Embed this noticepistolero (p@fsebugoutzone.org)'s status on Wednesday, 28-May-2025 16:12:36 JST
pistolero:mgsgb_0: Microsoft wants to take a continuous stream of screenshots of your system. They have been keylogging for a long time. :mgsgb_1: Microsoft has been using Copilot to spy. ( https://www.schneier.com/blog/archives/2024/02/microsoft-is-spying-on-users-of-its-ai-tools.html ) :mgsgb_2: Storage remains cheap; Microsoft's cluster is way up the list on top500.org. :mgsgb_3: Eating as much data as you can and keeping it around forever in case you figure out how to use it at some point in the future is espoused by Microsoft, Google, Facebook, everyone, all of the large tech companies. Data is considered an asset, active user metrics represent not just ad inventory but also a continued stream of data. :mgsgb_4: AI prompts are part of that data, kept around forever, besides being used to train new versions of the AI. :mgsgb_5: AI image generation tools produce unique images. :mgsgb_6: Image search is old tech. You can build it from scratch in your house: you normalize the image, you adjust color to maximize noise and then you boil it down to a 8x8 1bpp square first so that you have a 64-bit index, and that's often enough for a match, but if you have a lot of really similar images, you can just treat that as a constraint on the search space. It's not exactly CS-101, but it's something you can do maybe two or three years in. :mgsgb_7: Microsoft makes a big chunk of its money from government contracts. PhotoDNA came out of Microsoft Research. :mgsgb_8: AI-generated images are tied to an account: if not a credit card, at least an identity with IP/email/time, maybe a phone number for 2FA. :mgsgb_9: AI-generated images are almost certainly kept around. It is not hard at OpenAI's (i.e., Microsoft's) scale to retain a searchable corpus of all previously generated images. :mgsgb_a: People are using AI tools to make avatars. :mgsgb_b: None of those people are actually anonymous.
@p >Microsoft wants to take a continuous stream of screenshots of your system. They have been keylogging for a long time. how is this different from .bash_history
> :mgsgb_b: None of those people are actually anonymous.
:mgsgb_c: This would include making feds easy for feds to identify. :mgsgb_d: Government agents being able to identify each other isn't new; this is probably half the reason for the avatar libraries.
You think they're sending Copilot downstream or that they're taking your data upstream? You think they won't change their mind? Every program you write is sent upstream to Apple for cataloguing, Microsoft is already keylogging. Use your head. The first rule of security is you don't *rely* on incompetence.
Anyway, the important conclusions of the post do not hinge on whether or not Recall is always sent upstream or just sometimes or maybe just in the future.
@p since i decided to be a good boy after 2020 i dropped a lot of opsec and stopped analyzing so heavily. still feels weird having something like this slip by my radar, even though I'm trying much less.
@Nepiant Yeah. It is one of the very few Microsoft tools that completely drops Tor traffic. (LinkedIn and Github and most other Microsoft stuff doesn't mind Tor at all. Even a lot of Bing stuff, though not the translation tool.)
:mgsgb_e: This makes it easier to understand the government wanting to tamp down on Chinese AI and to ensure that the US corners the market. :mgsgb_f: People are using AIslop not just as avatars, but it's nearly the only thing people use for blog header images nowadays. (I wonder how many Telegram accounts, Gablins, ZeroHedge authors, etc., can be deanonymized right away by this kind of tactic.) :mgsgb_1::mgsgb_0: This extends to Tor: I don't think I have seen *any* of these image generation tools that allows access from Tor. No boats, even, just a 403.
and i don't mean that random people will occasionally come across the page. what i mean is there's a group of people who like to check for page updates on there regularly.
@moth_ball@Nepiant They were one of the weird ones. I seem to recall them being corkscrews but maybe that was pigs.
I really gotta get up to date on my animal dicks so that if I go somewhere and order food that I can't identify, I can at least tell if they've given me a bowl of dicks or not.
@p@Nepiant Yeah they were the corkscrews. The reference image showed that with geese the dick is even longer. With pigs it's this weird cable that goes all the way into the uterus. Apparently pig orgasms can last 30 minutes too.
And then there's the felines with the spiked dicks and... hey come back!
@p@moth_ball yeah, it's ducks that have corkscrew penises. basically natural selection between them favored the females who could choose their mates (by having deeper, more twisted vaginas... i think) and males who could choose their mates (rape) (by having longer, twistier dicks, to match those vaginas).
@phnt@p Maybe I should just give up on the internet... how the fuck are you people still posting these soyjack comics? I get a visceral disgust reaction every time I see one. You ever seen one of those botched trans surgeries? My disgust towards a soyjack is probably a few times over what those bring up. So again, how the fuck are you people still postings these abominations?
> how the fuck are you people still postings these abominations?
I'm not a huge fan of them, but it fits really well in this case. In some cases, they work really great. Here is a more appealing drawing and you can forget about the soyjacks. chompette.jpe
@alyx@p Don't worry I have like 7 memes with them in total. I'm don't like them much either, since they were hijacked by 13 year olds that think the new ones are somehow funny. Most of the time they no longer even look like the people they are supposed to resemble. firefox-crash-log.jpg
@pernia@p Bash history but it's every keystroke and is backed up on microsoft servers because Billy boy cares about you (you-> :scared_satania:💋 <- Billiam Gates))
@pernia@p -You add a space in front of the command or `set +o history`, or disable history and you no longer have bash history. - `history -c` will erase the current bash history permanently. -The history is only stored for so long, with later commands eventually erasing the oldest commands (this is also adjustable to be shorter or longer as the user desires). -The keys pressed are not keylogged, only the final command you execute is logged (so you can re-execute a previous command or later search for a command you previously did).
@p Also now that I think about generative AI. I wonder when they will intentionally start adding unicode characters that look the same or double spaces as a form of tracking like it is already done in government documents. They can leverage the fact that basically everything supports unicode now and use that to track spread of their generated responses across the Internet with almost 100% accuracy.
> They can leverage the fact that basically everything supports unicode now and use that to track spread of their generated responses across the Internet with almost 100% accuracy.
That's clever, but it seems like you could scan a document for these things pretty easily, couldn't you?
@p You could once you get a list of characters in unicode that look similar and add zero-width space/double space scanning on top of that. Once you start mixing languages in a single document, it gets slightly harder, but it's still possible.
The question is: Who will bother doing all of that? Ain't nobody got time for that. Whistleblowers probably will, but even the usual fedi user on this side probably won't. I know that, if it is a few sentences, I would probably rewrite that, but if we are talking about multiple paragraphs, I would probably copy and paste it.
@p And remember that if you are copy pasting a post that you don't know was written by an AI, you are unwillingly spreading it, so in a sense you are also tracked.
Fairy fluff (Oh christ I haven't said that in ages.) Does the margin calculate based on char count while the text is just like sat there bing-chilling?
Okay, in non-retard language:
I imagine there's not being a calculation applied to text beyond generic word-wrap / whatever the fuck CSS property is being used by the notification gets a calculation based on how long the perceived text would be then renders that in when it gets added to the document?
So if it's 200 chars it would go "Oh, this will need expanding!" then it will calculate how much to expand based on font-size and all that jazz. When it sees 5,000 (mostly ZWSP) chars though it goes "Nah, fuck that." and has a bit of a panic attack (it doesn't know they're zwsp and treats chars as monospace).
That would make sense to me since one can be left just to HTML5/CSS3 but one needs to require JS at SOME point to know how to look.
I'm probably WAAAY off but I could see myself implementing something similar.
@Kirino@p At least it works and doesn't break unless you try to. Misskey's UI completely shits itself into being unresponsive with ~1K of zero-width spaces.
Huh, that's fucking WEIRD. I genuinely appreciate you actually getting into the weeds though, I was just spit-balling so it's actually really cool to see how it words -- even if I'm now fueled by sheer rage at the weird fucking way it works!