only way to get around it, what if I retrain the last layer and copyright the resulting weights?
That wouldn't do anything. You can copyright the weights but not the model outputs, if this stands. Incidentally, I have no idea if slightly perturbing the weights of a model that falls under a certain license would then allow you to completely escape that license, my guess is that if the overall result is fairly similar the answer would be no.
Also, I do feel that this whole thing is a bit moot. If the author just claimed that they created the art they would probably be granted copyright protection.
As nice as it would be to have a total AI commons, the reasoning is kinda bullshit. Copyright office is arguing this technology is different because it's "random" and "not possible to predict", but it's actually deterministic and merely hard to predict. Expecting this to get overturned the moment AI gets used by a major film studio.
Yep this is a good read on the situation, I don't even think overturning it is the only way to get around it, what if I retrain the last layer and copyright the resulting weights?
That wouldn't do anything. You can copyright the weights but not the model outputs, if this stands. Incidentally, I have no idea if slightly perturbing the weights of a model that falls under a certain license would then allow you to completely escape that license, my guess is that if the overall result is fairly similar the answer would be no.
Also, I do feel that this whole thing is a bit moot. If the author just claimed that they created the art they would probably be granted copyright protection.