3 ms·

The point of this survey is to improve the deepfakes. They aren't very difficult to spot when you've been primed and are paying attention, but the purpose here

by rad88 5y ago

The point of this survey is to improve the deepfakes. They aren't very difficult to spot when you've been primed and are paying attention, but the purpose here isn't to show off how good the fakes are or even challenge you to spot the difference -- it's to study what it is that gives a deepfake away. This'll ultimately be used to make the fakes more believable.

mitchdoogle 5y ago
If that's the case, then I wonder if it's more about the content of what is being said
unixbane 5y ago
I know you really want to point out how the main purpose of this is polling (training), but you had to make the claim that we are just accustomed to seeing "deepfakes". So I'm gonna tell you why that's not the case. This is maybe the 3rd time in my life I looked at any of the recent AI generated stuff since the breakthrough in the 2010s and they were all obvious. I got 6/6 correct on my first try. Like maybe if I never saw any normal video of any sort (NTSC, HD, LCD, CRT, mpeg) in my life then I wouldn't be able to tell, because I would have no conception of what the other 500 glitches caused by common video techniques are and if this is one of them as opposed to an edit. The first nimrod's face is wobbling and glasses are 2D and move in a different direction from his face. This is what I noticed in the first fraction of a second after opening the website, before even figuring out if this is what the test is or not. This is literally like the 90s when people were like "WOAHHH these 3D effects are so real", and doom, etc. I really hope this isn't the state of the art people keep telling me about. It wouldn't surprise me though, as all modern politics are based on make believe concepts. There's also the fact that this stuff is ultra low resolution, and at higher resolutions, mistakes will be even more obvious.
overtonwhy 5y ago
Maybe it's using A/B testing to build a data set for the generation algorithms.

by rad88 5y ago

mitchdoogle 5y ago

If that's the case, then I wonder if it's more about the content of what is being said

unixbane 5y ago

I know you really want to point out how the main purpose of this is polling (training), but you had to make the claim that we are just accustomed to seeing "deepfakes". So I'm gonna tell you why that's not the case. This is maybe the 3rd time in my life I looked at any of the recent AI generated stuff since the breakthrough in the 2010s and they were all obvious. I got 6/6 correct on my first try. Like maybe if I never saw any normal video of any sort (NTSC, HD, LCD, CRT, mpeg) in my life then I wouldn't be able to tell, because I would have no conception of what the other 500 glitches caused by common video techniques are and if this is one of them as opposed to an edit. The first nimrod's face is wobbling and glasses are 2D and move in a different direction from his face. This is what I noticed in the first fraction of a second after opening the website, before even figuring out if this is what the test is or not. This is literally like the 90s when people were like "WOAHHH these 3D effects are so real", and doom, etc. I really hope this isn't the state of the art people keep telling me about. It wouldn't surprise me though, as all modern politics are based on make believe concepts. There's also the fact that this stuff is ultra low resolution, and at higher resolutions, mistakes will be even more obvious.

overtonwhy 5y ago

Maybe it's using A/B testing to build a data set for the generation algorithms.