News for hackers. Stuff that matters.
Re:Mark Twain filmed by Thomas Edison in 1909 by zamadatix on Sunday October 15, 2023 @09:26PM (#37895012)
At what point does it stop being film of Mark Twain and start being what someone/a computer imagines film of Mark Twain to be like? Most of the data in this, from the color to the frames themselves, did not actually exist and, while looking reasonable, very likely does not agree with what the actual data would have been.

One other downside of this kind of approach is it makes high quality versions of the source significantly harder to find (I doubt the 240p encoding with blocks bigger than Mark Twain's head is the best copy of the source available).

Re:Mark Twain filmed by Thomas Edison in 1909 by proc0 on Monday October 16, 2023 @12:59AM (#37896075)
Arguably the tech at the time did not allow Mark Twain to be properly captured at all, so even the original contains imperfections that distort the image. A high res AI version that restores it based on statistical training is still better and would show an "educated guess" of what he looked like.
Re:Mark Twain filmed by Thomas Edison in 1909 by imiric on Monday October 16, 2023 @03:10AM (#37896664)
To be fair, all photography is an approximation of reality. Whether that approximation is improved by color data added at time of capture or a century later, or if the image is cleaned up by AI instead of a human, doesn't change that fact.

Most images taken with a smartphone today don't capture reality reliably either, and you're allowed to alter them to look whichever way is most pleasing.

Even our eyes perceive reality slightly differently, and photography is not much different. If anything, looking at 2D reproductions of reality adds another layer of subjective perception.

Re:Mark Twain filmed by Thomas Edison in 1909 by bloak on Monday October 16, 2023 @06:50AM (#37897977)
For me there's a big difference between applying a simple algorithm to the data, a formula which someone could scribble down on the back of an envelope, on the one hand, and, on the other hand, some neural network thing with millions of weights that might be hallucinating.

Even compression algorithms can hallucinate: https://www.theregister.com/2013/08/06/xerox_copier_flaw_mea...

I want my historical records to be processed only with simple algorithms that help me see what is really there.