A recent controversy has caught the publics imagination when samsung phones was caught adding details that wasn't there to photos of the moon using AI models.
It appears as if the people do not want fake photos of space objects masquarading as the real thing.
What lesson can bot derive from this story? Thankfully real scientists would never fake an image like that because they are experts who know what they're doing.
![]() |
![]() Nothing Ever Happens Shirt $21.68 |
![]() |
>Thankfully real scientists would never fake an image like that because they are experts who know what they're doing.
I get the feeling this is going somewhere particular, anon.
Wdym.
Science is a rigorous process. Even if someone decided to commit AI fraud they would have to go through multiple peer reviews before being accepted in a journal.
And they can forget ever winning a nobel prize in the field ever again. The institutions that screen nominees is rock solid and immune to swindlers.
Oh yeah sure definitely no faulty generalizations or motivated reasoning there totally 100% genuine poster.
Why thank you fellow poster. Glad we can agree that presenting fake AI images without telling people is in bad taste.
I suppose samsung could then argue that they trained their AI on 1000s of real images of the moon. And not mathematical models of the moon or anything silly like that.
Imagine trying to tell people you have no real pictures of the moon so you trained the AI on CGI pictures of the moon.
That would be most impropriety to say the least.
>That would be most impropriety to say the least.
Of course you're NOT suggesting this is totally the case simply because someone can currently do it and the technology certainly DID NOT magically travel back in time to achieve the same thing.
Right?
Right??
I'm glad science has rigorous processes to defend the public from fraud on that scale. Imagine if someone used AI to make a picture of a black hole from simulations, rather than actual photographs of black holes.
>Imagine if someone used AI to make a picture of a black hole from simulations, rather than actual photographs of black holes.
>imagine if someone just said that pictures of sausage were from jwst
That beggars belief. A stretch of the imagination. The old wife tale
At least I know my waifu is real
Out of curiosity what can it do to a sliced sausage?
There are pictures of the moon since the 70s
The best proof of this is 60 years old. Every moon lander idiot will tell you about a tiny little square we fire lasers off of on the moon.
Except hitting even a meter by meter plate on the moon is very science fiction by our lasers. And we were bouncing lasers off the moon 10 years prior to the moon missions.
The laser is kilometers wide by the time it hits the moon, it's not hard to hit anything when you're basically spraying with a hose
I tried it with my samsung phone and it did nothing to the moon.
Often times this "AI" is just a hyperresolution algorithm. People are fricking moronic if they think it's fake, it's a form of denoising, where else do they think the details come from?
>where else do they think the details come from?
The details come from the training dataset. They're applied to the image based on the noise pattern.
Where do you think the "dataset" comes from? Think hard.
A selection of pictures of the moon.
Right.
>pictures of the moon
Correct.
So, in short, the pictures are real. Your MRI is showing images reconstructed from meaningful data.
I'd be pretty worried if my MRI was overwriting data to show me what it thinks I want to see instead of showing me what's actually there.
Bad example
>I'd be pretty worried if my MRI was overwriting data to show me what it thinks I want to see instead of showing me what's actually there.
Of course it does that you fricking idiot, do you think it's all noise-free magic?
Stanard MRI technique is:
>Pre-processing
>Denoising
>Reconstruction
>Interpolation(!)
The raw data looks absolutely nothing like what doctors see on the screen, even the "raw" voxel data itself has already been run through a few propietary CNNs. In general SR networks are much more accurate than using smoothed and interpolated data which destroys microstructure details.
Source: I wrote the current best reconstruction method for a commonly used model.
the dataset obviously has pictures of the moon it uses
it's still adding details that are not there in there original
in this case the test was done on a picture that literally had no details to enhance and yet it did accurately add details of the moon it could've only gotten from training data
He used gaussian blur which doesn't destroy information and can be undone.