sus

inverted totalitarianism

though it could also just be called good old oligarchy with a thin veneer of democracy

I actually kind of looked at (jpeg) compression artifacts, and it's indeed true to the extent that if you compress the image bad enough, it eventually makes it impossible to determine if the color was originally flat or not.

(eg. gif and dithering is a different matter, but it's very rare these days and you can distinguish it from the "AI noise" by noticing that dithering forms "regular" patterns while "AI noise" is random)

Though from a few tests I did, compression only adds noise to comic style images near "complex geometry", while removing noise in flat areas. This tracks with my rudimentary understanding of the discrete cosine tranform jpeg uses*, so any comic with a significantly large flat area is detectable as AI based on this method, assuming the compression quality setting is not unreasonably low

*(which should basically be a variant of the fourier transform)

I recreated most of the comic image by hand (using basic line and circle drawing tools, ha) and applied heavy compression. The flat areas remain perfectly flat (as you'd expect as a flat color is easier to compress)

But the AI image reveals a gradient that is invisible to the human eye (incidentally, the original comic does appear heavily jpeg'd, to the point I suspect it could actually be chatgpt adding artificial "fake compression artifacts" by mistake)

there's also weird "painting" behind the texts which serves no purpose (and why would a human paint almost indistinguishable white on white for no reason?)

the new ai generated comic has less compression, so the noise is much more obvious. There's still a lot of compression artifacts, but I think those artifacts are there because of the noise, as noise is almost by definition impossible to compress

The thing missing here is that usually when you do texture, you want to make it visible. The AI 'watercolor' is usually extremely subtle, only affecting the 1-2 least significant bits of the color, to the point even with a massive contrast increase it's hard to notice, and usually it varies pixel by pixel like I guess "white noise" instead of on a larger scale like you'd expect from watercolor

(it also affects the black lines, which starts being really odd)

I guess it isn't really a 100% proof, but it's at least 99% as I can't find a presumed-human made comic that has it, yet every single "looks like AI" comic seems to have it

I also noticed how it suddenly went from great to crap.

But the real reason I think is ironically AI. It used to be that you could easily crawl the web, but after the AI craze, images suddenly became valuable to crawl and every website gets bombarded by scrapers, and are adding more and more countermeasures to make it more difficult, so tineye's own image scraper probably can't compete with them and so can't find any new images

I agree that the "arm things" are wrong, as it's pretty clearly just an 'artistic choice' that a human could very much do.

But that said these images are 100% provable to be AI. If you haven't built up the intuition that immediately tells you it's AI (it's fair, most people don't have unlimited time for looking at AI images), these still have the trademark "subtle texture in flat colors" that basically never shows up in human-made digital art. The blacks aren't actually perfectly black, but have random noise, and the background color isn't perfectly uniform, but has random noise.

This is not visible to the human eye but it can be detected with tools, and it's an artifact caused by how (I believe diffusion) models work

1. They claim to respect privacy and - to date - have done nothing to suggest that they don’t.

If you ignore all the fast and loose they play with privacy, sure, there is "nothing to suggest" they don't respect it.

IT’S OPTIONAL (there goes the “aggressive push” bit)

It's not an aggressive push if you ignore the part where they repeatedly use the foot in the door technique where they first promise they won't do something, and then later do it anyways.

They claim it is optional but they just shove a pop-up in your face about AI, while misleading you about how it works. This is about 1 step away from how most companies "allow" you to "preserve your privacy" by carefully clicking "no" to a long list of popups suggesting you give them cookies and share your emails etc.

This may be easy to dismiss as "problem between keyboard and chair" but when it predictably leads to many users thinking it's off but being surprised when they find it turned on without them realizing it it's not much consolation

NOTHING EXCEPT FOR THE PROMPT IS SENT TO MISTRAL (there goes the “reads all emails” bit)

How do you figure that works? The server somehow corrects your spelling mistakes without reading the email containing the spelling mistake? Again, End-to-end encryption is a core advertised feature of protonmail, and this completely sidesteps it while actively misleading users into thinking it doesn't

Sure you can look at it as just a bit of politicking (if a poorly thought out one), but it's really just the tip of the iceberg. Proton hasn't done anything that clearly crosses an unacceptable line, but they've made a lot of other highly questionable decisions in a relatively short timespan

oh, actually now that I looked it up closer, starting about 9 months ago they did a foot in the door manuever (a survey with leading questions followed up by misrepresenting the results) and then aggressively pushed an AI service that, you guessed it, tries to read all the emails you write and receive, totally undermining the end-to-end encryption. (the claim is it works locally, but most users have their data processed on the proton servers unencrypted)
And the way they did it is straight out of the enshittification playbook where they first promise that it's "business only" and then later try to push it to all users, and claiming it's off by default while it's actually on by default

https://pivot-to-ai.com/2024/07/18/proton-mail-goes-ai-security-focused-userbase-goes-what-on-earth/

(this article only covers the early portion of the debacle)

this isn't even all the problems with proton either, though all the other things are pretty minor by comparison (eg. quitting mastodon "because it's too expensive to maintain" (?))

The most popular alternative seems to be tutanota, though there should be a lot of alternatives though they may be very niche

(it seems tuta has some technical limitations if you want to do automated emailing, and the UI is a bit clunky, but it's not a privacy or security problem)

technically I guess it's between Cosmic latte and Desert sand

or "light grayish yellow"

..I had to look those up but it's definitely not a plain white

well it's supposed to be the "passive" you (or "indeterminate" you)

but I think anyone can learn to notice the pattern of comics with this particular odd color style ^(assuming^ ^you^ ^would^ ^want^ ^to^ ^learn^ ^it^ ^for^ ^some^ ^reason,^ ^and^ ^if^ ^we^ ^ignore^ ^color^ ^blindness^ ^or^ ^other^ ^similar^ ^problems^ ^etc)^

the trick is of course to look for the most disliked comments. Here's a couple

All humans are basically pure evil.

You are probably cruel and violent to vulnerable individuals more than three times a day.

Older men having sex with sexually mature teen girls is fine

DUI laws are too strict. It shouldn’t be all or nothing at .08 BAC but more severe punishments for more severe inebriation. .08 is pretty low and people who drink regularly can function fine at that level.

The Lemmy users who call themselves “Leftists” are garbage human beings

And most of them aren’t really that left. Trying to talk about abolition of police and prisons is something they would never agree to

wait, is this a roundabout way of calling yourself a garbage human being?